QUANTIFYING GROUNDWATER RECHARGE DYNAMICS USING A 

PROCESS-BASED DISTRIBUTED HYDROLOGIC MODEL 

 
By 
 

Guoting Kang 

Michigan State University 

in partial fulfillment of the requirements 

for the degree of  

Civil Engineering—Doctor of Philosophy 

A DISSERTATION  

Submitted to  

2018 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 

 

 

 
 
 
 

 

 

ABSTRACT 

 

QUANTIFYING GROUNDWATER RECHARGE DYNAMICS USING A 

PROCESS-BASED DISTRIBUTED HYDROLOGIC MODEL 

 
By 
 

 

Guoting Kang 

Groundwater – the lifeblood of groundwater-dependent ecosystems and societies 

– is facing unprecedented threats from over-extraction, contamination, and 

changing climate. Groundwater recharge provides a sustainable source of water 

for aquifers and plays an important role in both surface and sub-surface domains. 

Understanding and accurately estimating the rate, location, and timing of major 

recharge events and their seasonal and inter-annual variability is key to safely 

matching societal needs of water and to maintaining healthy groundwater-

dependent ecosystems. This work attempts to understand and quantify recharge 

dynamics in an agricultural watershed in the Ottawa County, Michigan using field 

observations of baseflows, groundwater heads, satellite-based 

evapotranspiration (ET) products and an integrated, process-based hydrologic 

model. Specific objectives of the work are to: (1) understand the spatial and 

temporal distribution of high- and low-recharge events and (2) assess the relative 

impacts of climate, land use, soils, and topography on the spatiotemporal 

distribution of recharge within the region. County-wide synoptic and time-series 

baseflow data collected from over 40 small streams between July and November 

of 2015 were used to quantify the uncertainties in recharge estimation. 

Precipitation data represent important inputs to hydrologic models and have a 

 

major influence on model performance and the estimated recharge. Compared to 

data from a typical network of rain gauges, the Next-Generation Weather Radar 

(NEXRAD) provides precipitation data at a much higher spatial resolution. 

NEXRAD data were blended with traditional rain gauge data to estimate recharge 

and to evaluate differences relative to recharge estimated using rain gauge data 

alone. Results indicate that caution should be exercised in using NEXRAD 

precipitation data for recharge estimation. The representation of recharge and its 

variability within a numerical model are closely related to the representation of 

meteorological forcing fields and their spatial structure, land use and land cover, 

the hydraulic properties of underlying soils and aquifers as well as topography -- 

all of which are represented to varying degrees of accuracy depending on the 

mesh resolution employed and the algorithms used to represent sub grid-scale 

processes. To understand the effects of grid resolution on recharge and to 

identify optimal resolution relative to the size of the watershed, models were 

setup with different grid resolutions. Recharge patterns follow precipitation 

patterns more closely at coarse grid sizes since the characteristics of LULC, 

terrain and hydraulic properties are smoothed at this resolution. Insights gained 

from the study are expected to aid in the sustainable management of natural 

resources, particularly groundwater-dependent ecosystems. 

 
 
 
 
 
 
 
 

 

 
 
 
 

 
 
 
 
 
 
 
 

In loving memory of my grandmother, Zhufeng Wang  

 

iv 

 
 
 
 
 
 
 
 

 

 

ACKNOWLEDGEMENTS 

  

 

This dissertation research cannot be finished without the support and 

encouragement from many professors, colleagues, friends, and family.  

First, I want to give my gratitude to Dr. Mantha S. Phanikumar, for his 

fundamental role in my doctoral studies. Thank you for your patience and critical 

opinions that helped me successfully complete my dissertation research. Thank 

you for introducing me to the hydrologic study which enhances my research 

career prospects.  

In addition, I would like to thank my dissertation committee, Dr. Lifeng Luo, Dr. 

Yadu Pokhrel, and Dr. Shu-guang Li, for their invaluable comments and 

suggestions to improve this dissertation to its final form. Thank you for your 

understanding and believing in me especially during my hard time.  

I would like to acknowledge Dr. Huasheng Liao for assisting the analysis of 

groundwater data. I also want to express my thanks to my lab mates Tuan 

Nguyen, Ammar Safaie, and Han Qiu who offered great help with streamflow 

sampling. I also want to thank my friends in the department, Zach Curtis, 

Sanghoon Shin, Xiaoyu Zhang and Huiyun Wu for sharing your research and 

study with me.  

I am deeply thankful to my parents for their love, support, and sacrifices. I would 

pay special thanks to my husband, Jia Feng, for giving me strength and courage. 

Thanks to my daughter Nina Feng who always brings happiness into my life. 

 

v 

TABLE OF CONTENTS 

 

 

LIST OF TABLES ............................................................................................... viii 
 
LIST OF FIGURES .............................................................................................. ix 
 
CHAPTER 1 .......................................................................................................... 1 
Quantifying the High-Resolution Spatiotemporal Dynamics of Recharge in a 
Great Lakes Watershed ........................................................................................ 1 
1.1 Introduction .................................................................................................. 2 
1.2 Materials and Methods ................................................................................ 7 
1.2.1 Site Description ..................................................................................... 7 
1.2.2 Field Work ........................................................................................... 10 
1.2.3 PAWS Model ...................................................................................... 12 
1.2.4 Data Sources ...................................................................................... 15 
1.2.5 Criteria for Model Assessments .......................................................... 17 
1.3 Results and Discussion ............................................................................. 19 
1.3.1 Streamflow .......................................................................................... 20 
1.3.2 Evapotranspiration .............................................................................. 30 
1.3.3 Groundwater Heads ............................................................................ 34 
1.3.4 Spatial Variation in Recharge ............................................................. 36 
1.3.5. Temporal Dynamics of Recharge ...................................................... 38 
1.3.6. Recharge Dynamics for Various LULC .............................................. 43 
1.4 Conclusions ............................................................................................... 45 

 
CHAPTER 2 ........................................................................................................ 49 

A Comparison of Groundwater Recharge Estimated with gauge and NEXRAD 

Precipitation Data Using a Physically-Based Hydrologic Model .......................... 49 
2.1 Introduction ................................................................................................ 50 
2.2 Materials and Methods .............................................................................. 53 
2.2.1 Study Area .......................................................................................... 53 
2.2.2 Meteorological Data ............................................................................ 56 
2.2.2.1 Weather Station Data ................................................................... 56 
2.2.2.2 NEXRAD Precipitation Data ......................................................... 57 
2.2.3 Model Implementation......................................................................... 59 
2.3 Assessment of the NEXRAD MPE precipitation ........................................ 61 
2.3.1 Comparisons of the Gauge and NEXRAD MPE Precipitation ............. 61 
2.3.2 Comparisons of the Streamflow .......................................................... 65 
2.4 Results ...................................................................................................... 72 
2.4.1 The Spatial Distribution of Recharge .................................................. 72 
2.4.2 Temporal Change of the Areal Recharge Using Two Precipitation Data
 ..................................................................................................................... 75 
2.4.3 Water Budget ...................................................................................... 79 

 

vi 

2.5 Discussions and Conclusions .................................................................... 81 

 
CHAPTER 3 ........................................................................................................ 85 

Effects of Grid Resolution on Groundwater Recharge Estimation Using a 

Physically-Based Hydrologic Model .................................................................... 85 
3.1 Introduction ................................................................................................ 86 
3.2 Materials and Methods .............................................................................. 90 
3.2.1. Site Description .................................................................................. 90 
3.2.2. Model Descriptions and Setup ........................................................... 94 
3.3 Results ...................................................................................................... 98 
3.3.1. Grid-size Effects on the Spatial Pattern of Recharge ......................... 98 
3.3.2. Grid-size Effects on the Temporal Pattern of Recharge .................. 100 
3.3.3. Grid-size Effects on the Average Annual Water Budget .................. 103 
3.3.4. Grid-size Effects on Recharge over Different LULC ........................ 104 
3.3.5 Scaling Effects on Annual Recharge ................................................ 109 
3.4 Discussions and Conclusions .................................................................. 111 

 
APPENDICES ................................................................................................... 115 
APPENDIX A ................................................................................................. 116 
APPENDIX B ................................................................................................. 117 

 
BIBLIOGRAPHY ............................................................................................... 121 
 
 
 

 

 

 

vii 

 

LIST OF TABLES 

 
Table 1.1 U.S. Geological Survey (USGS) gauging stations .............................. 16 
 
Table 1.2 Model calibration parameters .............................................................. 20 
 
Table 1.3 Summary of one-time streamflow sampling ........................................ 30 
 
Table 2.1 U.S. Geological Survey (USGS) gauging stations .............................. 61 
 
Table 2.2 Statistical analysis of MPE precipitation based on the gauge 
precipitation at the three USGS gauges. ............................................................. 64 
 
Table 2.3 Statistical analyses of the discharge simulation performance at three 
USGS gauges for the period from January 2011 to November 2015. ................. 72 
 
Table 2.4 Statistical analyses of the discharge simulation performance at three 
USGS gauges for the period from January 2011 to December 2014. ................. 72 
 
Table 2.5 Annual water budgets (%P, percent of annual precipitation) ............... 81 
 
Table 3.1 Ottawa County watershed characteristics at different grid sizes ......... 97 
 
Table 3.2 Average annual water budgets (%P, percent of annual precipitation)
 .......................................................................................................................... 104 
 

 

viii 

LIST OF FIGURES 

 
Figure 1.1 Map of the modeling domain showing the major streams, National 
Climatic Data Center (NCDC) and Michigan Automated Weather Station Network 
(MAWN) weather stations, U.S. Geological Survey (USGS) gauging stations, 
National Hydrography Dataset (NHD) streams and the sampling locations for the 
streamflow. ........................................................................................................... 9 
 
Figure 1.2 Map of the land use and land cover for the modeling domain. ........... 10 
 
Figure 1.3 Process flow diagram......................................................................... 13 
 
Figure 1.4 Water budget in a soil column. ........................................................... 14 
 
Figure 1.5 Comparisons between the simulated and observed daily streamflow 
rates at three USGS gauging stations. Dark-blue lines represent the observed 
streamflow rates; dark-red lines represent the simulated streamflow rates, and 
the black bar represents precipitation. The hydrographs of a1, b1, and c1 are 
presented on a linear scale from 2011 to 2015. In a2 to c2 and a3 to c3, close-up 
views are presented for the selected simulation period represented within the 
dark-blue dashed lines. In a2 to c2, the enlarged plots are on a linear scale, and 
in a3 to c3, enlarged views are plotted on a semi-log scale. ............................... 23 
 
Figure 1.6 Comparisons between the simulated and measured daily flow at 
sampling locations. Red lines represent the simulated streamflow rates, and the 
blue dots represent the measured streamflow rates. The hydrographs are 
presented on a semi-log scale. ........................................................................... 27 
 
Figure 1.7 Maps of the average annual evapotranspiration for 2011–2014 based 
on (a) MODIS ET and (b) simulated ET. ............................................................. 32 
 
Figure 1.8 Monthly time-series of MODIS and simulated ET for 2011–2014. (Unit: 
mm/month) .......................................................................................................... 34 
 
Figure 1.9 Comparison between the simulated groundwater head and the 
wellogic database for the period of 2011-2015 ................................................... 35 
 
Figure 1.10 Map of the average annual recharge for the 4-year period of 2011–
2014. ................................................................................................................... 38 
 
Figure 1.11 Monthly recharge in different land uses throughout four hydrological 
years: September 2011–September 2015. ......................................................... 41 
 
Figure 1.12 Dynamics of monthly areal recharge for the 4-year period of 
September 2011–September 2015. .................................................................... 43 

 

ix 

Figure 1.13 Annual recharge in various types of land cover in the year of 2012 
and 2014 (Unit: mm/year) ................................................................................... 45 
 
Figure 2.1 Study area. (The top right map shows the location of the Ottawa 
County watershed in Michigan; the top left map shows the major streams in the 
watershed, the weather stations inside the watershed, the NEXRAD Radar 
HRAP grids (4 km spatial resolution), the Radar station and the UGSG gauges; 
the lower left map shows the spatial variability of elevation; the lower right map 
shows the primary land use and land cover types inside the watershed. ............ 55 
 
Figure 2.2 Scatter plots of the gauge and the NEXRAD MPE precipitation rates 
over the period of 2011-2015. (Unit: mm/day) .................................................... 62 
 
Figure 2.3 Comparisons between observed and simulated discharge at USGS 
04119400. (The dark green line depicts the simulation conducted with gauge-
MPE precipitation; the dark red line describes the simulation conducted with 
gauge precipitation. The presented simulation period is from January 2011 to 
November 2015. The selected window period is from January 2015 to November 
2015.) .................................................................................................................. 69 
 
Figure 2.4 Comparisons between observed and simulated discharge at USGS 
04108800. (The dark green line depicts the simulation conducted with gauge-
MPE precipitation; the dark red line describes the simulation conducted with 
gauge precipitation. The presented simulation period is from January 2011 to 
November 2015. The selected window period is from January 2015 to November 
2015.) .................................................................................................................. 70 
 
Figure 2.5 Comparisons between observed and simulated discharge at USGS 
04108600. (The dark green line depicts the simulation conducted with gauge-
MPE precipitation; the dark red line describes the simulation conducted with 
gauge precipitation. The presented simulation period is from January 2011 to 
November 2015. The selected window period is from January 2015 to November 
2015.) .................................................................................................................. 71 
 
Figure 2.6 The spatial maps of average annual precipitation and average annual 
recharge based on gauge and gauge-MPE precipitation over period 2011-2014.
 ............................................................................................................................ 74 
 
Figure 2.7 The comparison between gauge and gauge-MPE precipitation 
conducted areal monthly recharge. ..................................................................... 77 
 
Figure 2.8 The comparison between estimated seasonal recharge based on 
gauge and NEXRAD MPE precipitation data. (X-axis is time, “W” indicates winter, 
“S” indicates Summer) ........................................................................................ 78 
 
Figure 3.1 Climate and hydrologic stations in the study domain. ........................ 92 

 

x 

Figure 3.2 Map of land use and land cover in the watershed. ............................. 94 
 
Figure 3.3 The spatial patterns of (a) average annual ET; and (b) average annual 
recharge over period 2011–2014. ....................................................................... 99 
 
Figure 3.4 Monthly areal precipitation with 300 m grid size and monthly areal 
recharge over all the grid sizes of the period 2011-2014. (Units: mm/month) ... 101 
 
Figure 3.5 Annual areal recharge over all the grid sizes between 2011 and 2014.
 .......................................................................................................................... 102 
 
Figure 3.6 Comparison of daily discharge over all the grid sizes at USGS 
04108800 station. ................................................................................................. 3 
 
Figure 3.7 Comparisons of annual recharge among all the grid sizes in different 
land types in the drought year of 2012 .............................................................. 107 
 
Figure 3.8 Comparisons of annual recharge among all the grid sizes in different 
land types in the wet year of 2013 .................................................................... 108 
 
Figure 3.9 Comparisons of annual recharge (mm/year) between (a) the upscaled 
from 700 m to 1 km and 1 km grid size; (b) the upscaled from 500 m to 1 km and 
1 km grid size; (c) the upscaled from 300 m to 1 km and 1 km grid size; and (d) 
the upscaled from 300 m to 500 m and the 500 m grid size for the period 2011–
2014. ............................................................................................................... 1111 
 
Figure A1 Comparisons between observed and simulated discharge at USGS 
04118800. ......................................................................................................... 116 
 
Figure A2 Comparisons between observed and simulated discharge at USGS 
04108600. Dark green line depicts the simulation conducted with gauge-MPE 
precipitation; Dark blue dash line describes the observed discharge................ 116 
 
Figure B1 Comparison between daily discharge over four grid sizes with 
parameters calibrated at 300 m grid size and daily discharge at 1 km grid size 
with parameters calibrated at 1 km from 2011 to 2014. (Units: m3/s) (g300, g500, 
g700 and g1K indicate the parameters estimated at 300 m grid size, g1K_1K 
indicates the parameters estimated at 1 km grid size) .................................. 11919 
 
Figure B2 Comparison between monthly areal recharge over four grid sizes with 
parameters calibrated at 300 m grid size and monthly areal recharge at 1 km grid 
size with parameters calibrated at 1 km from 2011 to 2014.(Units: mm/month) 
(g300, g500, g700 and g1K indicate the parameters estimated at 300 m grid size, 
g1K_1K indicates the parameters estimated at 1 km grid size) ........................ 120 
 

 

xi 

Figure B3 Comparison between annual areal recharge over four grid sizes with 
parameters calibrated at 300 m grid size and annual areal recharge at 1 km grid 
size with parameters calibrated at 1 km over the period 2011-2014. (Units: 
mm/year) (g300, g500, g700 and g1K indicate the parameters estimated at 300 
m grid size, g1K_1K indicates the parameters estimated at 1 km grid size) ..... 120 
 
 

 

 

 

xii 

CHAPTER 1 

Quantifying the High-Resolution Spatiotemporal 

Dynamics of Recharge in a Great Lakes Watershed 

 

Groundwater recharge is a sustainable source for recharging and cleaning the 

groundwater and aids the integration of the surface and sub-surface water 

resources. Understanding and estimating the natural recharge rate, locations, 

and timing of major recharge events in the groundwater system are essential to 

meet human consumption and maintain a healthy groundwater-dependent 

ecosystem. This study attempts to understand and quantify the recharge 

dynamics in an agricultural watershed in Ottawa County, Michigan with an 

integrated, process-based hydrologic model with field observations of the 

streamflow rates, USGS streamflow rates, groundwater heads information, 

satellite image-based evapotranspiration (ET) products. The study objectives lie 

in: (1) quantifying the spatial and temporal distributions of large- and small-

recharge events in the focal region; and (2) assessing the impact of climate, 

LULC, soil, and topography factors on the spatiotemporal distribution of recharge 

events in the region. The spatial variation of the simulated recharge reveals that 

the geographical pattern of the simulated recharge follows a general trend 

whereby high or low values of recharge coincide with locations with high or low 

elevations, respectively, and high or low recharge locations also have low or high 

values of evapotranspiration, respectively. The temporal dynamics of the recharge 

 

1 

illustrate that recharge rates are dependent on the year-to-year climate variation 

corresponding to the hydrological year. Further, the recharge values are also 

affected by different land types and various soil types, whereby the recharge 

values were relatively high in agricultural land and low in the forested and 

impervious land.  

1.1 Introduction 

Groundwater resources worldwide are under increasing pressure as a result of 

rapid population growth, agricultural usage, industrial demands, and inefficient 

water management. Ensuring a reliable groundwater supply and minimizing 

groundwater contamination and its environmental impact are critical and urgent 

challenges for groundwater management (Comte et al., 2016; Gorelick and 

Zheng, 2015). As water levels decline in groundwater systems globally, 

quantifying the spatiotemporal variability of natural recharge has become key to 

protecting and managing both the quality and quantity of groundwater (Scanlon 

et al., 2006; Simmers, 2013). Further, understanding the spatial and temporal 

patterns of recharge is vital to evaluating contaminant transport in aquifer 

management and protection and further to managing the waste site (Glenn et al., 

2016; Healy and Scanlon, 2010; Niazi et al., 2017; Scanlon and Cook, 2002). 

However, estimating the temporal and spatial variability of recharge has been a 

challenge (Vries and Simmers, 2002), because the occurrence and spatial-

temporal variations is responsive to climatic factors, topography, the nature of the 

land use/land cover (LULC) types, and the hydraulic properties of the underlying 

soils ( Arnold et al., 1998; Chinnasamy et al., 2015; Crosbie et al., 2013; Freeze 

 

2 

and Cherry, 1979; Hunt et al., 2008; Jyrkama et al., 2002; Kendy, 2005; 

McMahon and Böhlke, 2006; Memon, 1995; Turkeltaub et al., 2014). 

Understanding the impact of these climatic and physical factors on recharge can 

provide a basis for improving groundwater flow analysis and for promoting 

recharge estimation in large-scale water balance studies, groundwater models, 

and climate models more generally (Bresloff et al., 2013; Kim et al., 2012).  

Because of the landscape variability in vegetation and soil, estimations of the 

spatial and temporal variability of recharge come with a high level of uncertainty, 

especially when the land surface and subsurface conditions are heterogeneous 

(Glenn et al., 2016; Niazi et al., 2017). To reduce the level of generalization in 

using low spatial and temporal resolution data, high-spatial-and-temporal-

resolution recharge estimation has the potential to help us model the reality of 

recharge explicitly. Physically-based, surface–subsurface integrated hydrologic 

models, which quantify all the relevant hydrologic processes contributing to 

recharge, play an essential role in estimating the spatial and temporal patterns of 

recharge (Awan and Ismaeel, 2014; Dakhlalla et al., 2016; Jyrkama et al., 2002; 

Scanlon and Cook, 2002). Models of this kind are viewed to be either semi-

distributed or fully distributed models, with either sequential coupling or full 

coupling schemes. Semi-distributed models lump the climatic factors and 

physical parameters into sub-basins or hydrologic response units (HRUs) and 

predict the average behavior of a catchment based on several small 

homogeneous units, e.g., HEC-HMS (U.S. Army Corps of Engineers, 2000) and 

SWAT-MODFLOW (Bailey et al., 2016). The finest spatial resolution depends on 

 

3 

the HRUs, and the temporal resolution is limited to the lumped and simplified 

conceptual model, which often proceeds at a regular and fixed time interval 

(Kampf and Burges, 2007). Fully distributed models discretize the watershed into 

grid cells, with the model parameters determined based on climatic factors, 

physical properties of the soil type, and the land use of the given cell, 

e.g. ,CATHY ( Camporese et al., 2010; Paniconi and Wood, 1993; Paniconi et 

al., 2003), ParFlow (Ashby and Falgout, 1996; Kollet and Maxwell, 2006), InHM 

(VanderKwaak and Loague, 2001), MIKE-SHE (DHI, 2004), and PAWS (Shen 

and Phanikumar, 2010). Therefore, fully distributed models are capable of 

describing the hydrological processes at various spatial and temporal scales and 

quantifying the hydrologic responses of the watershed to the variations in 

climatic, LULC, soil, and topographical factors (Biftu and Gan, 2001). Also, the 

coupling scheme is a criterion for selecting an appropriate integrated hydrologic 

model for a given application (Kampf and Burges, 2007). Sequential coupling 

scheme has a non-iterative coupling approach such as PAWS, HEC-HMS, and 

an iterative coupling approach such as CATHY. These two coupling approaches 

(especially the non-iterative one) are computationally efficient such that they can 

be applied to large-range watersheds ( Dagès et al., 2012; Shen and 

Phanikumar, 2010). However, the sequentially coupled models are sensitive to 

temporal discretization (Bailey et al., 2017; Camporese et al., 2010; Dagès et al., 

2012). Theoretically, full coupling scheme is more accurate than sequential 

coupling (Furman, 2008). However, this does not necessarily mean that full 

coupling is superior to sequential coupling. The fully coupled models are often 

 

4 

restricted to short simulation periods and small catchments ranging from just a 

few kilometers to several hundred kilometers, e.g.,  InHM,  (VanderKwaak and 

Loague 2001).  Furman (2008) argues that the full coupling method is more 

suitable for academic application studies than in the practical applications.  

Building a reliable and verifiable physically-based hydrologic model with a high 

spatiotemporal resolution is challenging as quantifying recharge entails 

substantial execution time and might also cause overparameterization (Beven, 

1996). PAWS + CLM (Common Land Model) in this study is a seamless 

integration of all the important processes of the hydrological cycle to account for 

the entire hydrology of a watershed/aquifer system. Moreover, PAWS+CLM is 

grid-based, which allows each component in water balance to be produced at 

very fine spatial scales and addresses problems at a local scale. Each process in 

the non-iterative method in PAWS has its own time step so that this model is 

considerably more computationally efficient than using a uniform time step for all 

the hydrologic processes involved. These processes vary in terms of their 

respective response time. For example, regarding rainfall events, surface flows 

respond very quickly, evapotranspiration (ET) reacts quite slowly, and 

groundwater responds even more slowly.  Multiple datasets, such as streamflow 

rates (discharge) from USGS gauges streamflow rates measured between July 

and November in 2015, wellogic groundwater heads, and MODIS ET products 

were used to validate the model. USGS gauges were installed in relatively large 

streams in the area. County-wide spatially distributed data of streamflow rates 

and time-series streamflow rates were collected from small streams. Such small 

 

5 

streamflow rates measured in small streams were used to reduce parameter 

uncertainties and limit overparameterization in the model. Streamflow in the 

summer can be considered the baseflow given that it approximately equals the 

recharge (Arnold et al., 2000; Cherkauer and Ansari, 2005; Gebert et al., 2007; 

Zomlot et al., 2015). The factors that influence recharge, such as the 

groundwater table, precipitation, vegetation, and soil texture, can be examined in 

detail in PAWS as well. Previous studies on various watersheds in Michigan (Niu 

et al., 2014; Riley and Shen, 2014; Shen and Phanikumar, 2010; Shen et al., 

2013) show that PAWS performs very well in terms of simulating the major 

components of the water budget for large- and small-scale watersheds. 

As a part of a project studying water resources in general in Ottawa County, 

Michigan, the county relies heavily on its groundwater for drinking water and for 

agricultural and industrial usage. A declining static water level in the glacial 

aquifer (south-central part of the county) and increasing chloride concentrations 

have been reported in the county (IWR, 2013:85). We observed that many small 

groundwater-dependent streams in the region dried up during our fieldwork in the 

summer of 2015. This loss in the groundwater table indicates that the volume of 

groundwater in Ottawa County is not sustainable. Understanding the 

spatiotemporal dynamics of recharge, therefore, is of the utmost importance in 

groundwater resource management in Ottawa County. Results from this study 

provide much-needed preliminary support for the management decision making. 

The primary objectives of this study are (1) to quantify the spatial and temporal 

distributions of large- and small-recharge events in the focal region between 

 

6 

2011 and 2015; and (2) to assess the relative impact of climate, LULC, soil, and 

topography on the spatiotemporal distribution of recharge in the region. 

Herein, we report on using PAWS + CLM to estimate recharge over five years at 

a fine spatial and temporal resolution. The study is organized as follows: Section 

one provides a general background of the motivation for this high- 

spatiotemporal-resolution groundwater study and a review of physically-based 

hydrologic models. Section two specifies the study area, methods, methodology, 

and data sources. We further show the details of the model calibration and 

validation in section three. While section four presents the results, and section 

five summarizes conclusions and discussions for the recharge estimation in 

Ottawa County. 

1.2 Materials and Methods 

1.2.1 Site Description 

Ottawa County, MI is located in the west-central part of the Lower Michigan 

(Figure 1.1). The primary river in the county is the Grand River, which drains 

westward into Lake Michigan. Streams in the southeast region of the county 

converge into the Rabbit River, as a tributary of the Kalamazoo River to the south 

of Ottawa County. The area on the southwest side of the county is drained by the 

Macatawa watershed. Therefore, in Ottawa County, the modeling domain, 

marked by a dark blue line in Figure 1.1, comprises part of the lower Grand River 

watershed, the Macatawa River watershed, and the Rabbit River watershed. The 

 

7 

modeling domain is approximately 1,900 square kilometers, and the elevation 

within its boundary ranges from 175 m to 325 m.  

Lake Michigan has a strong influence in the climate of the Ottawa County. During 

the winter time, a massive lake effect leads to an average of 1800 mm snow in 

Ottawa County each year. The county’s average annual rainfall is around 900 

mm (miOttawa.org), and the primary uniform rain season is between May and 

October. The lake effect climate is essential to the county given that agricultural 

land accounts for about 50% of the county’s total land mass and provides 

opportunities for diversified agricultural practices (Census of America, 2012). 

Before the lumber industry developed in Michigan, the county was almost entirely 

covered by forests. Today, large areas of forests can only be found along Lake 

Michigan, wetlands and rivers and scattered patches of forests intersperse with 

farmland. Most wetlands in the county are associated with the Grand River and 

its tributaries, while some wetlands are distributed along the Macatawa and 

Pigeon Rivers, scattered inland from the shoreline area, in the northeast. Figure 

1.2 presents the spatial variabilities of land use and land cover (LULC) for the 

entire modeling domain.  

 

 

 

8 

Figure 1.1 Map of the modeling domain showing the major streams, National 
Climatic Data Center (NCDC) and Michigan Automated Weather Station Network 
(MAWN) weather stations, U.S. Geological Survey (USGS) gauging stations, 
National Hydrography Dataset (NHD) streams and the sampling locations for the 
streamflow. 
 

 

 

 

9 

Figure 1.2 Map of the land use and land cover for the modeling domain. 
 

 

1.2.2 Field Work  

In this study, the SonTek RiverSurveyor M9 system and the OTT MF Pro current 

flow meter were used to measure the streamflow rates of small streams in 

Ottawa County. As a robust and highly precise Acoustic Doppler Current Profiler 

(ADCP), RiverSurveyor M9 can measure streams with a minimum depth of 20 

cm. OTT MF Pro is a low-maintenance discharge measurement system with an 

electromagnetic current meter that has a sensor head ideal for use in low-flow 

environments with a measurable depth range from 0 to 3.05 m. The 

 

10 

RiverSurveyor M9 system collected the streamflow rates of streams with a depth 

of more than 20 cm, while the OTT MF Pro was primarily used to collect the 

streamflow rates of shallow streams with a depth of less than 20 cm, but a width 

of at least 1.5 m.  

We performed two types of streamflow sampling in small streams in Ottawa 

County in 2015: “synoptic” or one-time streamflow sampling over ten days in late 

July and a time-series sampling over six months from June to November. 

Streamflow rates were collected in multiple shallow segments of the four major 

rivers (Grand River, Macatawa River, Pigeon River, and Rabbit River) and 44 of 

their tributaries. The one-time streamflow sampling between July 22nd and July 

31st was undertaken with the purpose of providing a snapshot of the spatial 

variations in the flow. In this 10-day period, 48 streamflow rates were collected, 

including 22 measurements collected by the SonTek RiverSurveyor M9, nine 

measurements collected by the OTT MF Pro current flow meter, and 17 dried-up 

streams recorded as streamflow rate of 0 (sampling locations presented in Figure 

1.1). The time-series sampling data for the small streams were collected for two 

principal reasons: (1) to provide a validation dataset in the form of discharge as a 

function of time; and (2) to provide a basis for examining the model’s 

performance for small streams. Channel width and depth measurements were 

also collected at the sampling locations to help correct the river properties in the 

model. Time-series streamflow rates were collected at 14 locations at either 

monthly or biweekly intervals, including 13 measurements collected by the 

 

11 

SonTek RiverSurveyor M9 ADCP and one measurement collected by the OTT 

MF Pro current flow meter.  

1.2.3 PAWS Model  

PAWS incorporates the overland flow, channel flow, unsaturated vadose zone, 

and saturated groundwater flow that cover various flow paths after a rainfall 

event (Shen and Phanikumar, 2010). In addition, PAWS, coupled with CLM, 

provides detailed, process-based vegetation dynamics (Niu et al., 2014; Thornton 

and Zimmermann, 2007), including factors, such as canopy interception, 

snowpack, biomass, depression storage, and evapotranspiration (Lawrence et 

al., 2011; Shen et al., 2013). Based on the entire hydrology of a 

watershed/aquifer system, PAWS makes it possible to capture the dynamic 

nature of surface and groundwater exchange as driven by the head gradient, 

thereby providing opportunities to evaluate the sensitivity of recharge rates to 

climate, land use, water use, and other natural and anthropogenic factors. 

Figure1.3 briefly describes the modeling framework, including the model 

calibration, the model evaluation, and the major results produced by each 

process of the model. Physical and weather data were the input for the model. 

We used the streamflow rates from the USGS gauging stations between August 

2009 and December 2010 to calibrate the model. We then use the streamflow 

rates from the USGS gauging stations from January 2011 to November 2015, 

streamflow rates measured during the fieldwork, and the wellogic groundwater 

data and MODIS ET to evaluate the model’s performance.  

 

12 

Although recharge observations are rarely available, recharge can be 

constrained based on an overall water budget through the surface water, vadose 

zone, and groundwater system. While precipitation and ET are the critical factors 

controlling recharge (Cao et al., 2016; Kendy et al., 2003; Tan et al., 2013), 

recharge estimates are reliable when the primary processes are determining the 

water budget, i.e., discharge and ET, are simulated accurately. An illustration of 

the water budget theory is presented in Figure 1.4. 

Figure 1.3 Process flow diagram.  
 

 

 

 

13 

 

 

Layers of soil 

Figure 1.4 Water budget in a soil column. 
 

Here,  =  −  − −∆ is the recharge;  is the precipitation, which is the 
model input;  is the runoff;  is the evapotranspiration; and ∆ is the changes 
in the soil water. , , and ∆ are computed by PAWS. Compared with the 
values of  and , the value of ∆ is small and for this reason the latter is 

 

neglected in some studies (Lee et al., 2007; Mo et al., 2009; Coelho et al., 2017). 

LU1, LU2, …N refer to the types of land use in one cell. In this study, the three 

dominant types of land use in each discretized cell are selected to represent the 

combination of the land use in one cell.  

 

14 

We applied the model at a 300 * 300 m2 spatial resolution across the entire study 

area, with an hourly time step for weather, vegetation, and evapotranspiration 

simulations, a minutely time step for channel flow and overland flow, and 

adaptive time steps ranging from 2 min to 1 h for the unsaturated vadose zone. 

We aggregated the results to daily, monthly, and yearly time steps for this 

analysis. Twenty-two vertical layers were used to discretize the vadose zone, 

and two layers were defined for the saturated groundwater domain where the top 

layer represents the unconfined aquifer, and the bottom layer is the confined 

aquifer. 

1.2.4 Data Sources 

The meteorological forcing data for the model are daily precipitation, maximum 

and minimum air temperature, wind speed, and relative humidity. The values for 

each of these were obtained from gauge observations collected from 54 stations 

provided by the National Climatic Data Center (NCDC, 

https://www.ncdc.noaa.gov/cdo-web/) and 18 stations provided by the Michigan 

Automated Weather Station Network (MAWN, 

http://www.agweather.geo.msu.edu/mawn/) around the study region over the 

study period (Locations of the weather stations in Figure 1.1).  

The 30 m resolution National Elevation Dataset (NED) from the U.S. Geological 

Survey (USGS, http://nationalmap.gov/elevation.html) was processed to 

generate average cell elevation and lowland-storage bottom elevation in the 

computational grid and used to calculate the slope and surface runoff routing of 

the watershed. The National Hydrography Dataset (NHD) from the USGS 

 

15 

(http://nhd.usgs.gov/data.html) provided the channel flow network and stream 

properties (Figure 1.1). The soil data were obtained from the Soil Survey 

Geographic Database (SSURGO) distributed by the U.S. Department of 

Agriculture, Natural Resources Conservation Service (USDA-NRCS). The soil 

data were processed by Rosetta to provide soil water retention and unsaturated 

conductivity information for the model (Schaap et al., 2001). 

The streamflow rates (discharge) for calibrating and validating the model were 

obtained from the U.S. Geological Survey, National Water Information System 

(USGS-NWIS, http://waterdata.usgs.gov/nwis/rt). Additional streamflow rates 

were collected by fieldwork to better estimate the model parameters. There are 

four USGS gauging stations for the simulation period in the study domain, of 

which three were used to calibrate and validate the model (USGS04119400, 

USGS04108800, and USGS04108600). The other station (USGS04119000), 

which is on the Grand River and close to the eastern boundary of the modeling 

domain, provided upstream inflow data for the model. Table 1.1 shows the 

locations of the four USGS gauges. 

Table 1.1 U.S. Geological Survey (USGS) gauging stations 

Gauge Number  Gauge Name 
04119000 
04119400 
04108800 
04108600 
 

Latitude 
Grand River at Grand Rapids, MI 
42.9644 
Grand River near Eastmanville, MI  43.0242 
42.7792 
Macatawa River near Zeeland, MI 
42.6422 
Rabbit River Near Hopkins, MI 
 
 

Longitude 
-85.6764 
-86.0264 
-86.0183 
-85.7219 
 

 
Groundwater head information was collected from the Wellogic database, which 

includes information such as the static water level (SWL), the depth of the well, 

and the aquifer type, as well as a vertical description of the sediments.  

 

16 

The LULC data used in the model are from the Integrated Forest Monitoring 

Assessment and Prescription (IFMAP) dataset (Figure 1.2) through the Michigan 

Department of Natural Resources (MDNR). IFMAP is derived from the 

classification of Landsat Thematic Mapper (TM) imagery for the period of 1997 to 

2001 and has a 30 m spatial resolution. Land use data were reclassified into 

several model classes, called plant functional types (PFTs) (Thornton and 

Zimmermann, 2007) based on the land use classification percentile values 

reported in previous studies (Jia et al., 2001; Lu and Weng, 2006; Niu et al., 

2014). Because the model cell size (300 m) is coarser than the IFMAP spatial 

resolution (30 m), a model cell possesses a mixture of land use/land cover types 

so three dominant PFTs were recorded and used in the model cells.  

Moderate resolution imaging spectroradiometer (MODIS) Global ET (MOD16) 

products were compared with the ET simulated by the PAWS model spatially and 

temporally. The MOD16 global ET datasets offer 1 km spatial resolution land 

surface ET data for global vegetated land areas at 8-day, monthly, and annual 

intervals for the 2000–2014 period. It should be noted that these datasets do not 

account for ET from either open water or urban areas. Data can be obtained from 

http://www.ntsg.umt.edu/project/modis/mod16.php. 

1.2.5 Criteria for Model Assessments 

Krause et al. (2005) recommend a combination of efficiency criteria for evaluating 

model performance. Accordingly, in this study, in addition to graphical 

techniques, the following statistical criteria were selected for model assessments: 

the coefficient of determination (, Eq. 1.1), the square-root-transformed NASH 

 

17 

metric (, Eq. 1.2) (Shen and Phanikumar, 2010) and the root-mean-
square error (, Eq. 1.3).  
is defined as the squared value of the Pearson’s product-moment correlation 
the line of best fit. Typically, values of greater than 0.5 are considered 

(Legates and McCabe, 1999) representing the percentage of the data closest to 

acceptable (Moriasi et al., 2007). The square-root transformed NASH (RNASH) 

was employed as a second criterion because NASH coefficient tends to give too 

much importance to the peak flows, while RNASH metric is designed to 

emphasize the baseflow (Shen and Phanikumar, 2010). In this study, examining 

baseflow simulation is more important than assessing the peak flows because 

baseflow is the groundwater contribution to streams and is, therefore, associated 

with recharge. RMSE is further used to estimate the difference between the 

measured and the simulated values: 

n



i

1


((
)(
O O S
i



i



S

))

n



i

1


(
O O



i

2

)

n



i

1


(

S

i



2

S

)




 




2

R

 
 

n

i




1

n

i

1


(

(

O
i



S

i

2

)

O
i



O
i

2

)

 

1
n

S O
i
i
n 
1
i

(

2

)

 

 

RNASH

1
 

 
 

RMSE



 

 

 

 

 

 

 

 

 

 

 

(Eq. 1.1) 

(Eq. 1.2) 

(Eq. 1.3) 

2









  

 

 

 

 

 

18 

Where,  is the th observed value,  is the mean of the observed data,  is 
the th simulated value, ̅ is the mean simulated value, and  is the total number 

of events. 

1.3 Results and Discussion 

The daily streamflow rates for the period of 08/01/2009–12/31/2010 from the 

USGS 041190400, USGS 04108800, and USGS 04108600 gauges were used to 

calibrate the PAWS model. The parameters estimated during calibration are 

listed in Table 1.2. The parameter optimization using the Differential Evolution 

(DE) algorithm (Price et al., 2005) was performed at the High-Performance 

Computing Center (HPCC) at Michigan State University. We used the streamflow 

rates at these USGS gauges from 01/01/2011 to 11/22/2015 to validate the 

model. Further, the measured streamflow rates from fieldwork spatially are used 

to validate the model performance on the streamflow rates in small streams. 

Also, the MODIS ET products are used to validate the simulated ET in both 

spatial and temporal terms. The comparison between the simulated and wellogic 

groundwater head is further used to verify the model performance for subsurface 

flow processes. 

 

 

 

19 


 



 

D

oh

K

 

 

 

rK
SK

 
 

N
hR
/k kr

 
 
 

yS

 

Groundwater hydraulic 

conductivity(m/day) 

River Leakances (m/day) 
Soil saturated hydraulic 

conductivity (m/day) 

van Genuchten parameter N 

River bed rise (m) 

Anisotropy ration for the bottom 
conductivity between unconfined 

and confined aquifer 

Specific yield for unconfined 

aquifer  

Adjustment  

method* 

3 

1 

1 

1 

3 

3 
3 

2 
1 
1 

3 

Upper 
Bound 

4 

1 

300 

5 

Lower 
Bound 
1.0e-01 

1.0e-06 

20 

0.1 

60/10** 

1.0e-10 

50 
10 

5 
2 

0.05 

0.5 

1.0e-08 

1.03 

0 

1.0e-10 

20 

5.0e-05 

Table 1.2 Model calibration parameters 
 

Symbol 

Parameter Meaning (unit) 

van Genuchten parameter Alpha 

Empirical fitting parameter for root 
efficiency function (Lai and Katul, 

Average distance from ponding 

domain to flow domain (m) 

overland flow ground interception 

(1/m) 

2000) 

(m) 

*: Adjustment method: 1, Assign the value to the parameter; 2, add the value to 
the parameter; 3, multiply the parameter by the value 
**: 60 and 10 are the upper bounds of the groundwater hydraulic conductivities of 
unconfined aquifer and confined aquifer respectively 
 

1.3.1 Streamflow  

We compared the daily observed and simulated streamflow data over five years 

at three USGS gauging stations (Figure 1.5). The zoom-in hydrographs at each 

station in both semi-log scale and linear scale are shown for the period of July 

1st, 2013–December 31st, 2013. The coefficient of determination, RNASH, and 

RMSE were used to examine the model performance for streamflow simulation. 

The coefficients of determination for all three stations were higher than 0.5, which 

is an acceptable value. The RNASH values ranged from 0.4 to 0.98. The lowest 

 

20 

RNASH was at USGS 04108600, and we think that might be partially because it 

is close to the watershed boundary. The values of RMSE at these three USGS 

gauges were small compared to the streamflow rates at corresponding USGS 

gauges. The comparison at USGS 04119400 (Grand River, Eastmaville) showed 

an outstanding performance on streamflow simulation because of the 

contribution of the upstream inflow input provided by UGSG 04119000, which is 

at the intersection of Grand River and the watershed boundary. All the statistics 

indicate that the model performed to an acceptable standard in simulating 

streamflow, especially during the low flow periods of 2011–2015. The 

mismatches in the hydrograph are primarily due to the underestimation of 

streamflow peaks in the spring. An examination of the weather records shows 

that the precipitation values are low for the underestimated peak periods, 

indicating that precipitation is not the direct cause of the high streamflow rates. 

And we also need to take into account that the county is affected by heavy lake-

effect snow during the winter time. Therefore, the mismatches identified may be 

correlated with the effects of melting snow. A close inspection of Figure 1.5 (b1) 

shows that the simulated streamflow at USGS 04108800 (Macatawa River, 

Zeeland) was underestimated at the peak every spring during the entire 

simulation period because the lake-effect snow was severe at this location, which 

is close to Lake Macatawa and Lake Michigan. There are two possible reasons 

for the uncertainties on the snow-melting periods: (1) the uncertainty of the 

measured snowfall depth induced by the measurement methods at the weather 

stations, and (2) the deficiency of the snow energy balance routine of the current 

 

21 

CLM model. The snowmelt events have the potential effects of reducing the 

correlation coefficient as well as the RNASH values.  

 

22 

USGS 04119400 (Grand River, Eastmanville)

1250
1000

750
500

R2 =0.96
RMSE =21.06
RNASH =0.98

06/30/11

12/27/11

06/24/12

12/21/12

06/19/13
103

Date (mm/dd/yy)
(a2)

12/16/13

06/14/14

12/11/14

06/09/15

(a1)

0
50

100
150

200
250
11/22/15

(a3)

250
0

01/01/11
200

150

100

50

0

07/01/13

102

08/30/13

10/29/13

12/28/13

101
07/01/13

08/30/13

10/29/13

12/28/13

 

Figure 1.5 Comparisons between the simulated and observed daily streamflow rates at three USGS gauging 
stations. Dark-blue lines represent the observed streamflow rates; dark-red lines represent the simulated 
streamflow rates, and the black bar represents precipitation. The hydrographs of a1, b1, and c1 are presented on a 
linear scale from 2011 to 2015. In a2 to c2 and a3 to c3, close-up views are presented for the selected simulation 
period represented within the dark-blue dashed lines. In a2 to c2, the enlarged plots are on a linear scale, and in a3 
to c3, enlarged views are plotted on a semi-log scale. 
 

 

 

23 

Figure 1.5 (cont’d) 

R2 =0.51
RMSE =3.75
RNASH =0.66

100

80

60

40

20

USGS 04108800 (Macatawa River, Zeeland)

(b1)

0

50

100

150

200

250
11/22/15

(b3)

0

01/01/11
28

21

14

7

0

07/01/13

 

 

06/30/11

12/27/11

06/24/12

12/21/12

06/19/13
102

Date (mm/dd/yy)
(b2)

12/16/13

06/14/14

12/11/14

06/09/15

100

08/30/13

10/29/13

12/28/13

10-2
07/01/13

08/30/13

10/29/13

12/28/13

 

 

24 

Figure 1.5 (cont’d) 

100

80

60
40

R2 =0.53
RMSE =1.78
RNASH =0.4

USGS 04108600 (Rabbit River, Hopkins)

06/30/11

12/27/11

06/24/12

12/21/12

06/19/13
101

Date (mm/dd/yy)
(c2)

12/16/13

06/14/14

12/11/14

06/09/15

20
0

01/01/11
12

9

6

3

0

07/01/13

100

08/30/13

10/29/13

12/28/13

10-1
07/01/13

08/30/13

10/29/13

12/28/13

 

(c1)

0

50

100
150

200
250
11/22/15

(c3)

 

25 

The time-series comparisons between the simulated and field-measured 

streamflow rates were analyzed, and the comparisons at six sampling locations 

scattered throughout the watershed are presented in Figure 1.6. As the 

streamflow rates in these small streams were very small (0~10 m3/s), the 

comparisons of the streamflow rates were plotted in the semi-log scale to 

examine the streamflow rates closely. Value 0.1 m3/s was added to the simulated 

and measured data at location M22 and M38 for plotting in semi-log scale due to 

the zero or close to zero values for some simulation periods. Overall, the 

simulated streamflow reached a good agreement with the measured streamflow 

at these sampling locations. One-time sampling was performed in July 2015 to 

measure the “baseflow,” as the ET values were high in the summer. The 

measured and simulated streamflow rates for the collected streamflow data are 

summarized in Table 1.3. The simulated streamflow rates are close to the 

measured streamflow rates. The results of the time-series sampling data and the 

one-time sampling indicate that the proposed model has (1) the ability to capture 

“baseflow;” and (2) the potential to accurately simulate recharge because 

baseflow can be considered a proxy for the recharge (Arnold et al., 2000; Risser 

et al., 2005; Zomlot et al., 2015).  

 

 

 

26 

 

 

Figure 1.6 Comparisons between the simulated and measured daily flow at 
sampling locations. Red lines represent the simulated streamflow rates, and the 
blue dots represent the measured streamflow rates. The hydrographs are 
presented on a semi-log scale. 
 

 

 

27 

Figure 1.6 (cont’d) 

 

 

 

 

28 

Figure 1.6 (cont’d) 

 

 

 

 

29 

Table 1.3 Summary of one-time streamflow sampling 
 

Instrument  SiteID 

Instrument 

SiteID 

G7 
G9 
G13 
G18 
G20 
G31 
G49* 
G66 
G69 
G77 

Obs  
0.16 
0.11 
0.005 
0.06 
0.05 
0.006 

0.12 
0.25 
0.03 

Sim  
0.12 
0.06 
0.008 
0.1 
0.03 
0.007 

0.15 
0.33 
0.05 

M9 
M9 
OTT  
OTT  
OTT  
OTT  
OTT  
M9 
M9 
M9 

P7 
P12 
R25 
M6 
M16 
M18 
M101 
M71* 
M81* 

 

Obs  
0.1 
0.44 
0.11 
0.01 
0.07 
0.03 

0,0,0 
0,0,0 

Sim  
0.04 
0.26 
0.11 
0.02  
0.05 
0.01 

0,0,0 
0,0,0 

 

0.11,0.06  0.06,0.01 

0.05, 0.05  0.02,0.06 

OTT  
M9 
OTT 
M9 
M9 
M9 
OTT  

 
 
 

G5, G34, G37, G38, G39, G41, G42, G54, G75; 

M5, M7, M8, M28, M51, M54 

 
Observed 

     0 

Simulated 

      0 

*more than one measurement was collected at this location. M9: RiverSurveyor M9. OTT: OTT 
MF Pro. 
 

1.3.2 Evapotranspiration 

To evaluate the performance of our model, we also used the monthly and yearly 

MODIS ET products over the four-year period of 2011 to 2014 to validate the 

simulated ET temporally and spatially. The comparison between the 4-year 

average annual MODIS and the simulated ET of the spatial map is presented in 

Figure 1.7.  

The simulated ET with a spatial resolution of 300 m provides more details than 

does the MODIS ET with a cell size of 1 km. For the simulated ET, the highest 

values occurred along the major rivers and in the wetlands, while the lowest 

values were found in urban areas. The spatial variation in the MODIS ET values 

was relatively moderate compared with those of the simulated ET because ET’s 

coarser resolution smooths the spatial variabilities of ET. Moreover, the MODIS 

ET products do not include the ET from open water or wetlands, where ET 

values appear to be the highest in our model. In addition, the MODIS ET is not 

 

30 

available for urban areas where our model estimates the lowest ET values (Mu et 

al., 2013) (The blank areas in the MODIS ET map in Figure 1.7 represent the 

urban areas). But in an overall comparison, the spatial pattern of the simulated 

average annual ET is similar to that of the average annual MODIS ET.  

In the study area, two regions showed significant differences in their average 

annual ET values. First, the northeast part of the modeling domain shows higher 

average annual simulated ET values than the MODIS ET. The Rough River, a 

wetland associated with Rough River, and the forested land cover in the 

northeast region lead to high ET values. The other region that showed a 

significant difference in its average annual ET values from the MODIS ET is in 

central Ottawa County where the percentage of sand in the soil is high. Water is 

readily brought up to the surface to produce high ET values in sandy soils. Also, 

the values of precipitation were higher than the values of precipitation in the 

surrounding areas. While the simulated ET values are high in most cells in this 

area, the MODIS ET has high ET values only in some cells in this area.  

 

 

 

31 

 
Figure 1.7 Maps of the average annual evapotranspiration for 2011–2014 based 
on (a) MODIS ET and (b) simulated ET. 
 
 
With a time-series ET constraint, the uncertainty of the recharge estimation 

decreases as compared to when a constraint of this nature is not included in the 

model (Xie et al., 2017). Thus, a comparison between the simulated and MODIS 

monthly ET was examined in this study. We aggregated the time series of the 

daily ET output to a time interval by months to compare with monthly MODIS ET 

products. The comparison of the areal average monthly simulated ET and the 

monthly MODIS ET is presented in Figure 1.8. The results for the simulated ET 

resemble the MODIS ET values. The R2 value for the 4-year comparison is 0.79. 

As expected, the highest value for both the simulated and the MODIS ET 

accrued during the summer each year and the lowest values accrued during the 

winter time. The 2012 drought leads to the low ET value shown for the summer 

of 2012.   

 

32 

The primary reason for the differences between the simulated ET and the MODIS 

ET may have arisen because of the different ET algorithms used in PAWS-CLM 

and in producing MODIS ET. PAWS-CLM uses a resistance approach to 

simulate ET based on the two-big leaf model (Dai et al., 2004), whereas the 

MODIS ET is produced by the revised Penman-Monteith formulation (Mu et al., 

2011). The second reason is that the areal monthly MODIS ET does not account 

for ET from either open water or urban areas. However, the areal monthly 

simulated ET includes the ET from both open water and urban areas.  

 

 

 

33 

) 0

h
t
n
o
m
m
m

/

(
 
n
o

i
t
a
t
i
p
i
c
e
r
P

)
h
t
n
o
m
m
m

/

(
 
T
E

50

100

150

200

250

150

120

90

60

30

R2 =0.79

 

PAWS
MODIS

0

 

Jan-2011 Jul-2011 Jan-2012 Jul-2012 Jan-2013 Jul-2013 Jan-2014 Jul-2014 Jan-2015

 
Figure 1.8 Monthly time-series of MODIS and simulated ET for 2011–2014. (Unit: 
mm/month) 
 

Date 

1.3.3 Groundwater Heads 

We derived the observed groundwater heads data from over 400 wells installed 

in unconfined aquifer within the Ottawa County. The estimated groundwater 

heads were extracted for the computation cells where the wells are located. The 

comparison between the estimated and the observed groundwater heads is 

presented in Figure 1.9. The scatter plot shows the groundwater heads in distinct 

colors for different years against the reference 1:1 45-degree line. The overall R2 

value is 0.88, indicating a good match both in spatial and temporal terms. Given 

the close match between the estimated and wellogic groundwater heads data, 

 

34 

we argue that the performance of the PAWS model also provides good estimates 

on the ground heads for this study.  

R2=0.88

240

220

200

180

160

160

2011
2012
2013
2014
2015

180

200

Wellogic GW head (m)

220

240

 

Figure 1.9 Comparison between the simulated groundwater head and the 
wellogic database for the period of 2011-2015 
 

Thus, based on the comparisons in the three fields of the streamflow, the 

evapotranspiration, and the groundwater heads between our estimated data and 

the published data from various sources, we are confident that the PAWS model 

performs well in estimating these variables. Meanwhile, the similar geographical 

and time-series patterns between the estimated and the published data further 

approve the capability of the PAWS model at producing good estimates both in a 

finer geographic scale (300 m) and temporal dynamics of various hydrologic 

 

35 

factors (daily temporal resolution). Despite that recharge is not the factor that we 

can use to compare between the estimated and real values, the water balance at 

the pixel level allows us to argue that if streamflow, ET, and groundwater heads 

are relatively close to the actual situation, recharge estimates should match with 

the actual recharge condition in our study area. Therefore, in the following 

sections, we are going to present the estimated recharge data without further 

justification for the quality of our modeled values. 

1.3.4 Spatial Variation in Recharge  

Figure 1.10 shows the spatial pattern of the average annual recharge over the 

four-year period of 2011–2014. The spatial distribution of the simulated recharge 

followed the general pattern whereby high and low recharge values occurred at 

high and low elevations, respectively. And the opposite relationship was found 

between the spatial patterns of ET and recharge except for urban areas where 

both ET and recharge values are low. The primary recharge areas for the glacial 

aquifer occurred in the northeastern and southeastern parts of the county, where 

the primary land use is agriculture and elevation is high. The central area of the 

county showed low recharge values for two main reasons. First, the high ET 

values were found in this area. Second, the elevation in this area is relatively low. 

The groundwater replenished by the recharge in the northeastern region of the 

county discharged to the Grand River. Also, the groundwater replenished by the 

recharge in the southeastern area of the county mainly discharged to the Rabbit 

River. Therefore, the primary recharge areas did not contribute much 

 

36 

groundwater into the central part of the county where has experienced the high-

level, unsustainable groundwater usage. 

The unexpected high values of recharge along the top eastern boundary of the 

computational domain, which we think are due to boundary issue that the 

watershed was composited based on the surface watershed composition so that 

the groundwater table at the boundary might have been affected by the 

groundwater flow outside of the model domain. And this is the only region in the 

study we think shows slightly different patterns of recharge from our impression 

in the study area.  

 

 

 

37 

Figure 1.10 Map of the average annual recharge for the 4-year period of 2011–
2014. 
 

 

1.3.5. Temporal Dynamics of Recharge 

Recharge is related to factors such as precipitation, land use types, soil types, 

topography, and the ways these factors are combined. The non-linear 

relationship between recharge and these climatic and physical factors is 

complex. Figure 1.11 presents the monthly dynamics of recharge with different 

types of soil for four kinds of land use. In this study, each pixel in the LULC map 

consists of three dominant LULC types expressed as a percentage, which means 

 

38 

different cells might have different combinations depending on their specific three 

dominant types. Soybean, corn, broadleaf, and needleleaf are the dominant land 

types in the selected cells for Figure 1.11. Agricultural land (soybean/corn) 

accounts for 90% of the selected cells, and forested land (broadleaf/needleleaf) 

takes 80–90% of the selected cell. Three major soil types—sand, clayloamy, and 

loam—were defined based on the relative proportions of sand, clay, and silt 

(NRCSS, USDA). The soil in the soybean land and broadleaf forest land for the 

selected cells was mostly sandy, whereas in the corn and needleleaf forest cells 

it mainly was clay loamy. 

The time series of the estimated monthly recharge based on various vegetation 

and soil types illustrate the various responses of recharge to different conditions 

of vegetation, soil, and rainfall. In agricultural land, the simulated monthly 

recharge generally had a late response to the monthly precipitation and 

increased after intense precipitation during the simulation period except for the 

year of 2012 (Figure 1.11(a) and (b)). Precipitation for the year 2012 was below 

average, which had a more significant impact on reducing the recharge in the 

soybean land with sandy soil than in the corn land with clay loamy soil, indicating 

that recharge in sandy soil was more responsive to precipitation. Also, high ET 

values are more often observed in sandy soils. The fluctuation of the monthly 

recharge values was greater in the forested land than in agricultural land. 

Further, the monthly recharge values in the forested land were low, especially in 

2012 due to a period of drought that took place in late spring to fall. The temporal 

fluctuation of recharge in forested land may have arisen at least in part of the 

 

39 

fluctuation in ET, which took a large portion of the precipitation (about 60% in this 

study area). The temporal pattern of ET is primarily affected by precipitation and 

the growing season in the agricultural land, as ET in the forested land is more 

dependent on precipitation given that forests do not have a clear growing 

season. Therefore, we argue that the temporal fluctuation of ET is correlated to 

the temporal fluctuation of recharge in the forested land. Further, the recharge in 

the forested land is also affected by the types of land use because the forested 

area accounted for 80–90% of the selected forested cell. As the model cell is 300 

m * 300 m, 10-20% of the other land use types could potentially cause noticeable 

impacts on recharge.  

 

 

 

40 

Agricultural Land (Soybean)

200

150

100

50

)
h
t
n
o
m
m
m

/

(
 
e
g
r
a
h
c
e
R

0

200

400

600

)
h
t
n
o
m
m
m

/

(
 
n
o

i
t
a
t
i
p
i
c
e
r
P

0
Sep-11 Jan-12 May-12 Sep-12 Jan-13 May-13 Sep-13 Jan-14 May-14 Sep-14 Jan-15 May-15 Sep-15

800

 

Agricultural Land (Corn)

200

150

100

50

)
h
t
n
o
m
m
m

/

(
 

e
g
r
a
h
c
e
R

0

200

400

600

)
h
t
n
o
m
m
m

/

(
 
n
o

i
t
a
t
i
p
c
e
r
P

i

800

0
Sep-11 Jan-12 May-12 Sep-12 Jan-13 May-13 Sep-13 Jan-14 May-14 Sep-14 Jan-15 May-15 Sep-15

 
Figure 1.11 Monthly recharge in different land uses throughout four hydrological 
years: September 2011–September 2015. 
 

 

 

41 

Figure 1.11 (cont’d) 

200

150

100

50

)
h
t
n
o
m
m
m

/

(
 

e
g
r
a
h
c
e
R

Forested Land (Broadleaf)

0

200

)
h
t
n
o
m
m
m

/

400

(
 
n
o

i
t
a
t
i
p
c
e
r
P

i

600

0
Sep-11 Jan-12 May-12 Sep-12 Jan-13 May-13 Sep-13 Jan-14 May-14 Sep-14 Jan-15 May-15 Sep-15

800

Forested Land (Needleleaf)

200

150

100

50

)
h
t
n
o
m
m
m

/

(
 

e
g
r
a
h
c
e
R

0

200

400

600

)
h
t
n
o
m
m
m

/

(
 
n
o

i
t
a
t
i
p
c
e
r
P

i

0
Sep-11 Jan-12 May-12 Sep-12 Jan-13 May-13 Sep-13 Jan-14 May-14 Sep-14 Jan-15 May-15 Sep-15

800

 

Figure 1.12 presents the time series of the average areal monthly recharge for 

the period of September 2011–September 2015. The cycle of the monthly 

recharge dynamics is responsive to the cycle of a hydrological year. However, 

the temporal distribution of the average areal monthly recharge for each 

hydrological year was different. For example, extremely high precipitation values 

were found in April 2013, which contributed to the peak values of recharge. 

However, the effect is not as strong in other hydrological years.  

 

42 

0

50

)
h
t
n
o
m
m
m

/

(
 
n
o

i
t
a
t
i
p
c
e
r

i

100

150

200

250P

20

15

10

5

)
h
t
n
o
m
m
m

/

(
 

e
g
r
a
h
c
e
R

0
Sep-11 Jan-12 May-12 Sep-12 Jan-13 May-13 Sep-13 Jan-14 May-14 Sep-14 Jan-15 May-15 Sep-15

Figure 1.12 Dynamics of monthly areal recharge for the 4-year period of 
September 2011–September 2015. 
 

1.3.6. Recharge Dynamics for Various LULC  

 

 

Boxplots of the annual recharge for four major land use types were created for 

the drought year of 2012 and the wet year of 2014 to examine the impact of 

LULC on recharge (Figure 1.13). We used these two years because the 

climatical factors represent a typical condition of the area as 2012 experienced 

significant drought, 2014 showed more of an average condition in the area. As 

noted in section 2.4, three dominant PFTs were modeled in each cell in this 

study. For example, one cell might be composed of 70% of soybean, 20% of corn 

and 10% of the broadleaf forest. In the boxplots, X-axis noted the percentage of 

one dominant LULC type in a cell. Taking soybean as an example, 10% in the X-

 

43 

axis represents that the cell contains between 0 and 10% of soybean, and 20% 

means that soybean constitutes to between 10 and 20% of the cell, and so forth. 

The 90% of soybean indicates that the cells have larger than 80% but smaller 

than 100% of soybeans. In each box, the central horizontal line indicates the 

median, and the bottom and top of the vertical bar indicate the first and third 

quartile metrics, respectively. The outliers are plotted individually using a “+” 

symbol. 

Figure 1.13 shows that the median values of recharge increased when the 

percentage of the soybean and corn land types in the cells increased for both the 

drought year of 2012 and the wet year of 2014. Conversely, the median values of 

the recharge decreased when the coverage percentages of the broadleaf and 

impervious land types in the cells increased for both years. The recharge values 

are lower in the cells with a larger proportion of forested land types, because the 

root system in the forested land retains a relatively high level of moisture in the 

soil and leads to a high ET value. Changes in the median recharge values over 

the percentage of all four land types are similar in the drought and wet years, but 

the recharge values for 2014 are higher than those for 2012.  

 

 

 

44 

 Figure 1.13 Annual recharge in various types of land cover in the year of 2012 
and 2014 (Unit: mm/year) 
 

 

1.4 Conclusions 

In this study, we explored estimating recharge at a very fine spatiotemporal 

resolution (300 m on a daily basis) at a watershed scale with a process-based 

hydrologic model (PAWS). To build a reliable and valid hydrologic model, we 

used the streamflow rates from USGS gauges and the county-wide streamflow 

rates collected from small streams to spatially and temporally validate the model. 

We also used wellogic groundwater heads data and MODIS ET products to 

validate the model's performance on recharge estimation indirectly. The results 

show that the model estimations are in agreement with the observations of the 

major hydrologic components, including streamflow, groundwater, and ET. 

 

45 

Further, the high spatial and temporal resolution allows us to see more local 

variations of these factors both in individual cells and the time series. Based on 

the validation with existing data, we argue that PAWS model is capable of 

producing a realistic representation of recharge. The capability of PAWS to 

simulate the hydrological process at such a fine resolution is due to two essential 

characteristics: (1) the model is process-based, which allows us to examine each 

factor individually on the fine-spatial-resolution cells; (2) the non-iterative 

sequential coupling method allows us to more efficiently carry out the modeling 

process, which enables us to run the model at such fine spatial and temporal 

resolution for such a long simulation period and a large watershed. The modeling 

process with multiple step-time intervals (from 1 min to daily) takes 3 hours for a 

one-year worth of analysis, which makes the total processing time to be within 24 

hours for the overall project after we made our decisions about parameterization.   

With the PAWS model, we produced daily outputs of recharge for the Ottawa 

County during the five-year study period, and the data is further aggregated to 

monthly averaged values for analysis purposes. The occurrence of recharge and 

its spatial and temporal variations were examined in response to climatic factors, 

land use, the hydraulic properties of the underlying soils, and topography. The 

spatial variation of recharge shows that the primary recharge areas do not 

contribute much groundwater to the central part of the county, which has 

experienced the high-level, unsustainable groundwater usage. The results of the 

temporal dynamics of recharge also advance our understanding of how 

vegetation types and soil contents influence recharge, which is critical for 

 

46 

managing land types rotations to sustain the groundwater usage. Likewise, the 

results strengthen the field's understanding of the long-term impact of forest 

restoration on aquifer recharge volumes and water resource management. The 

temporal dynamics of the monthly recharge show that recharge rates are 

dependent on the year-to-year climate variation corresponding to a specific 

hydrological year. The analysis of the monthly recharge simulation (derived from 

daily recharge data) allow us to have a much closer look at the recharge values, 

patterns, and trends in a local 300 m by 300 m cell size. And we also hope this 

study opens the opportunity for further fine spatial and temporal analysis on 

recharge and furthers the guidance on water resource management at a very fine 

spatial and temporal scale.  

Estimating recharge at a high spatial and temporal resolution is vital in 

apprehending and representing the reality of recharge at the local scale. The 

high spatial and temporal resolution of recharge estimation in this study not only 

makes it possible to design water resource management plans on a very fine 

local scale but also contributes to our understanding of the relationship between 

groundwater recharge and various climatical and physical factors at a fine 

resolution.  

In our study, we also understand that there is a lot left yet to be done. First, the 

modifiable areal unit problem (MAUP) remains an important issue as the fine cell 

size is arbitrarily defined in this study. A follow-up study will further the discussion 

on MAUP to compare the cell-size effect on recharge estimation using 300 m, 

500 m, 700 m, and 1 km cell sizes for the same period and same study area. 

 

47 

Second, we also discuss the effect of various precipitation data provided by 

different sources as well. Third, given the time frame of this study, we do not 

have the chance to carry out a study to analyze the "time-lag" between 

precipitation and recharge cases. We think this study will further our 

understanding of the relationship between precipitation and recharge, which can 

potentially provide more explicit guidance for the water resource management to 

improve the impacts of activities to increase recharge on a local scale by 

mimicking the precipitation-recharge process.  

 

 

48 

CHAPTER 2  

A Comparison of Groundwater Recharge Estimated with 

Gauge and NEXRAD Precipitation Data Using a 

Physically-Based Hydrologic Model 

 

This study aims to understand the differences on recharge estimation using two 

precipitation datasets based on a physically-based hydrologic model for Ottawa 

County, MI between 2011 and 2015. We first use the rain gauge data as the sole 

source for precipitation; and further, we combined the Stage IV NEXRAD data 

(also known as Multi-Sensor Precipitation Estimator, MPE) with the rain gauge 

data as our second data source for precipitation. With the gauge data as a 

reference, this study assessed the quality of NEXRAD MPE precipitation by 

examining the relationship between the two datasets and comparing the 

performance on discharge simulation at three USGS gauges. The results show 

that precipitation data from both sources produce a similar performance in the 

streamflow estimation in our model except for the year 2015. The spatiotemporal 

analysis of recharge shows that the spatial pattern of average annual recharge 

over the period of 2011-2014 was not sensitive to the different precipitation 

datasets and the temporal patterns of monthly recharge are also similar using the 

two datasets from 2011 to 2014. We conclude that NEXRAD MPE data have the 

 

49 

potential to be a supplemental source of precipitation data in understanding the 

impacts of precipitation on recharge.   

2.1 Introduction 

Groundwater recharge provides a sustainable source of water input for aquifers 

and plays an integral role in both surface and sub-surface water resources. 

Understanding and accurately estimating the rate, location, and timing of major 

recharge events and their seasonal and inter-annual variability is important to 

safely satisfy the societal needs of water and preserve the groundwater-

dependent ecosystem. As a major component of a watershed system, 

groundwater recharge is challenging to quantify. A physically-based hydrologic 

model provides a reliable and cost-efficient way to estimate recharge with all the 

relevant hydrological components. 

In many aspects, precipitation data remains one of the most critical factors in 

modeling hydrologic processes as precipitation serves as the direct driving force 

of the hydrologic models (Sexton-Sims et al. 2010; Xie and Xiong, 2011; Gao et 

al., 2017). Specifically, the spatial and temporal variation of precipitation, the 

spatial distribution of the stations gathering precipitation data, the spatial and 

temporal resolution of the precipitation data, and extreme precipitation cases all 

influence the performance of physically-based hydrologic models. A plethora of 

research (Larson and Peck 1974; Price et al., 2011; Cunha et al., 2012; Shen et 

al., 2015) shows that precipitation remains the primary source of uncertainty in 

watershed modeling, while the spatial and temporal variations of precipitation 

influence hydrologic responses, such as spatial and temporal dynamics of runoff, 

 

50 

evapotranspiration, and recharge in a watershed (Daly et al., 2002; Moore et al., 

2006; Cole and Moore, 2008; Mobley et al.2012). Existing studies also found that 

in a physical-based hydrologic model, the responses are sensitive to the 

locations of the rain gauges and the corresponding spatial variability of the 

rainfall data (Bell and Moore, 2000; Segond et al., 2007; Pechlivanidis et al., 

2017). In addition, the spatial and temporal resolutions of the rainfall data also 

have a great impact on hydrologic modeling (Fu et al., 2011; Vergara et al., 2013; 

Ochoa-Rodriguez et al., 2015). The impact of long-term spatial-temporal 

variabilities of precipitation on recharge was examined by Acworth et al. (2016)  

using gauge precipitation. Moreover, extreme precipitation plays a key role in 

determining recharge reported by Zhang et al. (2016) using reanalysis 

precipitation produced based on rain gauge data.  

There are a number of different precipitation products collected using different 

methods, such as rain gauges, ground-based radar (Next-Generation Weather 

Radar, NEXRAD), and satellite-based microwave and infrared sensors. The 

conventional precipitation data collected from the rain gauges are typically used 

as the ground truth to examine the qualities of the precipitation products derived 

by radars and satellites (Grimes et al., 1999; Habib et al., 2009; Price et al., 

2011; Serrat-Capdevila et al., 2014). As a primary source of climate data, a 

network of land-based National Climatic Data Center (NCDC) weather stations 

records the precipitation information collected from rain gauges. However, the 

low density of existing rain gauges and the missing data due to malfunctioning 

lead to large gaps in data and sometimes significant differences between the 

 

51 

data collected from nearby gauges. For large watersheds, the sparsely located 

rain gauges may not be enough to capture the overall variability of precipitation. 

Although radar has itsown limitations, such as range discrimination, pulse length, 

the radar system provides spatially continuous precipitation information at a high 

temporal resolution (Gao et al. 2017).The Stage IV NEXRAD data, also known as 

the multi-sensor precipitation estimator (MPE), is calibrated based on the data 

gathered by the high-accuracy rain gauges through the regional River Forecast 

Centers under the National Weather Service (NWS) and is further mosaicked to 

be a national product with a 4 km spatial resolution (Price et al., 2011). FAS crop 

analysts also argue that the radar-based precipitation data have superior quality 

compared to the other global precipitation data sets that cover the United States 

(Gebremichael and Hossain 2009). MPE has been widely used in hydrologic 

modeling (Kalin and Hantush, 2006; Habib et al., 2009; Price et al., 2011; 

Kitzmiller et al., 2013), but the use of MPE data in recharge estimation has not 

been fully explored. 

This study investigates whether the use of MPE data improves the accuracy of 

recharge estimation, evaluated based on the simulated streamflow data, using a 

physically-based hydrologic model (Process-based Adaptive Watershed 

Simulator, PAWS). This study developed two PAWS models based on data from 

two sources: first, we use the conventional precipitation data collected by gauges 

from the NCDC and MAWN (Enviro-weather Automated Weather Station 

Network). Further, the MPE precipitation is combined with the conventional rain 

gauge precipitation to represent the second source of the precipitation data. 

 

52 

Section 2 explains the data and methods of this study in details, while section 3 

focuses on the comparisons of the precipitation using two different data sources. 

We also compared the simulated and the observed streamflow to evaluate our 

adoption of the MPE precipitation data, as Neary et al. (2004) and Price et al. 

(2011) have shown that streamflow simulations can be used to evaluate the 

quality of MPE precipitation. Moreover, in section 4, we further investigate the 

simulated spatial distribution of recharge, the simulated temporal change of areal 

recharge, and the water budget for using two different data sources. Section 5 

presents the conclusions of this study.  

2.2 Materials and Methods  

2.2.1 Study Area 

Ottawa County, MI, located in the western-central part of the Lower Michigan 

(Figure 2.1), is located within a watershed (shown in Figure 2.1) that consists 

roughly of three sub-watersheds, including a part of the lower Grand River 

watershed, the Macatawa River watershed, and the Rabbit River watershed. The 

total area of the three sub-watersheds is about 1900 km2, and the elevation 

ranges from 175 m to 325 m. The primary land use types of Ottawa County, MI 

are agriculture (about 47%), forest (about 23%), grass (about 10%), urban (about 

10%), followed by water (about 4%) and wetlands (about 4%) (Figure 2.1). The 

agricultural products include mainly corn and soybean. This study focuses on the 

application of a physically-based hydrologic model to the total area of these three 

sub-watersheds.  

 

53 

Having a strong lake effect due to its contiguity with Lake Michigan, the Ottawa 

County has a climate that is strongly influenced by it. During the winter time, 

heavy lake-effect snow has introduced an average of 1800 mm of snowfall each 

year. The average annual rainfall in the county is around 900 mm (miOttawa.org) 

which is relatively uniformly distributed between May and October.  

 

 

 

54 

Figure 2.1 Study area. (The top right map shows the location of the Ottawa 
County watershed in Michigan; the top left map shows the major streams in the 
watershed, the weather stations inside the watershed, the NEXRAD Radar 
HRAP grids (4 km spatial resolution), the Radar station and the UGSG gauges; 
the lower left map shows the spatial variability of elevation; the lower right map 
shows the primary land use and land cover types inside the watershed. 

 

 

55 

2.2.2 Meteorological Data 

The weather stations (NCDC and MAWN) and the NEXRAD (Next-Generation 

Radar) are the two sources for obtaining the precipitation information in this 

study. The weather stations are unevenly distributed in the watershed and are 

concentrated around the urban areas. There are two stations in the southeastern 

watershed and three weather stations located in the central and western 

watershed (Figure 2.1). Daily NEXRAD precipitation products (MPE) produced by 

the Advanced Hydrologic Prediction Service National Weather Service (NWS) 

offer high temporal (daily) and spatial (4 km) resolution precipitation data. Since it 

is based on a field view of weather, precipitation data is available at every 

location within the 4 km by 4 km grid cells.  

2.2.2.1 Weather Station Data 

As meteorological data sets are needed to drive a physically-based hydrologic 

model, two organizations become our main sources: the GHCN-Daily and the 

MAWN. The meteorological data for the model input, including daily precipitation, 

maximum and minimum air temperature, wind speed, and relative humidity data, 

were obtained from GHCN (Global Historical Climatology Network) – Daily 

archive at the National Oceanic and Atmospheric Administration/ National 

Climatic Data Center (NOAA/NCDC, https://www.ncdc.noaa.gov/cdo-web/). 

GHCN-Daily is a composite of climate records from numerous sources that were 

merged and then subjected to a suite of quality assurance reviews (Menne et al., 

2012). However, about two-thirds of the weather stations report the precipitation 

data only. The additional meteorological data, such as air temperature, relative 

 

56 

humidity data, were collected through the Enviro-weather Automated Weather 

Station Network (MAWN: http://www.agweather.geo.msu.edu/mawn/). MAWN 

was built for making decisions strategically based on the weather information to 

improve the efficiency and productivity of agriculture. It provides detailed weather 

data and information, such as precipitation, air temperature, relative humidity, soil 

temperature, at hourly and daily temporal resolution. Putting the two main data 

sources together, there is a total of 72 weather stations around the study area 

over our simulation period (2009-2015), including 54 NCDC stations and 18 

MAWN stations. The locations of the weather stations are shown in Figure 2.1.   

Among all the meteorological data, precipitation data collected by the rain 

gauges from the weather stations as precipitation depth represents the near-

actual amount of precipitation at the scale of a localized point. The gauges are 

unevenly distributed in the study area and in our case, most of the gauges are 

concentrated around the urban areas, of Grand Rapids.  

2.2.2.2 NEXRAD Precipitation Data  

NEXRAD (Next-Generation Radar) system is a network of 160 high-resolution 

Doppler weather radars throughout the United States and select overseas 

locations. Operated by the National Weather Service (NWS), the NEXRAD 

system can detect the precipitation and wind using a field view based on a grid 

size of 4 km by 4 km. The NEXRAD precipitation products are the ground 

accumulated rainfall which is estimated based on a reflectivity and rainfall rate 

relationship (Fulton et al. 1998). The rainfall estimates are improved after 2010, 

due to the deployment of the Dual Polarization (Dual Pol) technology. Dual Pol 

 

57 

more accurately discerns the returns signal which allows the radar to better 

differentiate among the different types of precipitation (e.g., rain, hail, and snow) 

and reduce the non-weather artifacts. The KGRR (Grand Rapids, MI) site which 

is the closest radar site in the study domain, activated Dual Pol in December 

2011.   

The NEXRAD precipitation products are categorized into four product levels 

based on the amount of preprocessing, calibration, and quality control 

procedures performed (Xie et al. 2005; Sexton-Sims et al. 2010; Kang and 

Merwade 2014). The Stage IV NEXRAD (also known as Multi-Sensor 

Precipitation Estimator, MPE) daily precipitation products were utilized in this 

study as the main precipitation data source to be examined in this study. MPE 

products are the gauge-adjusted and manual quality-controlled precipitation 

products estimated by the NWS River Forecast Centers (RFCs) (Young et al., 

2000; Price et al., 2011; Chen et al., 2013). The data can be accessed from 

Advanced Hydrologic Prediction Service National Weather Service (NWS, 

https://water.weather.gov/precip/download.php). The spatial resolution of the 

NEXRAD precipitation data is approximately 4 km by 4 km which is referred to as 

an HRAP (Hydrologic Rainfall Analysis Project) grid, shown in Figure 2.1 (Reed 

and Maidment, 1999).  

NEXRAD can locate and follow precipitation up to over 400 km. However, 

NEXRAD signal attenuates as distance increases from the radar site (Doviak and 

Zrnić 1984), which can lead to poor performance on data collection. Luckily for 

 

58 

this study, one of the six radar stations covering Michigan is located very close to 

the central-eastern boundary of our study area (Station KGRR, Figure 2.1).  

2.2.3 Model Implementation 

PAWS is a physically-based hydrologic model, which incorporates overland, 

unsaturated vadose zone, saturated groundwater and channel flows (Shen and 

Phanikumar, 2010) that cover various flow paths in the watershed. In addition, 

PAWS coupled with the Community Land Model (CLM) 4.0 provides detailed, 

process-based vegetation dynamics (Thornton and Zimmermann, 2007; Niu et al. 

2014) while taking the canopy interception, snowpack, biomass, depression 

storage and evapotranspiration into consideration (Lawrence et al., 2011; Shen 

et al., 2013). Considering the entire hydrology of a watershed system, PAWS 

allows us to capture the dynamic nature of surface and groundwater exchange 

driven by the head gradient, thus to evaluate recharge by quantifying all of the 

recharge-relevant hydrologic processes in the system.  

Two PAWS models were built respectively for using the gauge precipitation data, 

and the combined gauge and NEXRAD MPE precipitation products (also 

mentioned as gauge-MPE in this article). The discharge information from the 

USGS 04119000 station (Figure 2.1 and Table 2.1) which is located at the 

junction between Grand River and watershed boundary served as the upstream 

inflow for the model. The grid size of the model is 300 m by 300 m which enables 

us to capture the detailed spatial variability of the simulated recharge. We used 

hourly time steps for weather, vegetation, and evapotranspiration simulations, 10 

minute time steps for channel flow, and overland flow, and adaptive time steps 

 

59 

ranging from 2 min to 1 h for the unsaturated vadose zone. The results were 

aggregated to daily, monthly and yearly time steps for further analysis.  

In this study, we first calibrated the model based on precipitation data obtained 

from the weather stations only. The same calibration parameters and model 

inputs except for the precipitation data were used for each of the models. 

Therefore, deviations in streamflow, ET, and recharge can be expected to be 

based solely on the precipitation variations. The daily streamflow rates for the 

period of 08/01/2009–12/31/2010 from the USGS 041190400, USGS 04108800, 

and USGS 04108600 gauges (Figure 2.1, Table 2.1) which are scattered across 

the county were used to calibrate the PAWS model. The parameter optimization 

using the Differential Evolution (DE) algorithm (Price et al., 2005) was performed 

at the High-Performance Computing Center (HPCC) at Michigan State 

University. We used the streamflow rates at these USGS gauges from 

01/01/2011 to 11/22/2015 to validate the model. Further, the measured 

streamflow rates from fieldwork spatially are used to validate the model 

performance on the streamflow rates in small streams. In addition, the MODIS 

ET products are used to validate the simulated ET spatially and temporally. The 

comparison between the simulated and wellogic groundwater head information 

verifies the model performance on the subsurface flow processes. All the 

analyses of this article are based on the period between January 2011 and 

November 2015. 

 

 

 

60 

Table 2.1 U.S. Geological Survey (USGS) gauging stations 
 

Gauge Number 

Gauge Name 

 04119000 
 04119400 
 04108800 
 04108600 

Grand River at Grand Rapids, MI 
Grand River near Eastmanville, MI 
Macatawa River near Zeeland, MI 

Rabbit River Near Hopkins, MI 

Latitude 
42.9644 
43.0242 
42.7792 
42.6422 

Longitude 
-85.6764 
-86.0264 
-86.0183 
-85.7219 

 

2.3 Assessment of the NEXRAD MPE precipitation 

2.3.1 Comparisons of the Gauge and NEXRAD MPE Precipitation 

Two daily precipitation datasets were collected from the weather stations and the 

NEXRAD weather radars from 2009 to 2015. Twenty-two weather stations are 

inside the watershed area, including the NCDC and MAWN rain gauges. The rain 

gauges are unevenly distributed in the watershed as most are concentrated 

around the urban areas. And the gauges have a density of 0.012 gauges/km2 in 

the study area, which is much better than the average gauge density in the 

United States (0.0013 gauges/km2) (Linsley 1992).The NEXRAD MPE daily 

precipitation products with a 4 km spatial resolution are adjusted by the gauge 

precipitation with high accuracy (Price et al., 2011). To assess the quality of the 

NEXRAD MPE daily precipitation data, we compared the NEXRAD MPE 

precipitation products with the corresponding gauge-based precipitation data 

close to the only three USGS gauging stations which provide streamflow data for 

model calibration and validation. The weather stations are US1MIOW0030, 

US1MIAN0001, and USW00004839, which are labeled as W3, W2, and W1 in 

Figure 2.1. The comparisons between the gauging precipitation and MPE 

precipitation products over 2011-2015 are presented in the scatterplots in Figure 

 

61 

2.2. The daily precipitation rate comparisons are scattered around the 1:1 line at 

the weather station US1MIOW0030 and weather station US1MIAN0001. 

However, NEXRAD MPE shows some extreme high values of precipitation when 

the gauging precipitation is small. And, the precipitation comparison scatterplot at 

the weather station USW00004839 is also scattered around the 1:1 line but with 

a relatively larger cloud.  

 

Figure 2.2 Scatter plots of the gauge and the NEXRAD MPE precipitation rates 
over the period of 2011-2015. (Unit: mm/day) 
 
 
Further, we used several metrics to measure the relationship between the 

gauges and the MPE precipitation datasets. Specifically, three metrics are used: 

(1) Pearson’s correlation coefficient (, Eq. 2.1) is a measurement of the linear 

correlation between the two precipitation datasets; (2) percent bias (B, Eq. 2.2) is 

used to examine the systematic differences between the two datasets; and (3) 

relative normalized root mean square differences (NRMSD, Eq. 2.3) is a 

measure of the random differences between the two datasets. 

 

62 




n



i

1


(

R R G G
i

)(





i

)

n



i

1


(

)
R R
i



2

n



i

1


(
G G
i



)

2

 

 

 

 

  

n



i

1


B



(

R G
i
i



2

)

n



i

1


G
i

(100)



 

 

 

 

 

 

 

 

 

 

 

 

(Eq. 2.1) 

(Eq. 2.2) 

n



1
n

(

R G
i
i



2

)

i

 

 

1




  

G

NRMSD

Where,  and  represent the  th day precipitation rate of rain gauges and 
MPE, respectively (mm/day);   and ̅ are the mean values of precipitation over 
the whole study period;  is the number of gauge and weather radar pairs. ρ is 

(Eq. 2.3) 

 

 

 

the Person’s correlation coefficient; B is the percent bias; NRMSD is the relative 

normalized root mean square difference. Table 2.2 shows a summary of the 

metrics for the NEXRAD MPE precipitation data against the gauge precipitation 

data at the three selected USGS gauges between January 2011 and November 

2015. It shows that the Pearson’s correlation between the NEXRAD MPE 

precipitation and the gauge precipitation was high at weather station 

US1MIAN0001 and weather station US1MIOW0030, and relatively low at 

weather station USW00004839. The percent bias and the  presents the 
same trend as the Pearson’s correlation, the values of bias and   at 

weather station US1MIAN0001 and weather station US1MIOW0030 were small, 

 

63 

and the values of bias and  at weather station USW00004839 was the 

largest among these three weather stations. Generally speaking, we see a strong 

correlation and little bias between the two data sets at the US1MIAN0001 and 

US1MIOW0030 weather stations. The statistical analysis illustrates that the 

values of NEXRAD MPE precipitation were close to the gauge precipitation at 

US1MIAN0001 and US1MIOW0030 weather stations and had a significant 

difference from the gauge precipitation at USW00004839.  

Table 2.2 Statistical analysis of MPE precipitation based on the gauge 
precipitation at the three USGS gauges. 
 

 

USW00004839 
US1MIOW0030 
US1MIAN0001 
 

 
0.70 
0.90 
0.91 

B (%) 
37.84 
15.65 
15.93 

N R M S D  

2.07 
1.07 
1.03 

This result is consistent with the processing of NEXRAD MPE precipitation 

products which are adjusted based on the gauge precipitation. Precipitation pairs 

at US1MIOW0030 and US1MIAN0001 weather stations which are about 27~33 

km away from the NEXRAD site have a strong correlation. And, precipitation 

pairs at the USW00004839 weather station which is about 45km away from the 

NEXRAD Site – KGRR, has a relatively poor correlation, but the Pearson’s 

correlation remains quite high between the two data sets. The relatively smaller 

correlation might be caused by the long distance away from the radar site. As 

mentioned earlier, NEXRAD signal attenuates as distance increases from the 

radar site.  

 

64 

2.3.2 Comparisons of the Streamflow 

As discussed earlier, quantifying the uncertainties of precipitation data obtained 

from the rain gauges and NEXRAD MPE is not straightforward. In this study, we 

also compared the observed and the simulated discharge to aid in evaluating the 

use of these two precipitation datasets in our physically-based hydrologic model 

(Figure 2.3). This approach is one of the predominant approaches to evaluate 

precipitation data sources (Neary et al., 2004; Kalin and Hantush, 2006; Sexton-

Sims et al., 2010; Price et al., 2011). Also, the base flow is a proxy of recharge 

and therefore, accurately simulated base flow indicates that the model has the 

potential to accurately simulate recharge. 

The criteria used to examine the streamflow (discharge) estimation performances 

are (1) the coefficient of determination (R2, Eq. 2.4); (2) the root-mean-square 

error (RMSE, Eq. 2.5); and (3) the square-root-transformed NASH metric 

(RNASH, Eq. 2.6) which is designed to emphasize the assessment of the 

baseflow (Shen and Phanikumar, 2010) as a proxy of recharge since the NASH 

metric is biased toward peak flows.  

n



i

1


((
)(
O O S
i



i



S

))

n



i

1


(
O O



i

2

)

n



i

1


(

S

i



2

S

)




 




2

R

 

RMSE



1
n

S O
i
i
n 
1
i

(

2

)

 

 

 

2









  

 

 

 

65 

 

 

 

 

 

 

(Eq. 2.5) 

(Eq. 2.6) 

(

O
i



S

i

2

)

2

n

i




1

n

i

1


RNASH

1
 

(

)



O
i

O
i

Where,  is the th observed value,  is the mean of the observed data,  is 
the th simulated value, ̅ is the mean simulated value, and  is the total number 

(Eq. 2.7) 

 

 

 

 

 

 

of events. 

Figure 2.3, Figure 2.4, and Figure 2.5 show the comparisons between the 

observed discharge and the two sets of model-simulated discharge with the 

gauge precipitation data and gauge-MPE precipitation data at three USGS 

gauges. The metrics used to test the comparisons with the two above-mentioned 

precipitation datasets were summarized in Table 2.3. Both simulated sets of 

results of discharge with the gauge-based precipitation and the gauge-MPE-

based precipitation data show agreements with the observed discharge at the 

USGS 04119400 station on the Grand River. The R2 (0.96), RNASH (0.98) 

between the simulated and observed streamflow using gauge precipitation data 

are very high, while the R2 (0.97) and RNASH (0.97) are also very high based on 

the gauge-MPE precipitation data. Therefore, we can argue that both models 

have produced outstanding performance at USGS 04119400. Close inspection of 

Figure 2.3 reveals that the simulation conducted with the gauge-MPE 

precipitation data led to an overestimated discharge value in 2015, especially in 

the fall. However, the discharge simulation at the USGS 04119400 station has a 

relatively good performance because of the significant contribution of the input 

upstream inflow data provided by the UGSG 04119000 station which is at the 

 

66 

intersection of Grand River and the watershed boundary (point of entry of the 

Grand River into the Ottawa County used as a boundary condition in the model). 

Comparing the NEXRAD MPE precipitation to the gauge precipitation for the 

period from January 2015 to November 2015, the values of NEXRAD MPE 

precipitation were consistently larger than the values of gauge precipitation 

(Figure 2.3). As we take the conventional gauge precipitation data as the 

standard input for the model, the NEXRAD MPE data are consistent with the 

gauge data in the years between 2011 and 2014. But we do detect some 

disagreement for the year 2015, further research is needed to investigate the 

causes and explanations of this unexpected situation.   

Figure 2.4 and Figure 2.5 also illustrate that the simulation conducted with the 

gauge-MPE precipitation data led to an overestimated discharge value in 2015. 

The metrics from our statistical analyses illustrate that the discharge simulation 

conducted with the gauge precipitation data has a higher correlation coefficient 

with the observed discharge than that conducted with the gauge-MPE 

precipitation data at the USGS 04108800 station. The R2 of the gauge 

precipitation simulated-observed discharge is 0.51 which is acceptable indicating 

a strong correlation with values larger than 0.5 (Moriasi et al., 2007). The R2 is 

0.42 when we use gauge-MPE precipitation data (Table 2.3). The RNASH has 

the same trend as the R2 as the value is 0.66 for using the gauge precipitation 

data and 0.54 with the gauge-MPE precipitation data (Table 2.3). The USGS 

04108800 station did not have discharge records for the period from January 

2015 to mid-March 2015. The comparison between the simulated and observed 

 

67 

discharge data at the USGS 04108600 station also shows that the R2 and 

RNASH are smaller when we combine with NEXRAD MPE precipitation data 

(Table 2.3). When we further recalibrated the model based on a combination of 

the gauge precipitation and NEXRAD MPE precipitation data, we still observe the 

general offsetting trend of the streamflow estimation in 2015 (APPENDIX A). We 

suspect that the NEXRAD MPE has some data offset in the year of 2015, but 

further research is needed to understand this discrepancy in the NEXRAD MPE 

dataset fully.   

Since 2015 was the only year we observed a large offset in our analysis, we also 

summarized the comparison for the years between 2011 and 2014. The metrics 

for this period are summarized in Table 2.4. Both the R2 and RNASH increase 

significantly based on the NEXRAD MPE precipitation data. The R2 with the 

NEXRAD MPE precipitation increase from 0.43 to 0.53 at the USGS 04108600 

station. Moreover, the RNASH increase from 0.28 to 0.38. And the R2 and 

RNASH with the NEXRAD MPE precipitation at the USGS 04108800 station 

increase from 0.42 to 0.46, and from 0. 54 to 0.67, respectively.  

Based on the series of comparisons drawn between the simulated discharge 

using two data sources and the observed discharge, we can argue that the 

NEXRAD MPE data leads to relatively similar performance as the conventional 

gauge-based precipitation data in our study area. However, in some years, the 

similarity is not robust as consistent overestimated results have been observed 

for one of the five years in the model.  

 

 

68 

Figure 2.3 Comparisons between observed and simulated discharge at USGS 
04119400. (The dark green line depicts the simulation conducted with gauge-
MPE precipitation; the dark red line describes the simulation conducted with 
gauge precipitation. The presented simulation period is from January 2011 to 
November 2015. The selected window period is from January 2015 to November 
2015.) 
 

 

 

 

69 

Figure 2.4 Comparisons between observed and simulated discharge at USGS 
04108800. (The dark green line depicts the simulation conducted with gauge-
MPE precipitation; the dark red line describes the simulation conducted with 
gauge precipitation. The presented simulation period is from January 2011 to 
November 2015. The selected window period is from January 2015 to November 
2015.) 
 

 

 

70 

Figure 2.5 Comparisons between observed and simulated discharge at USGS 
04108600. (The dark green line depicts the simulation conducted with gauge-
MPE precipitation; the dark red line describes the simulation conducted with 
gauge precipitation. The presented simulation period is from January 2011 to 
November 2015. The selected window period is from January 2015 to November 
2015.) 
 
 

 

 

 

71 

Table 2.3 Statistical analyses of the discharge simulation performance at three 
USGS gauges for the period from January 2011 to November 2015. 
 

 

 
R2 

RMSE 
RNASH 

04119400 

04108800 

04108600 

Gauge 
0.96 
21.06 
0.98 

MPE 
0.97 
20.22 
0.97 

Gauge 
0.51 
3.75 
0.66 

MPE 
0.42 
3.99 
0.54 

Gauge 
0.53 
1.78 
0.4 

MPE 
0.43 
1.89 
0.28 

 
 
Table 2.4 Statistical analyses of the discharge simulation performance at three 
USGS gauges for the period from January 2011 to December 2014. 
 

04119400 

04108800 

04108600 

Gauge 
0.98 
22.95 
0.98 

MPE 
0.98 
20.76 
0.98 

Gauge 
0.54 
3.66 
0.68 

MPE 
0.46 
3.86 
0.67 

Gauge 
0.55 
1.82 
0.45 

MPE 
0.53 
1.88 
0.38 

 
 
R2 

RMSE 
RNASH 

 

2.4 Results 

2.4.1 The Spatial Distribution of Recharge 

The spatial maps of the average annual precipitation and the average annual 

recharge obtained based on the gauge precipitation data and the NEXRAD MPE 

precipitation data between 2011 and 2014 are shown in Figure 2.6. The spatial 

variabilities of recharge are affected by the spatial characteristics of precipitation, 

land use and land cover (LULC), terrain characteristics, and soil types. Figure 2.6 

shows that the general spatial patterns of the simulated recharge using both 

precipitation datasets roughly follow the spatial pattern of elevation (DEM), 

whereby high values of recharge occur in places with high elevation, and low 

values of recharge occur in places with a relatively low elevation. Agricultural 

 

72 

land generally has larger values of recharge than forested land does for the 

recharge maps produced based on both precipitation datasets. Also, we can 

clearly see that the precipitation map obtained from the NEXRAD MPE is more 

fine-grained than the gauge-based precipitation map. On the one hand, this 

situation reflects the high spatial resolution of the NEXRAD MPE dataset, and on 

the other hand, the NEXRAD MPE offers the possibility to study and understand 

the impacts of precipitation on recharge on a fine scale. And further, if NEXRAD 

MPE data lead to consistent results with the conventional gauging data, the finer-

resolution maps on various hydrologic variables will enable various water 

resource management projects to be implemented on a fine scale. 

 

73 

 

 

Figure 2.6 The spatial maps of average annual precipitation and average annual 
recharge based on gauge and gauge-MPE precipitation over period 2011-2014. 
 
 
The northern part of the study area with high estimated recharge values, the 

values of the recharge estimated with the NEXRAD MPE precipitation were 

consistently higher than the values of recharge obtained with the gauge 

precipitation data which is probably because the values of the NEXRAD MPE 

precipitation (between 1050 mm/year and 1200 mm/year) were much higher than 

 

74 

the values of gauge precipitation values (between 950 mm/year and 980 

mm/year) in this area. The impact of high precipitation from the NEXRAD MPE 

dataset on recharge can be seen in the estimated recharge value in the northern 

part of the study area. The areal averaged annual precipitation based on the 

gauge precipitation data, and the NEXRAD MPE precipitation data were 911.0 

mm/year and 968.8 mm/year, respectively. The areal averaged annual recharge 

obtained with the gauge precipitation and NEXRAD MPE precipitation was 112.1 

mm/year and 125.0 mm/year. The values of the NEXRAD MPE precipitation 

were 1.06 times larger than the values of gauge precipitation. However, the 

values of recharge estimated with the NEXRAD MPE precipitation data were 

1.12 times higher than the values of recharge estimated with the gauge 

precipitation data. Since all the other factors are held constant, the difference in 

the estimated recharge is attributed solely to the difference in the precipitation 

data inferring that the higher precipitation values presented in the NEXRAD MPE 

precipitation data lead to a higher estimation of the recharge.  

2.4.2 Temporal Change of the Areal Recharge Using Two Precipitation Data 

Figure 2.7 presents a side-by-side comparison of both the monthly areal 

precipitation (bars on the top) and estimated monthly areal recharge values 

(lines) of the two precipitation datasets: gauge data and the NEXRAD MPE data. 

The changes of the estimated recharge based on the two precipitation datasets 

follow the cycle of the hydrologic year starting from October and ending in 

September of the following year. The values of the simulated monthly areal 

recharge based on two precipitation datasets were similar over the period from 

 

75 

September 2011 to November 2013. The estimated recharge with the NEXRAD 

MPE precipitation data was larger than the one derived with the gauge 

precipitation data after December 2013. In 2014, the values of the estimated 

recharge using the NEXRAD MPE precipitation data were consistently larger 

than the values from the gauge precipitation data, while the temporal patterns of 

the simulated recharge remained similar. In 2015, simulated recharge with the 

NEXRAD MPE precipitation data kept increasing with a fluctuating precipitation 

data. 

When we compare the values of the gauge precipitation data to the values of the 

NEXRAD MPE precipitation data, the values were similar from September 2011 

to September 2013 (Figure 2.7). However, the NEXRAD MPE precipitation data 

were smaller than the values of gauge precipitation when the precipitation was 

heavy, while the NEXRAD MPE precipitation data were greater than the values 

of gauge precipitation when the precipitation was light. In previous studies, 

NEXRAD MPE precipitation was found to have underestimated heavy rain and 

overestimated light rain (Jiang et al. 2006; Skinner et al., 2009). The values of 

NEXRAD MPE precipitation were larger than the values of gauge precipitation 

starting from October 2013, the values of the NEXRAD MPE precipitation were 

significantly and consistently greater than the values of the gauge precipitation in 

2015. The large values of precipitation made a significant contribution to the peak 

values of recharge. In a hydrologic year from October 2013 to September 2014, 

the annual gauge precipitation was 924.9 mm. The NEXRAD MPE-based areal 

precipitation over the period from October 2013 to September 2014 was 

 

76 

1092.6mm. But for the year of 2015, the cumulative precipitation from January to 

November was 663 mm based on the gauging data and 1217 mm from the 

NEXRAD MPE. The values of the NEXRAD MPE precipitation were much larger 

than the values of gauge precipitation in the year 2015. Given the constraint of 

this study, future research is needed to investigate the significant discrepancy 

between the two datasets. 

Figure 2.7 The comparison between gauge and gauge-MPE precipitation conducted 
areal monthly recharge.  
 

 

Figure 2.8 describes the seasonal changes of the estimated recharge based on 

the gauge precipitation data and the NEXRAD MPE precipitation data. Here, 

seasons were defined as northern meteorological seasons. A calendar year is 

divided into four meteorological seasons of three months each: Spring is from 

March 1 to May 31; Summer is from June 1 to August 31; Fall is from 

September 1 to November 30; and, Winter is from December 1 to February 28 

(February 29 in a leap year). The seasonal changes of the estimated recharge 

using gauge precipitation data (Figure 2.8) indicate that winter was the primary 

recharge season and summer was the low recharge season except for the 

 

77 

hydrologic year of 2012-2013 right after the drought year of 2012. The values of 

the estimated recharge increased from fall 2012 to spring 2013 and then 

decreased until fall 2013. The large values of the estimated recharge in Spring 

were contributed by the extremely high precipitation in April 2013, and this impact 

lasted until fall 2013. The seasonal changes of recharge with the NEXRAD MPE 

precipitation had a similar trend as one using the gauge precipitation data. But in 

2015, there are large differences in precipitation from the two data sources. The 

intense high values of precipitation from the NEXRAD MPE precipitation data in 

the spring and summer of 2015 (Figure 2.7) leads to an extremely high estimated 

recharge values in the spring and summer of 2015. But the estimated recharge 

based on the conventional gauge data illustrates relatively stable and consistent 

seasonal changes between 2011 and 2015. We suspect errors in the 2015 

NEXRAD MPE data, but further research is needed to fully understand the 

discrepancies between the two precipitation datasets from these two sources.   

 

Figure 2.8 The comparison between estimated seasonal recharge based on 
gauge and NEXRAD MPE precipitation data. (X-axis is time, “W” indicates winter, 
“S” indicates Summer) 
 

 

 

78 

2.4.3 Water Budget 

To further evaluate and compare the model performances using the gauge 

precipitation data and the NEXRAD MPE precipitation data, this study also 

calculated the water budget between 2011 and 2014. Table 2.5 summarizes the 

annual water budget in a typical drought year of 2012, in the typical wet year of 

2014, and the average annual water budget over the period from 2011 to 2014. 

The annual water budget of 2012 and 2014 and the average annual water budget 

between 2011 and 2014 were all 5% to 20% larger than the values of precipitin 

which is acceptable because of the inputs/outputs or storage mechanisms in the 

watershed. In the drought year of 2012, the areal annual gauge precipitation was 

755.1 mm/year, while the areal annual NEXRAD MPE precipitation was 

728.8mm/year. The ratios of recharge (symbol – Dperc, which stands for "deep 

percolation" another term of recharge) to precipitation are the same in the year 

2012. But, in the wet year of 2014, the ratio of recharge to precipitation based on 

the NEXRAD MPE precipitation (16.4%) was larger than that with the gauge 

precipitation data (14.3%). The areal annual gauge precipitation was 893.9 

mm/year in 2014, while the areal annual NEXRAD MPE precipitation was 

1072.8mm/year in 2014. This observation confirms the study by Zhang et al. 

(2016), who argued that the extreme values of precipitation have stronger 

impacts on recharge estimation. 

Further, we also compared the simulated ET values in the water budget by using 

the two different precipitation data. The percentages of ET based on the two 

datasets appeared to be similar in 2012 water budget, and the averaged four-

 

79 

year water budget. The annual areal gauge precipitation in 2012 was 755.1 

mm/year, which is close to the value (728.8mm/year) of NEXRAD MPE 

precipitation in 2012. Moreover, the averaged annual areal gauge precipitation 

(911.0 mm/year) was also similar to the averaged annual areal NEXRAD MPE 

precipitation (968.8 mm/year) for the period of 2011-2014. The percentage of ET 

appeared to be similar smaller when we used the NEXRAD MPE data than that 

from the gauge precipitation data in 2014. But the values of the simulated ETs 

were similar based on the two precipitation datasets, 542.6mm/year with the 

gauge data and 560.0 mm/year with the NEXRAD MPE precipitation data which 

is because that annual areal precipitation of the gauge precipitation (893.9 

mm/year) was much smaller than that of the NEXRAD MPE precipitation (1072.8 

mm/year) in 2014. However, the ratios of recharge to precipitation conducted 

with the two precipitation datasets for the water budget of 2012, 2014 and the 

average four-year period were similar. In 2014, the high precipitation from 

NEXRAD MPE produced large values of recharge. To sum up, ET estimation, 

taking up a large portion of the water budget, was not as sensitive to the high 

precipitation as the estimated recharge was, but recharge has a significant 

response to the high values of precipitation. 

 

 

 

80 

Table 2.5 Annual water budgets (%P, percent of annual precipitation) 
 

            2012 

                2014 

       Ave annual 
MPE 
Gauge 
100 
       100 
       60.9 
61.9 
       12.9 
12.3 
25.8 
       23.7 
         2.4 
1.0 
4.1 
         5.5 
      105.4 
105.1 

 
 
P 
ET 

Dperc 
Qoc 
Qgc 
SUBM 
Total 

Gauge 
100 
75.9 
13.0 
22.3 
  1.1 
  1.9 
114.2 

    MPE 
      100 
     78.6 
     13.0 
     18.4 
       2.7 
       3.4 
    116.1 

    Gauge 
         100 
        60.7 
        14.3 
        27.2 
          1.1 
          5.8 
      105.1 

   MPE 
   100 
  52.2 
  16.4 
  30.9 
    2.5 
    7.5 
110.9 

P: precipitation, Dperc: percentage of recharge in the water budget, Qoc: surface runoff 
percentage in the water budget, Qgc: groundwater outflow percentage in the water budget, 
SUBM: sublimation percentage in the water budget. 
 

2.5 Discussions and Conclusions 

This study compared and examined two datasets for obtaining the precipitation 

data from the conventional gauge stations and the radar-based NEXRAD. We 

used a physically-based hydrologic model, PAWS, to test the performance of 

these two datasets on recharge estimation in the watershed around the Ottawa 

County, MI during the period of 2011 – 2015. The obvious advantages of the 

NEXRAD dataset lie in its high spatial and temporal resolution, with the complete 

coverage of the U.S. But the current research has revealed that the NEXRAD 

data may have discrepancies with the conventional gauge-based precipitation 

data during some periods. This study used PAWS, a physically-based 

hydrological model, to examine how NEXRAD precipitation data differed from or 

agreed with the conventional gauge-based precipitation data, and further to 

investigate how the two precipitation datasets performed in estimating the 

discharge, the spatial patterns of recharge, and the time-series recharge, and 

water budgets. 

 

81 

First, we compared the precipitation data from both sources at three weather 

stations over the period of 2011-2015. The results show that the two datasets 

have a high correlation at the two weather stations which are relatively close to 

the existing NEXRAD radar station. This confirms the general consensus of the 

radar-based NEXRAD data in that the farther the location is away from the radar 

station, the lower the accuracy of the data. However, although we have found 

discrepancies between the two datasets, their correlation is consistently strong at 

all three weather stations. Thus, we argue that NEXRAD MPE datasets would 

serve as a good supplement to the conventional gauge precipitation data for 

locations not too far away from the radar station.  

Second, we compared the estimated streamflow data using two datasets with the 

observed streamflow data at the three USGS gauge stations. The results show 

that the estimated streamflows have a relatively strong correlation with the 

observed data. Generally, precipitation data from both sources produce a similar 

performance in the streamflow estimation in our model with the exception of the 

year 2015.  

Further, the maps of the average annual estimated recharge based on the two 

precipitation datasets over period of 2011-2014 show very similar patterns in the 

study area. When we examined the average annual recharge maps, we noted 

that the areas with higher precipitation values from the NEXRAD MPE data 

caused the higher estimated average annual recharge. However, both recharge 

estimation maps based on the two datasets have similar spatial patterns, which 

further indicates that the NEXRAD data do not have a consistent spatial bias in 

 

82 

estimating recharge in a physically-based hydrologic model. Thus, NEXRAD 

MPE data opens up the possibility to study and understand the impacts of 

precipitation on recharge.   

The examination of the time-series of the areal monthly recharge based on the 

two precipitation datasets shows that the two monthly recharge time-series agree 

with each other between 2011 and 2014. But there appears to be a sharp 

difference between the two in the year 2015. Further investigation reveals that 

the NEXRAD MPE precipitation data is systematically and consistently higher 

than the gauge-based precipitation data. The difference can be clearly seen in 

Figure 2.7, and we suspect an error in the NEXRAD MPE data for 2015, but 

further research is needed to understand the exact reasons for this difference. 

The comparison of the estimated seasonal recharge shows similar patterns in 

that the two estimations roughly agree with each other before 2015, but the 2015 

estimation is also clearly affected by the high values in the NEXRAD MPE data. 

Last but not least, the total water budget is consistent within the acceptable 

range for both two precipitation datasets. The estimated water budget 

composition also reveals that the percentages of various estimated components 

in the water budget are similar using two precipitation datasets. However, the 

extreme values of precipitation have stronger impacts on recharge estimation. 

Thus, all in all, the precipitation datasets from the conventional gauge stations 

and from the radar-based NEXRAD MPE have shown a strong correlation with 

each other. The strongest correlation of the precipitation values occurs at 

locations that are closer to the radar station. Spatially, the two datasets produce 

 

83 

similar patterns of the estimated recharge for the study region. Also, temporally, 

the two datasets show a strong correlation between 2011 and 2014. Therefore, 

we argue that NEXRAD MPE generally provides a reasonable estimate of 

precipitation for hydrologic modeling in areas close to the radar stations. The 

high spatial resolution of the NEXRAD data also provides the possibility to help 

implement water management plans in the area with sparse weather 

stations/with weather stations which have missing records. 

 

 

 

84 

CHAPTER 3 

Effects of Grid Resolution on Groundwater Recharge 

Estimation Using a Physically-Based Hydrologic Model 

 

Results of numerical modeling based on distributed hydrologic models are 

sensitive to the grid size adopted, as the local climate and physical 

characteristics of the study area, such as precipitation, land use, and elevation 

can vary significantly based on different cell sizes in a physically-based 

hydrologic model. The grid size used in a model not only influences the overall 

representation of the spatial variability of the physical characteristics, but also 

lead to large differences in recharge estimation. The grid size in a model is also 

closely related to the computational efficiency and computer storage which are 

essential factors in numerical modeling. In this study, we used a physically-based 

hydrologic model, PAWS, to examine the effects of four grid sizes, 300 m, 500 m, 

700 m, and 1 km on the recharge estimation in Ottawa County, Michigan 

between 2011 and 2014. The results showed that the values of estimated 

recharge were related to grid sizes while the spatial and temporal patterns of 

recharge remained consistent for the four grid sizes. Finer grid size can capture 

the detailed spatial variabilities of recharge as the baseflow comparisons showed 

that the recharge was more accurately simulated with a grid size less than 500 

m. Moreover, the analysis of scaling effects illustrated that a finer grid size was 

able to represent the spatial continuity in recharge as well. These findings 

 

85 

demonstrated that a grid size below 500 m well captured the spatial variabilities 

of the recharge in the study domain. The spatial and temporal patterns of annual 

and monthly recharge were related to precipitation, Land Use and Land Cover 

(LULC) types, and topography as the recharge was more strongly related to 

precipitation at a coarser grid size. The analysis of the grid size effects on 

recharge over different LULC types demonstrated that the impacts of different 

LULC types on recharge were not significant across the four grid sizes.  

3.1 Introduction 

Groundwater is extracted for agricultural, industrial, and municipal uses globally 

and remains as the primary source of freshwater for approximately one-third of 

the world population (Alley, 2006; Giordano, 2009; Gleeson, Wada, Bierkens, & 

Beek, 2012; Kundzewicz & Döll, 2009; Richey et al., 2015; Siebert et al., 2010). 

Groundwater is also a key factor to maintain the composition and functions of 

groundwater-dependent ecosystems to regulate the water levels, water 

temperature and nutrition (Brown, Bach, Aldous, Wyers, & DeGagné, 2011; 

Howard and Merrifield, 2010; Murray et al., 2003). However, over-extraction and 

contamination of groundwater also lead to problems, such as aquifer depletion, 

land subsidence, cessation of base flow, and damaged natural ecosystems 

(Aeschbach-Hertig & Gleeson, 2012; Brown et al., 2011; Rodell, Velicogna, & 

Famiglietti, 2009). Groundwater recharge is the sustainable source for recharging 

and cleaning groundwater and plays an integral role in surface and sub-surface 

water resources (Hartmann, Gleeson, Wada, & Wagener, 2017; Vries & 

Simmers, 2002). Therefore, understanding and accurately estimating the natural 

 

86 

recharge is essential for sustainable management and preservation of the 

groundwater for the human activities and natural ecosystems. In particular, 

estimations of recharge are important to groundwater modelers because 

recharge is one of the primary inputs to groundwater modeling studies (Jyrkama, 

Sykes, & Normani, 2002). 

Groundwater recharge is a difficult component to be quantified in the hydrologic 

budget. Sparse information and indirect interpretation of observed groundwater 

recharge make the determination of recharge uncertain. Moreover, the 

heterogeneity of the climate and the land-surface and subsurface properties, 

such as land use, topography, hydraulic conductivities or porosities, increases 

the uncertainty of the recharge estimation (Glenn, Jarchow, & Waugh, 2016; 

Hartmann et al., 2017; Niazi, Bentley, & Hayashi, 2017). Physically-based 

integrated hydrologic models offer a relatively efficient and reliable way to 

estimate recharge by quantifying all the relevant hydrologic properties 

contributing to recharge (Awan & Ismaeel, 2014; Bailey, Rathjens, Bieger, 

Chaubey, & Arnold, 2017, 2016; Camporese, Paniconi, Putti, & Orlandini, 2010; 

Dakhlalla, Parajuli, Ouyang, & Schmitz, 2016; Jyrkama et al., 2002; Kollet & 

Maxwell, 2006; Paniconi & Wood, 1993; Scanlon & Cook, 2002; Shen & 

Phanikumar, 2010; VanderKwaak & Loague, 2001). Physically-based hydrologic 

models describe the vegetation dynamics, overland runoff, streamflow, soil 

water, groundwater flow, and energy cycle as well. However, due to the inherent 

nonlinearity of the processes, model results are sensitive to how spatial patterns 

of climate, and watershed physical characteristics are defined and represented 

 

87 

on a grid cell in the model (Mo, Liu, Chen, Lin, Guo, & Wang, 2009; Molnár & 

Julien, 2000; Rojas, Velleux, Julien, & Johnson, 2008). The grid size used in a 

model would affect the overall representations of variability in physical 

characteristics ( Molnár & Julien, 2000).  

Grid-size effects on the major components of the hydrological cycles are 

complex. Existing studies have shown the grid-size effects on various 

hydrological components, such as the evapotranspiration (ET), soil moisture, and 

surface runoff, and illustrated that the sensitivity of ET to grid size is smaller than 

that of soil moisture and runoff to grid size (Cerdan et al., 2004; Kuo et al., 1999; 

Hessel, 2005; Liang, Guo, & Leung, 2004; Mo et al., 2009; Molnár & Julien, 

2000). Zhang et al. (2008) focused on a 12 km2 study area and illustrated that a 

coarser grid size (100-200 m) better captures the peak flow than a finer grid size 

(10-30 m) does. However, the grid-size effect on recharge is still yet to be 

studied.   

Grid sizes also contribute to scaling issues over the heterogeneous land surface 

(Sridhar, Elliott, & Chen, 2003). A coarse grid size may make the spatial variation 

in the model hidden in the representation (Kuo et al., 1999; Schoorl, Sonneveld, 

& Veldkamp, 2000). In recent years, high-resolution remote sensing and GIS 

hydrologic products contribute to advances in hydrological modeling and provide 

us more options when a wide diversity of models are applied over a broad range 

of different scales (Sulis, Paniconi, & Camporese, 2011). The grid sizes for 

different scales of watersheds can vary from meters to kilometers. (Bormann, 

2006) argued that there might be a critical grid size at which the scaling effects 

 

88 

became significant, therefore, an appropriate grid size should also take the scale 

of the study area into consideration. 

Nevertheless, grid size choices are also closely related to the computational 

efficiency and may require large computer storage space, which is another 

essential factor in the numerical models (Molnár & Julien, 2000; Vieux, 2016). 

While a smaller grid size can significantly increase the accuracy to describe the 

surface-subsurface heterogeneity and further estimate detailed recharge in the 

model; as a tradeoff, a smaller grid size also requires longer computation time 

and a higher demand for computer storage. In general, the basic strategy to find 

a suitable grid size on a watershed scale is to make a proper compromise 

between estimation accuracy and computation efficiency.  

While it is clear that understanding the effects of grid size in recharge estimation 

is very important especially in the physically-based hydrological models for both 

modeling and groundwater management practices, the grid size effect on 

recharge estimation remains an understudied topic in the literature. This study 

used an integrated surface-subsurface hydrologic model, PAWS + CLM, to 

evaluate the grid-size effects on recharge estimation in the Ottawa County, MI.  

PAWS is a grid-based hydrologic model with efficient computational algorithms, 

which enables the simulations ranging from coarse grid size to fine grid size. The 

model was calibrated using streamflow rates from USGS gauges over the period 

2009-2010 and further validated using streamflow rates from USGS gauges 

between 2011-2015, streamflow rates measured in 2015, MODIS 

evapotranspiration (ET), and wellogic groundwater head datasets. The objectives 

 

89 

of this study are to (1) examine the effects of grid size variation on the spatial and 

temporal patterns of recharge simulation; (2) analyze the precipitation and LULC 

impacts on recharge estimation using four grid sizes of 300 m, 500 m, 700 m, 

and 1 km; (3) evaluate the grid size selection for physically-based hydrological 

model.  

In the next section, we illustrate the modeling domain, the PAWS approach of 

modeling recharge, and the model input datasets. The third section presents our 

results showing the grid-size effects on the spatial and temporal patterns of 

recharge in the Ottawa County, MI. And, we also examine the impacts of various 

LULC on recharge simulation over the four grid sizes and the scaling effects on 

the annual recharge simulation and estimation. The conclusions and arguments 

are drawn in section four. 

3.2 Materials and Methods 

3.2.1. Site Description 

In this study, Ottawa County, MI was selected as our study site where 

groundwater faces issues such as water table declination and contamination 

(IWR, 2013:85). The Ottawa County in Michigan heavily relies on its groundwater 

for drinking water, agriculture, and industry. To sustain the increasing demand for 

water accompanying the burgeoning population, pumping wells are being dug 

deeper and larger, which may have induced pollution into the rock formations 

(Sommers, 1984).  And the declining water table may further elevate the 

concentration of contaminants. Existing studies show that the percentage of 

 

90 

domestic wells having chloride concentrations exceeding 250 mg/l 

(recommended level for drinking water) has increased from less than 4% before 

2000 to 6-8% between 2000 and 2010 (IWR, 2013:95). Our fieldwork also 

observed that many small streams dried up in this region in the summertime of 

2015.   

Having a total area of 1460 square kilometers, the Ottawa County, MI is located 

on the west-central edge of the lower peninsula of Michigan along the eastern 

shoreline of Lake Michigan (Figure 3.1). Groundwater is important to the Great 

Lakes ecosystem because it provides a reservoir for storing water and directly 

replenishing the lakes in the form of base flow and indirectly as base flow to the 

tributaries which eventually discharge to the lakes (Sommers, 1984). Serving as 

the most important groundwater resource, the Marshall aquifer is an important 

source in the total water supply (Westjohn & Weaver, 1998). Grand River, as the 

primary river in the county, flows westward into Lake Michigan. The area to the 

north of Grand River and the immediate contiguous region to the south of Grand 

River are drained through the Grand River watershed into Lake Michigan. 

Streams in the southeast region of the county converge to Rabbit River, which 

further drains into Kalamazoo River watershed to the south of Ottawa County, 

MI. And the remaining area on the southwest of the county is drained by 

Macatawa watershed. Ottawa County, MI is located in the Traverse Carbonate 

reservoir which is one of the major Carbonate reservoirs in Michigan producing 

oil and gas (Catacosinos & Daniels, 1991). As significant quantities of oil and gas 

 

91 

produced in this region, the extractions of oil and gas may also introduce 

potential pollution to the groundwater system. 

 

Figure 3.1 Climate and hydrologic stations in the study domain. 
 
 

 

The climate in Ottawa County, MI is strongly influenced by Lake Michigan. During 

winter time, the county is affected by heavy lake-effect snow. Ottawa County, MI 

witnesses an average of 1800 mm of snow each year. Annual snowfall, typically 

in months between October and April, can range from less than 500 mm to 

 

92 

accumulations exceeding 2500 mm per year (miottawa.org). The average annual 

rainfall in the county is around 900 mm (miOttawa.org). The primary rain season 

is well distributed from May to October. Sitting on the eastern shore of Lake 

Michigan, the temperatures in Ottawa Country are cooler during springs and 

early summers and milder in falls and winters than the temperatures in the rest of 

the area of the county. This special climate is essential to the county where 

agricultural land takes about fifty percent of the county's total land mass and 

provides opportunities to the diversified agriculture in the county (USCB, 2012).  

Ottawa County, MI has a great diversity of land covers (Figure 3.2). Agricultural 

land dominates in the central, northwest, and southeast parts of the county. And 

large areas of forests and wetlands are predominant in the western portions. A 

large area of forests can be found along Lake Michigan as well. Scattered 

patches of forest are interspersed with farmland, along with wetlands, streams, 

and roads. Northern and southern Michigan forest types overlap in this transition 

area. Four major urbanized areas, Holland, Zeeland, Grandville (extending from 

Grand Rapids), and Grand Haven, are located throughout the county. 

 

 

 

93 

 

Figure 3.2 Map of land use and land cover in the watershed.  
 

3.2.2. Model Descriptions and Setup 

This study uses the PAWS + CLM model to analyze the grid effects on 

groundwater recharge estimations. PAWS (Shen & Phanikumar, 2010) + CLM 

(Oleson et al., 2013) is an integration of all important processes of the 

hydrological cycle considering the entire hydrology of a watershed system, 

explicitly solving the physically-based governing partial differential equations for 

overland flow, channel flow, wetlands, the subsurface flow, and the interactions 

among these components. The model is capable of constraining the recharge 

 

94 

simulations by using all the recharge-relevant components in the water budget. 

PAWS+CLM is a grid-based model allowing each component of the water 

balance to be produced at very fine spatial scales and addresses problems at the 

local scale. The non-iterative method of coupling used in PAWS makes 

computation efficient and each process with its individual time step also reduces 

the computational efforts compared to using a uniform time step for all the 

hydrologic processes. The non-iterative method and the adoption of a structured 

grid allow the applicability of the fine-grid-size model for large watersheds over a 

long temporal period.  

The coupling between PAWS and CLM is based on the conservation of mass. 

Vertically and horizontally discretized topography, soil, and geology information 

in PAWS are ported to CLM to keep the same definitions of computational units. 

At each time step, climate forcing is passed into CLM to compute energy and 

water fluxes with ET computed based on the resistance concept. Then, the 

resulting ET fluxes, as a function of depths, are passed back to PAWS as a 

source/sink term for the unsaturated zone (Richards equation). Soil temperature, 

ice content, canopy storage, ground precipitation, dew, and snowmelt amounts 

are also passed from CLM to PAWS. The soil hydrology module in CLM is 

substituted by PAWS, which is the key for coupling these two models.  

To analyze the grid effects on recharge simulation, PAWS + CLM was applied to 

estimate the recharge over Ottawa County, MI between 2011 and 2014 over four 

grid sizes: 300 m, 500 m, 700 m, and 1 km. As the grid size decreases, the 

numbers of grid cell increase dramatically. The total areas vary using different 

 

95 

grid sizes (Table 3.1) because a finer grid size allows the study area to capture 

more areas at the boundary than the total area based on a coarser grid size. For 

the PAWS + CLM model, seven years daily weather data (January 2009 – 

November 2015) were obtained from the National Climatic Data Center (NCDC) 

and Michigan Automated Weather Station Network (MAWN). The 30 m resolution 

national elevation dataset (NED) from U.S. Geological Survey (USGS) was 

preprocessed to generate average cell elevation, lowland-storage bottom 

elevation, average slope, and surface runoff routing of the watershed in the 

computational grids. National Hydrography Dataset (NHD) from USGS provided 

the channel flow network and streams properties. Soil data were obtained from 

the Soil Survey Geographic Database (SSURGO) which is distributed by the U.S. 

Department of Agriculture, Natural Resources Conservation Service (USDA-

NRCS). The soil data were processed by Rosetta to provide soil water retention 

and unsaturated conductivity information for the model (Schaap, Leij, & van 

Genuchten, 2001). The land use and land cover (LULC) data used in the model 

is the Integrated Forest Monitoring Assessment and Prescription (IFMAP) 

dataset and is available from the Michigan Department of Natural Resources 

(MDNR). Land use data were reclassified into several model classes, called plant 

functional types (PFTs) (Thornton & Zimmermann, 2007) based on the land use 

classification percentile values reported in previous studies (Niu, Shen, Li, & 

Phanikumar, 2014; Jia, Ni, Kawahara, & Suetsugi, 2001; Lu & Weng, 2006). As 

the model cell size is coarser than the IFMAP spatial resolution (30 m), a model 

cell possesses a mixture of land use/land cover types and three dominant PFTs 

 

96 

were used in a model cell. The coverage percentage of the ten representative 

LULC in the watershed for the four grid sizes are listed in Table 3.1. 

Table 3.1 Ottawa County Watershed characteristics at different grid sizes  
 

Characteristic 
300m 
Watershed Area (*103 km2)  1.900 
37082 
Total number of grid cells 
19.7% 
Broadleaf Forest Land 
4.6% 
Needleleaf Forest Land 
Shrub 
1.6% 
9.7% 
Grass 
19.1% 
Corn 
14% 
Soybean 
Forage Crops 
13.9% 
10.6% 
Impervious Land 
2.2% 
Water 
Wetland 
4.6% 

 

500m 
1.901 
13347 
20.4% 
4.3% 
1.0% 
9.7% 
19% 
14% 
14.4% 
10.1% 
2.8% 
4.3% 

700m 
1.896 
6818 
20.9% 
4.0% 
0.7% 
9.7% 
19.4% 
14.3% 
14.1% 
9.8% 
3.2% 
3.9% 

1km 
1.887 
3351 
21.3% 
3.8% 
0.4% 
9.6% 
20% 
13.6% 
14.1% 
9.5% 
4.0% 
3.7% 

The model calibration based on the Differential Evolution (DE) (Price, 2005) 

algorithm was done using the High Performance Computing Center (HPCC) 

resources at Michigan State University. Parameters estimated during calibration 

are listed in Table 1.2 in Chapter 1 Section 1.3. The model was calibrated 

against streamflow rates (discharge) from three USGS gauging stations at a 300 

m grid size. The simulations were performed from 2009 to 2015 and the period 

from August 01, 2009 to December 31, 2010 was used to calibrate the model. 

The streamflow rates from USGS gauging stations of the period of 2011 - 2015 

were used to validate the model. In addition, we carried out fieldwork between 

June and November of 2015 to collect streamflow data to provide a set of first-

hand validation data for validating the model results. The SonTek RiverSurveyor 

M9 system and the OTT MF Pro current flow meter were used to measure 

streamflow rates of small streams at 48 sampling locations in Ottawa County, MI. 

 

97 

Also, we used the moderate resolution imaging spectroradiometer (MODIS) 

Global ET (MOD16) products to validate the ET simulated by the PAWS model 

spatially and temporally between 2011 and 2014.   

Model calibration was carried out for 300 m grid size and the calibrated set of 

parameters was then transferred to the other three grid sizes, 500m, 700m, and 

1km. Although there may be a bias in running all the models with various grid 

sizes using the same set of calibrated parameters, it was reasonable to use the 

same values of parameters at all the grid sizes because parameters were 

resampled for all the grid sizes and were adjusted using “global multipliers” within 

the framework of a linear transformation. We also calibrated model at 1km grid 

size to examine the potential bias brought by parameters. Results demonstrate 

that we can use the parameter set that calibrated based on 300m grid size 

maintain the modeling consistency. 

3.3 Results 

3.3.1. Grid-size Effects on the Spatial Pattern of Recharge  

The spatial patterns of the average annual ET and average annual recharge over 

four grid sizes are shown in Figure 3.3. The simulated recharge presents obvious 

spatial variability over all the grid sizes. The spatial patterns of the average 

annual recharge were negatively related to the spatial patterns of average annual 

ET for all the grid sizes, whereby high ET values corresponded to low recharge 

values and vice versa. The spatial patterns of recharge also followed the general 

trend that high and low recharge values occur in high and low elevations 

 

98 

respectively (terrain information displayed in Figure 3.1). In Figure 3.3, the 

negative recharge (discharge) occurs close to the locations of stream channels 

and wetlands. With the grid size increasing, the stream channel boundaries 

became unidentifiable because coarser grid sizes averaged the spatial details of 

LULC. The 300 m and 500 m grid sizes offered the most details of spatial 

variability of recharge among all the grid sizes. The fine grid sizes indicated 

strong local spatial variabilities of the recharge in the watershed. Nevertheless, 

the general spatial patterns of recharge over the watershed were similar for all 

four grid sizes, indicating that the overall spatial pattern of recharge was not 

sensitive to the grid sizes at the watershed scale considered in this study.  

(a) 

(b) 

Figure 3.3 The spatial patterns of (a) average annual ET; and (b) average annual 
recharge over period 2011–2014. 
 
 

 

 

99 

3.3.2. Grid-size Effects on the Temporal Pattern of Recharge 

Figure 3.4 and Figure 3.5 display the differences in spatially averaged monthly 

recharge and annual recharge based on the four grid sizes between 2011 and 

2014. The monthly patterns of recharge using different grid sizes had similar 

trends, as the values of recharge increased from fall to spring and decreased 

from spring to fall in a hydrologic year (Figure 3.4). However, the monthly 

recharge values decreased when the grid size changed from coarse to fine. The 

difference between the monthly recharge values with 300 m grid size and the 

monthly recharge values with the other three grid sizes was the greatest, 

whereas the differences based on 500 m, 700 m, and 1 km grid sizes were small, 

especially during low recharge periods. Recharge increased sharply in Spring 

2011 and Spring 2013 due to the high volume of precipitation, especially with the 

coarse grid sizes demonstrating that recharge estimation was more sensitive to 

precipitation using a coarser grid size. And, by comparing the values of annual 

recharge for different years, the differences of the recharge values among the 

four grid sizes in a drought year (2012) were smaller than the differences of the 

recharge values in a more humid year of 2013 (Figure 3.5). The time-series of 

the annual recharge (Figure 3.5) also indicates that recharge has a positive 

relationship with precipitation for all the four grid sizes. The recharge values 

presented in the drought year of 2012, which had an areal precipitation of 

755mm, were less than the recharge values in the other years. In addition, the 

recharge was the largest in 2013 when the areal precipitation was 1023mm, 

which was the highest among the years. The values of the average annual 

 

100 

recharge for all of the four grid sizes are comparable to the annual average 

values reported by USGS (Neff, Piggott, & Sheets, 2005). Specifically, the values 

of recharge at coarse grid size were closer to the values of the recharge reported 

by USGS because the recharge derived by USGS were represented by a coarse 

resolution 8-digit hydrologic unit code.  

0

50

100

150

200

250

 
Figure 3.4 Monthly areal precipitation with 300 m grid size and monthly areal 
recharge over all the grid sizes of the period 2011-2014. (Units: mm/month)  
 

 

 

101 

280

210

140

70

0

g300
g500
g700
g1K

186.9

160.7

144.3

106.5

166.4

149.4

136.1

98.4

235.8

212.1

189.1

218.4

200.8

185.8

138

127.4

2011

2012

2013

2014

Date

 

Figure 3.5 Annual areal recharge over all the grid sizes between 2011 and 2014. 
 
 
Base flow is a moderate proxy of recharge. To better understand the grid size 

effects on recharge, comparisons of discharge among four grid sizes were 

generated at the USGS 04108800 station (Figure 3.6).  The RNASH (square-

root-transformed NASH) metric, focusing on baseflow, was employed to examine 

the baseflow performance over various grid sizes. RNASH is defined below 

(Shen & Phanikumar, 2010): 

(

O
i



S

i

2

)

RNASH

1
 

n

i




1

n

i

1


(



O
i

where,  O i is the th observed value; 

O
i

)

 

 

 

2

              (Eq.1) 

iS  is the th simulated value; and  is the 

total number of events. 

The values of RNASH at USGS station 04108800 are relatively high at 300 m 

grid size (RNASH = 0.66) and at 500 m grid size (RNASH = 0.68), whereas the 

values are low using a 700 m grid size (RNASH = 0.62) and at 1 km grid size 

 

102 

(RNASH = 0.6). The grid size of 500 m appears to be optimum from the point of 

baseflow simulation.  

)
s
/
3

m

(
 

e
g
r
a
h
c
s
D

i

102

101

100

10-1

10-2

10-3
01/01/11

 

07/03/11

01/02/12

USGS04108800  (Macatawa River, Zeeland)

 

01/02/13

07/03/12
  
USGS04108800  (Macatawa River, Zeeland)

Date (mm/dd/yy)

07/05/14

07/04/13

01/03/14

obs
g300
g500
g700
g1K

01/04/15

07/06/15 11/22/15

 

102

101

100

10-1

10-2

)
s
/
3

m

(
 
e
g
r
a
h
c
s
i
D

10-3
07/13/11

 

08/27/11

10/11/11

Date (mm/dd/yy)

11/25/11

01/02/12

 

Figure 3.6 Comparison of daily discharge over all the grid sizes at USGS 
04108800 station.  
 

3.3.3. Grid-size Effects on the Average Annual Water Budget  

The annual averages of the major water budget components over the period 

2011-2014 using the four grid sizes are reported in Table 3.2. Spatially-averaged 

recharge consisted of a smaller percentage of the precipitation when the grid size 

changed from coarse to fine. The other components in the water budget were not 

as sensitive to the grid size as the recharge. The sums of ET, recharge, surface 

runoff and groundwater outflow, and sublimation over various grid sizes were 

 

103 

about 100% ~ 120% of the precipitation which was acceptable because there 

may be some other inputs/outputs or storage mechanisms in the watershed (Niu 

et al., 2014). The differences among the total budgets over the four grid sizes 

very well reflected the differences of the recharge percentage. The sum of the 

water budget at 300 m grid size was the closest to 100%, and the largest sum of 

the water budget occurred at a grid size of 1 km with the value of 115.6%, which 

was also within the acceptable range as noted earlier. 

Table 3.2 Average annual water budgets (%P, percent of annual precipitation) 
 

 
P 
ET 
Dperc 
Qoc 
Qgc 
SUBM 
Total 
P: precipitation, Dperc: recharge, Qoc: surface runoff, Qgc: groundwater outflow, SUBM: 

700m 
100 
60.6 
19.8 
25.2 
3.2 
4.2 
113 

1km 
100 
59.9 
22.1 
24.1 
5.3 
4.2 
115.6 

300m 
100 
61.9 
12.3 
25.8 
1.0 
4.1 
105.1 

500m 
100 
61 
17.9 
24.8 
3.0 
4.2 
110.9 

sublimation. 

 
3.3.4. Grid-size Effects on Recharge over Different LULC  

Three representative LULC types in this study domain: corn, broadleaf and 

impervious land were selected to analyze the influences of LULC on recharge 

estimation over various grid sizes.  

 Boxplots of annual recharge among the four grid sizes over the three 

representative land types were created at a drought year (2012) and a wet year 

(2013) (Figure 3.7 and Figure 3.8). In this study, three dominant PFTs were 

modeled in each cell as noted in section 2.2. In the boxplots, X-axis noted the 

percentage of one dominant LULC type in a cell. For example, in the first figure 

 

104 

showing the land type of corn, 0.1 in the X-axis means that the percentage of 

corn in a cell is below 10%, and 0.2 means the percentage of corn is between 

10% and 20%, and so forth. And, 0.9 includes the cells with the percentage of 

corn between 80%-100%. Boxes of different grid sizes are shown using distinct 

colors. On each box, the central horizontal line indicated the median, and the 

bottom and top of the vertical line of the box were the first and third quartile 

metric, respectively. The outliers were plotted individually using the “+“ symbol. 

Both Figure 3.7 and Figure 3.8 show that the medians of the recharge over all 

the four grid sizes increased when the coverage percentages of corn land in the 

cells increased in both the drought year of 2012 and the wet year of 2013. 

Oppositely, the medians of the recharge over all the four grid sizes decreased 

when the coverage percentages of broadleaf and impervious land types in the 

cells increased in both the drought year of 2012 and the wet year of 2013. For all 

the grid sizes, the values of recharge remained higher in the cells with a higher 

coverage percentage of cropland (corn) than that in the cells with high coverage 

percentage of forest (broadleaf) and impervious land. The values of recharge in 

the wet year of 2013 were greater than those in the drought year of 2012. Thus 

the changes in the medians of recharge over the coverage percentage of all 

these representative three land types in 2013 were more significant than that in 

2012.  

The medians of recharge for cells with over 60% coverage of corn were similar 

among different grid sizes, indicating that the recharge estimations were not 

sensitive to the grid sizes when corn dominated the cells. Showing a similar 

 

105 

pattern, the medians of recharge for cells dominated by impervious land were 

also not sensitive to different grid sizes. When broadleaf forest land was the 

dominant land type in the cells (above 60% coverage percentage), the medians 

of recharge fluctuated over the four different grid sizes. The fluctuations of the 

medians in the wet year of 2013 were more significant than that in the drought 

year of 2012. The differences in the elevation for major broadleaf forest areas 

were relatively large, whereas, the differences of the elevations for corn was 

small. Thus, the spatial variability of terrain may need to be considered when the 

area experiences high volumes of precipitation.  

The number of the outliers decreased as the grid size increased for all the 

selected land use types, which demonstrated that the spatial variation of 

recharge was more pronounced in the finer grid size while a coarser grid size 

averaged out the spatial outliers. The outliers for the cells dominated by the 

broadleaf forests were more than that in other land cover types due to the spatial 

variabilities of the terrain. Broadleaf forests were distributed mainly in the western 

part of the watershed (close to Lake Michigan) with a lower elevation and eastern 

part of the watershed with a higher elevation. Broadleaf trees can also be found 

in the agricultural land or wetland areas as well. 

 

 

106 

Figure 3.7 Comparisons of annual recharge among all the grid sizes in different 
land types in the drought year of 2012 
 

 

 

 

107 

Figure 3.8 Comparisons of annual recharge among all the grid sizes in different 
land types in the wet year of 2013 
 

 

 

 

108 

3.3.5 Scaling Effects on Annual Recharge  

To further understand the scale effects associated with recharge, this study 

adopted an approach used by Mo et al. (2009) to draw four comparisons 

between the “resampled” recharge data and the simulated recharge data. The 

2011-2014 simulated annual recharge values based on 300 m, 500 m and 700 m 

were resampled to 1 km grid size using the nearest neighbor resampling 

scheme. These three resampled 1 km recharge data were further compared with 

the simulated 1 km recharge data, which we assumed represented the true value 

from our physically-based model.  We also resampled the simulated 300 m 

recharge data to a 500 m grid size and tested it against the simulated recharge 

based on a 500 m grid size. The comparison between the recharge resampled to 

1 km grid size from 300 m, 500 m and 700 m and the simulated recharge based 

on 1 km grid size in 2011 to 2014 were illustrated in Figure 3.9(a) (b), and (c). 

The comparison between the simulated recharge with a 500 m grid size and the 

recharge resampled from a 300 m to a 500 m grid size is shown in Figure 3.9(d).  

The correlation coefficient (R2) was used to analyze the scaling effects. The 

value of R2 ranges from 0 to 1, with 1 representing the two data have a very high 

correlation, meaning a weak scaling effect (Mo et al., 2009). The grid size effect 

testing on ET (Mo et al., 2009) had observed R2 values between a weak 

relationship of 0.6 and a strong relationship of 0.98. And, in our comparison, a 

relatively weak relationship (R2 = 0.48 - 0.62) was revealed, which showed that 

the simulated annual recharge was sensitive to grid size changes. Because of 

the existing grid size differences in the resampling process, the R2 value was 

 

109 

higher when the two grid sizes were farther apart. For example, the climate and 

physical characteristics using a 300 m grid size were more fragmented compared 

to ones using a 500 m, 700 m, or 1 km grid size. And interestingly, the R2 

between the resampled recharge from a 300 m to a 500 m grid size and the 

simulated recharge values based on a 500 m grid size (R2 = 0.51 – 0.54) was as 

strong as the R2 between the resampling from a 300 m to a 1 km grid size and 

the simulated recharge values based on a 1 km grid size (R2 = 0.48 – 0.54).  

Therefore, the scaling effect was strong for recharge among the grid sizes we 

tested of 300 m, 500 m, 700 m, and 1 km, which implies that the grid size should 

be very carefully selected for the physically-based hydrologic model based on 

specific purposes of the study.  

 
 
 
 
 
 
 
 
 
 
 
 
 
 

 

110 

(a) 

(b) 

(c) 

(d) 

 
Figure 3.9 Comparisons of annual recharge (mm/year) between (a) the upscaled 
from 700 m to 1 km and 1 km grid size; (b) the upscaled from 500 m to 1 km and 
1 km grid size; (c) the upscaled from 300 m to 1 km and 1 km grid size; and (d) 
the upscaled from 300 m to 500 m and the 500 m grid size for the period 2011–
2014.  
 
3.4 Discussions and Conclusions 

This study used a physically-based hydrologic model (PAWS) to examine the 

grid-size effects on recharge estimation over the grid sizes of 300 m, 500 m, 700 

m, and 1 km. Based on a watershed scale, we examined how different grid sizes 

impact on recharge estimation and representation. Specifically, we focused on 

the grid-size effects on (1) the spatial variation and representation of the 

simulated recharge and watershed structures, such as stream channels and 

 

111 

wetlands; (2) the time-series study of the recharge over humid and drought 

years; (3) the relation between LULC types and recharge estimation; and (4) the 

consistency of the annual recharge estimation across different spatial scales.  

First, the simulated recharge based on all of the four grid sizes showed obvious 

spatial variability. Fine grid sizes (300 m and 500 m) provided greater details of 

spatial variability of recharge and can capture the locations of major stream 

channels and wetlands. Therefore, the grid sizes below 500 m are recommended 

for simulating and presenting the detailed spatial variabilities of recharge at the 

watershed considered in this study. For resource management purposes on a 

county level, a smaller grid size also provides the feasibility to carry out detailed 

management plans. But any of the four grid sizes can be used to illustrate the 

general spatial pattern of recharge on a watershed scale as well.  

Second, the time-series patterns of both monthly and annual areal recharge over 

the four-year period between 2011 and 2014 were similar across all of the four 

grid sizes. But, both the monthly and annual values of areal recharge increased 

as the grid size increased. The fluctuations of annual precipitation in different 

years did not affect the spatial pattern of recharge, but the simulated recharge 

maps revealed a positive relation between the recharge and precipitation. The 

differences in the recharge values using different grid sizes were especially 

strong when the precipitation was high in a particular year. This implies that in a 

humid year or in areas with high precipitation, recharge simulation using a 

physically-based hydrologic model based on a coarser grid size might produce 

different results from one based on a finer grid size.  

 

112 

Further, this study analyzed the grid-size effect on simulating recharge based on 

three representative LULC categories of corn, broadleaf forest, and impervious 

land. The analysis of recharge estimation over different LULC types 

demonstrated that the impacts of different LULC types on recharge showed 

similar effects with all the four grid sizes. The recharge values remained high 

when the cropland (corn) was the dominant LULC component in the cells over all 

the four different grid sizes. The values of recharge were low when forests and 

impervious land types dominated the cells. However, the relation between the 

recharge values and the three representative LULC types was consistent for all 

four grid sizes, which indicated that the grid size effect was minor in illustrating 

the relationship between LULC types and recharge. Therefore, a coarser grid 

size is recommended in investigating the impacts of LULC types for computation 

efficiency purposes.  

Last but not least, we also tested whether the recharge estimations using 

different grid sizes remained consistent based on the correlation coefficient (R2) 

for four pairs of the “resampled” recharge values from a finer grid size and the 

simulated recharge values. The results showed that the resampled recharge from 

a finer grid size and the simulated recharge did not have a strong correlation (R2 

in the 0.48-0.62 range), which indicated that the resampled values of recharge 

might be inconsistent with the simulated recharge values. This situation implies 

that the scaling effect is strong in recharge estimation and the grid size needs to 

be carefully selected for recharge estimation based on the purpose of the study. 

For resource management purposes, the recharge estimation based on a 

 

113 

coarser grid size should be reviewed before detailed management plans can be 

applied to a relatively small watershed.  

Based on this study, we conclude that the grid size plays an important role in 

estimating recharge values. And seeking and justifying a proper and robust grid 

size for a study or research is always challenging. Based on the results, we can 

see that choosing a grid size properly is very important in representing the 

recharge values, investigating times series of recharge estimation, and carrying 

out a physically-based hydrologic model for recharge estimation. And, the 

relationship between recharge and LULC types remains consistent across 

different grid sizes. Further investigations are also needed to understand the 

application of specific resource management plans based on the recharge 

estimation using various grid sizes and how finer grid sizes can assist the 

decision-making process in groundwater and water resource management.  

 

 

 

 

 

114 

APPENDICES 

 

115 

APPENDIX A 

 
Figure A1 Comparisons between observed and simulated discharge at USGS 
04118800. Dark green line depicts the simulation conducted with gauge-MPE 
precipitation; Dark blue dash line describes the observed discharge. 
 

 
Figure A2 Comparisons between observed and simulated discharge at USGS 
04108600. Dark green line depicts the simulation conducted with gauge-MPE 
precipitation; Dark blue dash line describes the observed discharge. 

 

116 

APPENDIX B 

In this study, we calibrated our model based on 300 m and 1 km grid size and 

adopted the parameter set calibrated at 300 m for all grid sizes to maintain the 

modeling consistency based on two main reasons. First, we evaluated the model 

performance based on the discharge comparison on the 1 km grid size with two 

set of parameters calibrated at either the 300 m or the 1 km grid sizes. The 

comparisons with the observed data show great similarities. Second, we 

compared the simulated recharge at 1 km grid size with the parameter set 

calibrated at either the 300 m or the 1 km grid sizes and observed that the results 

illustrate similar temporal patterns. Thus, for modeling efficiency and consistency 

purposes, we adopted the parameter set calibrated at 300 m for all the models at 

four different grid sizes. The detailed comparison and decision-making process is 

illustrated here in Appendix B.  

Figure B1 shows the comparisons of daily discharge over four grid sizes with 

parameters calibrated at the 300 m grid size and daily discharge at the 1 km grid 

size with parameters calibrated at 1 km at USGS 04108800. The value of 

RNASH between estimated discharge and observed discharge at the 1 km grid 

size with parameters calibrated at the 300 m grid size is 0.6, while the value of 

RNASH between the two at the 1 km grid size with parameters calibrated at 1 km 

is 0.54. The RNASH comparison shows that the model performs better when we 

use the 300 m based parameter set than the one calibrated on the 1 km grid 

size. And, we conclude that the parameter set calibrated at the 300 m grid size 

can be used for the model with the 1 km grid size.  

 

117 

Figure B2 shows that the temporal patterns of the monthly areal simulated 

recharge based on the parameter sets calibrated at 1 km and at 300 m are 

similar. Figure B3 shows that both the simulated recharge at the 1 km grid size 

with parameters calibrated at the 300 m grid size and the 1 km grid size also 

show similar patterns and strong agreements when compared with recharge 

based on other grid sizes. So, we think the parameter set calibrated at 300 m can 

be used for all the grid sizes in this study for the modeling efficiency and 

consistency purposes.   

 

 

118 

USGS04108800  (Macatawa River, Zeeland)

 

10-3
01/01/11

 

07/03/11

01/02/12

07/03/12

01/02/13

07/04/13

Date (mm/dd/yy)

  
USGS04108800  (Macatawa River, Zeeland)

102

101

100

10-1

10-2

)
s

/
3

m

(
 

e
g
r
a
h
c
s
D

i

)
s

/
3

m

(
 

e
g
r
a
h
c
s
D

i

102

101

100

10-1

10-2

10-3
07/13/11

 

obs
g300
g500
g700
g1K
g1K_1K

01/03/14

07/05/14

01/04/15

07/06/15 11/22/15

 

08/27/11

10/11/11

Date (mm/dd/yy)

11/25/11

01/02/12

 

Figure B1 Comparison between daily discharge over four grid sizes with 
parameters calibrated at 300 m grid size and daily discharge at 1 km grid size 
with parameters calibrated at 1 km from 2011 to 2014. (Units: m3/s) (g300, g500, 
g700 and g1K indicate the parameters estimated at 300 m grid size, g1K_1K 
indicates the parameters estimated at 1 km grid size) 
 
 

 

 

119 

Figure B2 Comparison between monthly areal recharge over four grid sizes with 
parameters calibrated at 300 m grid size and monthly areal recharge at 1 km grid 
size with parameters calibrated at 1 km from 2011 to 2014.(Units: mm/month) 
(g300, g500, g700 and g1K indicate the parameters estimated at 300 m grid size, 
g1K_1K indicates the parameters estimated at 1 km grid size) 
 

 

Figure B3 Comparison between annual areal recharge over four grid sizes with 
parameters calibrated at 300 m grid size and annual areal recharge at 1 km grid 
size with parameters calibrated at 1 km over the period 2011-2014. (Units: 
mm/year) (g300, g500, g700 and g1K indicate the parameters estimated at 300 
m grid size, g1K_1K indicates the parameters estimated at 1 km grid size) 

 

 

120 

BIBLIOGRAPHY 

 

121 

 

 

BIBLIOGRAPHY 

Aeschbach-Hertig, Werner, and Tom Gleeson. 2012. Regional Strategies for the 

Accelerating Global Problem of Groundwater Depletion. Nature 
Geoscience 5 (December): 853–61.  

 
Alley, William M.. 2006. Tracking U.S. Groundwater: Reserves for the Future? 

Environment: Science and Policy for Sustainable Development 48 (3): 10–25.  

 
Apple, Beth A, and Howard W Reeves. 2007. Summary of Hydrogeologic Conditions by 

County for the State of Michigan. U.S. Geology Survey, 87. 

 
Arnold, J. G, R. S Muttiah, R Srinivasan, and P. M Allen. 2000. Regional Estimation of 

Base Flow and Groundwater Recharge in the Upper Mississippi River Basin. 
Journal of Hydrology 227 (1–4): 21–40.  

 
Arnold, J. G., R. Srinivasan, R. S. Muttiah, and J. R. Williams. 1998. Large Area 

Hydrologic Modeling and Assessment Part I: Model Development1. JAWRA 
Journal of the American Water Resources Association 34 (1): 73–89.  

 
Ashby, Steven F., and Robert D. Falgout. 1996. A Parallel Multigrid Preconditioned 

Conjugate Gradient Algorithm for Groundwater Flow Simulations. Nuclear 
Science and Engineering. Vol. 124, 1996, Issue 1.  

 
Awan, Usman Khalid, and Ali Ismaeel, 2014. A New Technique to Map Groundwater 
Recharge in Irrigated Areas Using a SWAT Model under Changing Climate. 
Journal of Hydrology 519, Part B (November): 1368–82.  

 
Bailey, Ryan, Hendrik Rathjens, Katrin Bieger, Indrajeet Chaubey, and Jeffrey Arnold. 
2017. SWATMOD-Prep: Graphical User Interface for Preparing Coupled SWAT-
MODFLOW Simulations. JAWRA Journal of the American Water Resources 
Association, February.  

 
Bailey, Ryan T., Tyler C. Wible, Mazdak Arabi, Rosemary M. Records, and Jeffrey Ditty. 

2016. Assessing Regional-Scale Spatio-Temporal Patterns of Groundwater–
Surface Water Interactions Using a Coupled SWAT-MODFLOW Model. 
Hydrological Processes 30 (23): 4420–33.  

 
Bell, V.A. and Moore, R.J. 2000. The Sensitivity of Catchment Runoff Models to Rainfall 

Data at Different Spatial Scales. Hydrology and Earth System Sciences 4 
(December).  

 
Biftu, G. F, and T. Y Gan. 2001. Semi-Distributed, Physically Based, Hydrologic 

Modeling of the Paddle River Basin, Alberta, Using Remotely Sensed Data. 
Journal of Hydrology 244 (3): 137–56.  

 

 

122 

Bormann, H. 2006. Impact of Spatial Data Resolution on Simulated Catchment Water 
Balances and Model Performance of the Multi-Scale TOPLATS Model. Hydrol. 
Earth Syst. Sci. 10 (2): 165–79.  

 
Bresloff, Cynthia J., Uyen Nguyen, Edward P. Glenn, Jody Waugh, and Pamela L. 
Nagler. 2013. Effects of Grazing on Leaf Area Index, Fractional Cover and 
Evapotranspiration by a Desert Phreatophyte Community at a Former Uranium 
Mill Site on the Colorado Plateau. Journal of Environmental Management 114 
(January): 92–104.  

 
Brown, Jenny, Leslie Bach, Allison Aldous, Abby Wyers, and Julia DeGagné. 2011. 

Groundwater-Dependent Ecosystems in Oregon: An Assessment of Their 
Distribution and Associated Threats. Frontiers in Ecology and the Environment 9 
(2): 97–102.  

 
Camporese, M., C. Paniconi, M. Putti, and S. Orlandini. 2010. Surface-Subsurface Flow 
Modeling with Path-Based Runoff Routing, Boundary Condition-Based Coupling, 
and Assimilation of Multisource Observation Data. Water Resources Research 
46 (2): W02512.  

 
Cao, Guoliang, Bridget R. Scanlon, Dongmei Han, and Chunmiao Zheng. 2016. Impacts 

of Thickening Unsaturated Zone on Groundwater Recharge in the North China 
Plain. Journal of Hydrology 537 (June): 260–70.  

 
Catacosinos, Paul A., and Paul A. Daniels. 1991. Early Sedimentary Evolution of the 

Michigan Basin. Geological Society of America. Vol. 256.  

 
Cerdan, O., Y. Le Bissonnais, G. Govers, V. Lecomte, K. van Oost, A. Couturier, C. 

King, and N. Dubreuil. 2004. Scale Effect on Runoff from Experimental Plots to 
Catchments in Agricultural Areas in Normandy. Journal of Hydrology 299 (1): 4–
14.  

Chen, Sheng, Jonathan J. Gourley, Yang Hong, P. E. Kirstetter, Jian Zhang, Kenneth 
Howard, Zachary L. Flamig, Junjun Hu, and Youcun Qi. 2013. Evaluation and 
Uncertainty Estimation of NOAA/NSSL Next-Generation National Mosaic 
Quantitative Precipitation Estimation Product (Q2) over the Continental United 
States. Journal of Hydrometeorology 14 (4): 1308–22.  

 
Cherkauer, Douglas S., and Sajjad A. Ansari. 2005. Estimating Ground Water Recharge 
from Topography, Hydrogeology, and Land Cover. Ground Water 43 (1): 102–12.  

 
Chinnasamy, Pennan, Gourav Misra, Tushaar Shah, Basant Maheshwari, and 

Sanmugam Prathapar. 2015. Evaluating the Effectiveness of Water 
Infrastructures for Increasing Groundwater Recharge and Agricultural Production 
– A Case Study of Gujarat, India. Agricultural Water Management 158 (August): 
179–88.  

 
Coelho, Victor Hugo R., Suzana Montenegro, Cristiano N. Almeida, Bernardo B. Silva, 

Leidjane M. Oliveira, Ana Cláudia V. Gusmão, Emerson S. Freitas, and Abelardo 
A. A. Montenegro. 2017. Alluvial Groundwater Recharge Estimation in Semi-Arid 

 

123 

Environment Using Remotely Sensed Data. Journal of Hydrology 548 (May): 1–
15.  

 
Cole, Steven J., and Robert J. Moore. 2008. Hydrological Modelling Using Raingauge- 

and Radar-Based Estimators of Areal Rainfall. Journal of Hydrology 358 (3): 
159–81.  

 
Comte, Jean-Christophe, Rachel Cassidy, Joy Obando, Nicholas Robins, Kassim 

Ibrahim, Simon Melchioly, Ibrahimu Mjemah, et al. 2016. Challenges in 
Groundwater Resource Management in Coastal Aquifers of East Africa: 
Investigations and Lessons Learnt in the Comoros Islands, Kenya and Tanzania. 
Journal of Hydrology: Regional Studies 5 (March): 179–99.  

 
Crosbie, Russell S., Bridget R. Scanlon, Freddie S. Mpelasoka, Robert C. Reedy, John 

B. Gates, and Lu Zhang. 2013. Potential Climate Change Effects on 
Groundwater Recharge in the High Plains Aquifer, USA. Water Resources 
Research 49 (7): 3936–51.  

 
Cunha, Luciana K., Pradeep V. Mandapaka, Witold F. Krajewski, Ricardo Mantilla, and 

Allen A. Bradley. 2012. Impact of Radar-Rainfall Error Structure on Estimated 
Flood Magnitude across Scales: An Investigation Based on a Parsimonious 
Distributed Hydrological Model.” Water Resources Research 48 (10): W10515.  

 
Dagès, Cécile, Claudio Paniconi, and Mauro Sulis. 2012. Analysis of Coupling Errors in 
a Physically-Based Integrated Surface Water–Groundwater Model. Advances in 
Water Resources 49 (December): 86–96.  

 
Dai, Yongjiu, Robert E. Dickinson, and Ying-Ping Wang. 2004. A Two-Big-Leaf Model for 

Canopy Temperature, Photosynthesis, and Stomatal Conductance. Journal of 
Climate 17 (12): 2281–99.  

 
Dakhlalla, Abdullah O., Prem B. Parajuli, Ying Ouyang, and Darrel W. Schmitz. 2016. 

Evaluating the Impacts of Crop Rotations on Groundwater Storage and Recharge 
in an Agricultural Watershed. Agricultural Water Management 163 (January): 
332–43.  

 
Daly, Christopher, Wayne P. Gibson, George H. Taylor, Gregory L. Johnson, and Phillip 

Pasteris. 2002. A Knowledge-Based Approach to the Statistical Mapping of 
Climate. Climate Research 22 (2): 99–113.  

 
Doviak, R.J., and Zrnić, D.S. 1984. Doppler Radar and Weather Observations. 

Academic Press.  

 
Doviak, R.J. and Zrnic, D.S. 1995. New Uncertainty Concepts in Hydrology and Water 

Resources. Cambridge University Press.  

 
Ebel, B.A., Mirus, B.B., Christopher S. Heppner, Joel E. VanderKwaak, and Keith 

Loague. 2009. First-Order Exchange Coefficient Coupling for Simulating Surface 
Water–Groundwater Interactions: Parameter Sensitivity and Consistency with a 
Physics-Based Approach. Hydrological Processes 23 (13): 1949–59.  

 

124 

Fu, Suhua, Torben O. Sonnenborg, Karsten H. Jensen, and Xin He. 2011. Impact of 

Precipitation Spatial Resolution on the Hydrological Response of an Integrated 
Distributed Water Resources Model. Vadose Zone Journal 10 (1): 25–36.  

 
Fulton, Richard A., Jay P. Breidenbach, Dong-Jun Seo, Dennis A. Miller, and Timothy 

O’Bannon. 1998. The WSR-88D Rainfall Algorithm. Weather and Forecasting 13 
(2): 377–95.  

 
Furman, Alex. 2008. Modeling Coupled Surface–Subsurface Flow Processes: A Review. 

Vadose Zone Journal 7 (2): 741–56.  

 
Gao, Jungang, Aleksey Y. Sheshukov, Haw Yen, and Michael J. White. 2017. Impacts of 

Alternative Climate Information on Hydrologic Processes with SWAT: A 
Comparison of NCDC, PRISM and NEXRAD Datasets. CATENA 156 
(Supplement C): 353–64.  

 
Gebert, Warren A., Mandy J. Radloff, Ellen J. Considine, and James L. Kennedy. 2007. 

Use of Streamflow Data to Estimate Base Flow/Ground-Water Recharge For 
Wisconsin1. JAWRA Journal of the American Water Resources Association 43 
(1): 220–36.  

 
Giordano, Mark. 2009. Global Groundwater? Issues and Solutions. Annual Review of 

Environment and Resources 34 (1): 153–78.  

 
Gleeson, Tom, Yoshihide Wada, Marc F. P. Bierkens, and Ludovicus P. H. van Beek. 
2012. Water Balance of Global Aquifers Revealed by Groundwater Footprint. 
Nature 488 (7410): 197–200. https://doi.org/10.1038/nature11295. 

 
Glenn, Edward P., Christopher J. Jarchow, and W. Joseph Waugh. 2016. 

Evapotranspiration Dynamics and Effects on Groundwater Recharge and 
Discharge at an Arid Waste Disposal Site. Journal of Arid Environments 133 
(October): 1–9.  

 
Gorelick, Steven M., and Chunmiao Zheng. 2015. Global Change and the Groundwater 

Management Challenge. Water Resources Research 51 (5): 3031–51.  

 
Grimes, D. I. F, E Pardo-Igúzquiza, and R Bonifacio. 1999. Optimal Areal Rainfall 

Estimation Using Raingauges and Satellite Data. Journal of Hydrology 222 (1): 
93–108.  

 
Allan Freeze and John Cherry,1979. Groundwater. Prentice-Hall, the University of 

Michigan.  

 
Habib, Emad, Boone F. Larson, and Jeffrey Graschel. 2009. Validation of NEXRAD 

Multisensor Precipitation Estimates Using an Experimental Dense Rain Gauge 
Network in South Louisiana. Journal of Hydrology 373 (3–4): 463–78.  

 
Hanson, R.L. 1991. Evapotranspiration and Droughts. Hydrologic Events and Floods 
and Droughts: U.S. Geological Survey Water-Supply Paper 2375, p. 99-104. 

 

125 

Hartmann, Andreas, Tom Gleeson, Yoshihide Wada, and Thorsten Wagener. 2017. 
Enhanced Groundwater Recharge Rates and Altered Recharge Sensitivity to 
Climate Variability through Subsurface Heterogeneity. Proceedings of the 
National Academy of Sciences 114 (11): 2842–47.  

 
Healy, Richard W., and Bridget R. Scanlon. 2010. Estimating Groundwater Recharge. 

Cambridge University Press.  

 
Hessel, Rudi. 2005. Effects of Grid Cell Size and Time Step Length on Simulation 

Results of the Limburg Soil Erosion Model (LISEM). Hydrological Processes 19 
(15): 3037–49.  

 
Howard, Jeanette, and Matt Merrifield. 2010. Mapping Groundwater Dependent 

Ecosystems in California. PLOS ONE 5 (6): e11249.  

 
Hunt, Randall J., David E. Prudic, John F. Walker, and Mary P. Anderson. 2008. 

Importance of Unsaturated Zone Flow for Simulating Recharge in a Humid 
Climate. Groundwater 46 (4): 551–60.  

 
Isaaks, Edward H., and R. Mohan Srivastava. 1990. An Introduction to Applied 

Geostatistics Edward H. Isaaks; R. Mohan Srivastava, Oxford University Press.  

 
Jiang, Haiyan, Peter G. Black, Edward J. Zipser, Frank D. Marks, and Eric W. Uhlhorn. 

2006. Validation of Rain-Rate Estimation in Hurricanes from the Stepped 
Frequency Microwave Radiometer: Algorithm Correction and Error Analysis. 
Journal of the Atmospheric Sciences 63 (1): 252–67.  

 
Johnson Dennis, Smith Michael, Koren Victor, and Finnerty Bryce. 1999. Comparing 
Mean Areal Precipitation Estimates from NEXRAD and Rain Gauge Networks. 
Journal of Hydrologic Engineering 4 (2): 117–24.  

 
Jyrkama, Mikko I., Jon F. Sykes, and Stefano D. Normani. 2002. Recharge Estimation 

for Transient Ground Water Modeling. Ground Water 40 (6): 638–48.  

 
Kalin Latif, and Hantush Mohamed M. 2006. Hydrologic Modeling of an Eastern 
Pennsylvania Watershed with NEXRAD and Rain Gauge Data. Journal of 
Hydrologic Engineering 11 (6): 555–69.  

 
Kampf, Stephanie K., and Stephen J. Burges. 2007. A Framework for Classifying and 

Comparing Distributed Hillslope and Catchment Hydrologic Models. Water 
Resources Research 43 (5): W05423.  

 
Kang, Kwangmin, and Venkatesh Merwade. 2014. The Effect of Spatially Uniform and 

Non-Uniform Precipitation Bias Correction Methods on Improving NEXRAD 
Rainfall Accuracy for Distributed Hydrologic Modeling. Hydrology Research 45 
(1): 23–42.  

 
Kendy, Eloise. 2005. The False Promise of Sustainable Pumping Rates. Groundwater 

41 (1): 2–4.  

 

 

126 

Kendy, Eloise, Pierre Gérard‐Marchant, M. Todd Walter, Yongqiang Zhang, Changming 

Liu, and Tammo S. Steenhuis. 2003. A Soil-Water-Balance Approach to Quantify 
Groundwater Recharge from Irrigated Cropland in the North China Plain. 
Hydrological Processes 17 (10): 2011–31.  

 
Kim, H. John, and Jackson, B. Robert. 2012. A Global Analysis of Groundwater 

Recharge for Vegetation, Climate, and Soils. Vadose Zone Journal  
 

Kollet, Stefan J., and Reed M. Maxwell. 2006. Integrated Surface–Groundwater Flow 

Modeling: A Free-Surface Overland Flow Boundary Condition in a Parallel 
Groundwater Flow Model. Advances in Water Resources 29 (7): 945–58.  

 
Kollet, Stefan J., and Reed M. Maxwell. 2008. Capturing the Influence of Groundwater 

Dynamics on Land Surface Processes Using an Integrated, Distributed 
Watershed Model. Water Resources Research 44 (2): W02402.  

 
Komatsu, Hikaru, Tomonori Kume, and Kyoichi Otsuki. 2010. Increasing Annual 

Runoff—Broadleaf or Coniferous Forests? Hydrological Processes 25 (2): 302–
18.  

 
KUNDZEWICZ, ZBIGNIEW W., and PETRA DÖLL. 2009.“Will Groundwater Ease 

Freshwater Stress under Climate Change.” Hydrological Sciences Journal 54 (4): 
665–75.  

 
Kuo, Wen-Ling, Tammo S. Steenhuis, Charles E. McCulloch, Charles L. Mohler, David 
A. Weinstein, Stephen D. DeGloria, and Dennis P. Swaney. 1999. Effect of Grid 
Size on Runoff and Soil Moisture for a Variable-Source-Area Hydrology Model. 
Water Resources Research 35 (11): 3419–28.  

 
Larson, Lee W., and Eugene L. Peck. 1974. Accuracy of Precipitation Measurements for 

Hydrologic Modeling. Water Resources Research 10 (4): 857–63.  

 
Lawrence, David M, Keith W Oleson, Mark G Flanner, Peter E Thornton, Sean C 

Swenson, Peter J Lawrence, Xubin Zeng, et al. 2011. Parameterization 
Improvements and Functional and Structural Advances in Version 4 of the 
Community Land Model. Journal of Advances in Modeling Earth Systems 3 (1): 
M03001.  

 
Lee, Cheng-Haw, Wei-Ping Chen, and Ru-Huang Lee. 2007. Estimation of Groundwater 
Recharge Using Water Balance Coupled with Base-Flow-Record Estimation and 
Stable-Base-Flow Analysis. Environmental Geology 51 (5): 869–869.  

 
Liang, Xu, Jianzhong Guo, and L. Ruby Leung. 2004. Assessment of the Effects of 

Spatial Resolutions on Daily Water Flux Simulations. Journal of Hydrology, 298 
(1): 287–310.  

 
Linsley, Ray K. 1992. Water Resources Engineering 4th Edition. New York: McGraw-Hill 

Publishing Co. 

 
 

 

 

127 

McMahon, P. B., and J. K. Böhlke. 2006. Regional Patterns in the Isotopic Composition 

of Natural and Anthropogenic Nitrate in Groundwater, High Plains, U.S.A. 
Environmental Science & Technology 40 (9): 2965–70.  

 
MDNR (Michigan Department of Natural Resources), 

http://www.mcgi.state.mi.us/mgdl/?rel=thext&action=thmname&cid=5&cat=Land+
Cover+2001 

 
Memon, B. A. 1995. Quantitative Analysis of Springs. Environmental Geology 26 (2): 

111–20.  

 
Menne, Matthew J., Imke Durre, Russell S. Vose, Byron E. Gleason, and Tamara G. 

Houston. 2012. An Overview of the Global Historical Climatology Network-Daily 
Database. Journal of Atmospheric and Oceanic Technology 29 (7): 897–910.  

 
Mo, X.G., Liu, S.X., Chen, D., Lin, Z.H., Guo, R.P., and Wang, K.. 2009. Grid-Size 

Effects on Estimation of Evapotranspiration and Gross Primary Production over a 
Large Loess Plateau Basin, China. Hydrological Sciences Journal 54 (1): 160–
73.  

 
Mobley John T., Culver Teresa B., and Burgholzer Robert W. 2012. Understanding 

Precipitation Fidelity in Hydrological Modeling. Journal of Hydrologic Engineering 
17 (12): 1315–24.  

 
Molnár, D.K. and Julien, P.Y. 2000. Grid-Size Effects on Surface Runoff Modeling. 

Journal of Hydrologic Engineering 5 (1): 8–16.  

 
Moore, R. J., Bell, V.A., Cole, S.J., and Jones, D.A. 2006. Spatio-Temporal Rainfall 

Datasets and Their Use in Evaluating the Extreme Event Performance of 
Hydrological Models. Publication - Report. August 2006.  

 
Moriasi, D. N., Arnold, J. G., Van Liew, M. W. and Bingner, R.L., Harmel, R. D. and 

Veith, T. L. 2007. Model Evaluation Guidelines for Systematic Quantification of 
Accuracy in Watershed Simulations. Transactions of the ASABE, no. 50: 885–
900.  

 
Mu, Qiaozhen, Faith Ann Heinsch, Maosheng Zhao, and Steven W. Running. 2007. 
Development of a Global Evapotranspiration Algorithm Based on MODIS and 
Global Meteorology Data. Remote Sensing of Environment 111 (4): 519–36.  

 
Mu, Qiaozhen, Maosheng Zhao, and Steven Running. 2013. MODIS Global Terrestrial 

Evapotranspiration (ET) Product (NASA MOD16A2/A3) Collection 5. NASA 
Headquarters. Numerical Terradynamic Simulation Group Publications, 
November.  

 
Mu, Qiaozhen, Maosheng Zhao, and Steven W. Running. 2011. Improvements to a 
MODIS Global Terrestrial Evapotranspiration Algorithm. Remote Sensing of 
Environment 115 (8): 1781–1800.  

 
 

 

 

128 

Murray, By Brad R., Melanie J. B. Zeppel, Grant C. Hose, and Derek Eamus. 2003. 

Groundwater-Dependent Ecosystems in Australia: It’s More than Just Water for 
Rivers. Ecological Management & Restoration 4 (2): 110–13.  

 
Neary V. S., Habib E., and Fleming M. 2004. Hydrologic Modeling with NEXRAD 

Precipitation in Middle Tennessee. Journal of Hydrologic Engineering 9 (5): 339–
49.  

 
Neff, B.P., A.R. Piggott, and R.A. Sheets. 2005. Estimation of Shallow Ground-Water 

Recharge in the Great Lakes Basin. 2005.  

 
Niazi, Amir, Laurence R. Bentley, and Masaki Hayashi. 2017. Estimation of Spatial 

Distribution of Groundwater Recharge from Stream Baseflow and Groundwater 
Chloride. Journal of Hydrology 546 (March): 380–92.  

 
Niu, Jie, Chaopeng Shen, Shu-Guang Li, and Mantha S. Phanikumar. 2014. Quantifying 

Storage Changes in Regional Great Lakes Watersheds Using a Coupled 
Subsurface-Land Surface Process Model and GRACE, MODIS Products. Water 
Resources Research 50 (9): 7359–77.  

 
Ochoa-Rodriguez, Susana, Li-Pen Wang, Auguste Gires, Rui Daniel Pina, Ricardo 
Reinoso-Rondinel, Guendalina Bruni, Abdellah Ichiba, et al. 2015. Impact of 
Spatial and Temporal Resolution of Rainfall Inputs on Urban Hydrodynamic 
Modelling Outputs: A Multi-Catchment Investigation. Journal of Hydrology, 
Hydrologic Applications of Weather Radar, 531 (Part 2): 389–407.  

 
P. Krause, D.P. Boyle, and F. Base. 2005. Comparison of Different Efficiency Criteria for 

Hydrological Model Assessment. Advances in Geosciences 5: 89–97. 

 
Panday, Sorab, and Peter S. Huyakorn. 2004. A Fully Coupled Physically-Based 

Spatially-Distributed Model for Evaluating Surface/Subsurface Flow. Advances in 
Water Resources, 27 (4): 361–82.  

 
Paniconi, Claudio, Marino Marrocu, Mario Putti, and Mark Verbunt. 2003. Newtonian 

Nudging for a Richards Equation-Based Distributed Hydrological Model. 
Advances in Water Resources 26 (2): 161–78.  

 
Paniconi, Claudio, and Eric F. Wood. 1993. A Detailed Model for Simulation of 

Catchment Scale Subsurface Hydrologic Processes. Water Resources Research 
29 (6): 1601–20.  

 
Pechlivanidis, I. G., N. McIntyre, and H. S. Wheater. 2017. The Significance of Spatial 
Variability of Rainfall on Simulated Runoff: An Evaluation Based on the Upper 
Lee Catchment, UK. Hydrology Research 48 (4): 1118–30.  

 
Price, K., S. T. Purucker, and S. R. Kraemer. 2011. Multi-Scale Comparison of Stage IV 

Nexrad (MPE) and Gauge Precipitation Data for Watershed Modeling. Georgia 
Institute of Technology.  

 
 

 

 

129 

Reed Seann M., and Maidment David R. 1999. Coordinate Transformations for Using 

NEXRAD Data in GIS-Based Hydrologic Modeling. Journal of Hydrologic 
Engineering 4 (2): 174–82.  

 
Richey, Alexandra S., Brian F. Thomas, Min-Hui Lo, John T. Reager, James S. 

Famiglietti, Katalyn Voss, Sean Swenson, and Matthew Rodell. 2015. 
Quantifying Renewable Groundwater Stress with GRACE. Water Resources 
Research 51 (7): 5217–38.  

 
Riley, W. J., and C. Shen. 2014. Characterizing Coarse-Resolution Watershed Soil 

Moisture Heterogeneity Using Fine-Scale Simulations. Hydrol. Earth Syst. Sci. 18 
(7): 2463–83.  

 
Risser, Dennis W, William J Gburek, and Gordon J Folmar. 2005. Comparison of 

Methods for Estimating Ground-Water Recharge and Base Flow at a 
Small Watershed Underlain by Fractured Bedrock in the Eastern United 
States. Scientific Investigations Report 2005-5038.  

 
Rodell, Matthew, Isabella Velicogna, and James S. Famiglietti. 2009. Satellite-Based 
Estimates of Groundwater Depletion in India. Nature 460 (7258): 999–1002.  

 
Rosalia Rojas, Mark Velleux, Pierre Y. Julien, and Billy E. Johnson. 2008. Grid Scale 

Effects on Watershed Soil Erosion Models. Journal of Hydrologic Engineering 13 
(9): 793–802.  

 
Scanlon, Bridget R., and Peter G. Cook. 2002. Theme Issue on Groundwater Recharge. 

Hydrogeology Journal 10 (1): 3–4.  

 
Scanlon, Bridget R., Kelley E. Keese, Alan L. Flint, Lorraine E. Flint, Cheikh B. Gaye, W. 

Michael Edmunds, and Ian Simmers. 2006. Global Synthesis of Groundwater 
Recharge in Semiarid and Arid Regions. Hydrological Processes 20 (15): 3335–
70.  

 
Schaap, Marcel G., Feike J. Leij, and Martinus Th. van Genuchten. 2001. Rosetta: A 
Computer Program for Estimating Soil Hydraulic Parameters with Hierarchical 
Pedotransfer Functions. Journal of Hydrology 251 (3): 163–76.  

 
Schoorl, J.M., M.P.W. Sonneveld, and A Veldkamp. 2000. Three-Dimensional 

Landscape Process Modelling: The Effect of DEM Resolution. Earth Surface 
Processes and Landforms 25 (August): 1025–34.  

 
Segond, Marie-Laure, Howard S. Wheater, and Christian Onof. 2007. The Significance 

of Spatial Rainfall Representation for Flood Runoff Estimation: A Numerical 
Evaluation Based on the Lee Catchment, UK. Journal of Hydrology 347 (1): 116–
31.  

 
Serrat-Capdevila, Aleix, Juan B. Valdes, and Eugene Z. Stakhiv. 2014. Water 

Management Applications for Satellite Precipitation Products: Synthesis and 
Recommendations. Journal of the American Water Resources Association 50 
(2): 509–25.  

 

130 

Sexton-Sims, Aisha, Ali Sadeghi, Xuesong Zhang, and Adel Shirmohammadi. 2010. 

Using NEXRAD and Rain Gauge Precipitation Data for Hydrologic Calibration of 
SWAT in a Northeastern Watershed. Transactions of the ASABE. 53(5): 1501-
1510.  

 
Shen, Chaopeng, Jie Niu, and Mantha S. Phanikumar. 2013. Evaluating Controls on 
Coupled Hydrologic and Vegetation Dynamics in a Humid Continental Climate 
Watershed Using a Subsurface-Land Surface Processes Model. Water 
Resources Research 49 (5): 2552–72.  

 
Shen, Chaopeng, and Mantha S. Phanikumar. 2010. A Process-Based, Distributed 
Hydrologic Model Based on a Large-Scale Method for Surface–Subsurface 
Coupling. Advances in Water Resources 33 (12): 1524–41.  

 
Shen, Z., H. Xie, L. Chen, J. Qiu, and Y. Zhong. 2015. Uncertainty Analysis for Nonpoint 

Source Pollution Modeling: Implications for Watershed Models. International 
Journal of Environmental Science and Technology 12 (2): 739–46.  

 
Siebert, S., J. Burke, J. M. Faures, K. Frenken, J. Hoogeveen, P. Döll, and F. T. 

Portmann. 2010. Groundwater Use for Irrigation – a Global Inventory. Hydrol. 
Earth Syst. Sci. 14 (10): 1863–80.  

 
Simmers, I. 2013. Estimation of Natural Groundwater Recharge. Springer Science & 

Business Media. 

 
Skinner Courtney, Bloetscher Frederick, and Pathak Chandra S. 2009. Comparison of 

NEXRAD and Rain Gauge Precipitation Measurements in South Florida. Journal 
of Hydrologic Engineering 14 (3): 248–60.  

 
Sommers, Lawrence M. 1984. Michigan, a Geography. Geographies of the United 

States. Boulder, Colo: Westview Press. 

 
Sulis, Mauro, Claudio Paniconi, and Matteo Camporese. 2011. Impact of Grid Resolution 

on the Integrated and Distributed Response of a Coupled Surface–Subsurface 
Hydrological Model for the Des Anglais Catchment, Quebec. Hydrological 
Processes 25 (12): 1853–65.  

 
Tan, Xiu-Cui, Jing-Wei Wu, Shu-Ying Cai, and Jin-Zhong Yang. 2013. Characteristics of 
Groundwater Recharge on the North China Plain. Groundwater 52 (5): 798–807.  

 
Thornton, Peter E., and Niklaus E. Zimmermann. 2007. An Improved Canopy Integration 
Scheme for a Land Surface Model with Prognostic Canopy Structure. Journal of 
Climate 20 (15): 3902–23.  

 
Tobler, W. R. 1970. A Computer Movie Simulating Urban Growth in the Detroit Region. 

Economic Geography 46 (sup1): 234–40.  

 
Turkeltaub, T., O. Dahan, and D. Kurtzman. 2014. Investigation of Groundwater 

Recharge under Agricultural Fields Using Transient Deep Vadose Zone Data. 
Vadose Zone Journal 13 (4).  

 

 

131 

VanderKwaak, Joel E., and Keith Loague. 2001. Hydrologic-Response Simulations for 

the R-5 Catchment with a Comprehensive Physics-Based Model. Water 
Resources Research 37 (4): 999–1013.  

 
Vergara, Humberto, Yang Hong, Jonathan J. Gourley, Emmanouil N. Anagnostou, 

Viviana Maggioni, Dimitrios Stampoulis, and Pierre-Emmanuel Kirstetter. 2013. 
Effects of Resolution of Satellite-Based Rainfall Estimates on Hydrologic 
Modeling Skill at Different Scales. Journal of Hydrometeorology 15 (2): 593–613.  

 
Vieux, Baxter E. 2016. Distributed Hydrologic Modeling Using GIS. Springer. 
 
Vries, Jacobus J. de, and Ian Simmers. 2002. Groundwater Recharge: An Overview of 

Processes and Challenges. Hydrogeology Journal 10 (1): 5–17.  

 
IWR and Department of Civil & Environmental Engineering, 2013. Ottawa County 

Water Resources Study Finial Report, Michigan State University. 

 
Wellogic, 
http://gismichigan.opendata.arcgis.com/datasets?q=Wellogic&sort_by=relevance 
 
Westjohn, D. B., and Thomas Lynn Weaver. 1998. Regional Aquifer-System Analysis: 

Michigan Basin Aquifer System. Hydrogeologic Framework of the Michigan Basin 
Regional Aquifer System. United States Geological Survey. 

 
Xie, Hongjie, Xiaobing Zhou, Enrique R. Vivoni, Jan M.H. Hendrickx, and Eric E. Small. 

2005. GIS-Based NEXRAD Stage III Precipitation Database: Automated 
Approaches for Data Processing and Visualization. Computers & Geosciences 
31 (1): 65–76.  

 
Xie, Pingping, and An-Yuan Xiong. 2011. A Conceptual Model for Constructing High-

Resolution Gauge-Satellite Merged Precipitation Analyses. Journal of 
Geophysical Research: Atmospheres 116 (D21): D21106.  

 
Xie, Yueqing, Peter G. Cook, Craig T. Simmons, Daniel Partington, Russell Crosbie, and 

Okke Batelaan. 2017. Uncertainty of Groundwater Recharge Estimated from a 
Water and Energy Balance Model. Journal of Hydrology, August.  

 
Young, C. Bryan, A. Allen Bradley, Witold F. Krajewski, Anton Kruger, and Mark L. 

Morrissey. 2000. Evaluating NEXRAD Multisensor Precipitation Estimates for 
Operational Hydrologic Forecasting. Journal of Hydrometeorology 1 (3): 241–54.  

 
Zhang, Jien, Benjamin S. Felzer, and Tara J. Troy. 2016. Extreme Precipitation Drives 

Groundwater Recharge: The Northern High Plains Aquifer, Central United States, 
1950–2010. Hydrological Processes 30 (14): 2533–45.  

 
Zhang, Zhiqiang, Shenping Wang, Ge Sun, Steven G. McNulty, Huayong Zhang, Jianlao 

Li, Manliang Zhang, Eduard Klaghofer, and Peter Strauss. 2008. Evaluation of 
the MIKE SHE Model for Application in the Loess Plateau, China. Journal of The 
American Water Resources Association, Vol. 44(5): 1108-1120.  

 

 

132 

Zomlot, Z., B. Verbeiren, M. Huysmans, and O. Batelaan. 2015. Spatial 

Distribution of Groundwater Recharge and Base Flow: Assessment of 
Controlling Factors. Journal of Hydrology: Regional Studies 4, Part B 
(September): 349–68.  

 

133