Content uploaded by James J. Butler Jr.
Author content
All content in this area was uploaded by James J. Butler Jr. on Mar 04, 2024
Content may be subject to copyright.
Zipper et al. | Irrigation OpenET | 1 of 52
Estimating irrigation water use from remotely sensed evapotranspiration
1
data: Accuracy and uncertainties across spatial scales
2
3
Authors: Sam Zipper1,2,*, Jude Kastens3, Timothy Foster4, Brownie Wilson1, Forrest Melton5,
4
Ashley Grinstead1,6, Jillian M. Deines7, James J. Butler1, Landon T. Marston8
5
6
Affiliations:
7
1. Kansas Geological Survey, University of Kansas, Lawrence KS 66047
8
2. Department of Geology, University of Kansas, Lawrence KS 66045
9
3. Kansas Biological Survey & Center for Ecological Research, University of Kansas,
10
Lawrence KS 66047
11
4. School of Engineering, University of Manchester, Manchester, UK
12
5. Atmospheric Science Branch, Earth Science Division, NASA Ames Research Center,
13
Moffett Field, CA 94035
14
6. Department of Natural Resources and the Environment, University of Connecticut,
15
Storrs, CT 06269, United States
16
7. Earth Systems Predictability and Resiliency Group, Pacific Northwest National
17
Laboratory, Richland, WA 99354
18
8. Department of Civil and Environmental Engineering, Virginia Tech, Blacksburg VA
19
24061
20
*Correspondence to samzipper@ku.edu
21
22
Highlights [3-5, max 85 characters]:
23
● Compared ET-based irrigation volumes to reported water use data at multiple scales
24
● ET-based irrigation volumes and reports agreed best when averaged over multiple years
25
● Soil moisture storage change can bias ET-based irrigation volumes in individual years
26
● Variation among ET models is substantial relative to irrigation management actions
27
● ET-based irrigation tracking is promising but application-relevant uncertainties exist
28
29
30
This is a draft manuscript submitted to Agricultural Water Management for peer review
31
(February 28, 2024).
32
33
Zipper et al. | Irrigation OpenET | 2 of 52
Abstract: Irrigated agriculture is the dominant use of water globally, but most water withdrawals
34
are not monitored or reported. As a result, it is largely unknown when, where, and how much
35
water is used for irrigation. Here, we evaluated the ability of remotely sensed evapotranspiration
36
(ET) data, integrated with other datasets, to calculate irrigation water withdrawals and
37
applications in an intensively-irrigated portion of the central United States. We compared
38
irrigation calculations based on OpenET data with reported groundwater withdrawals from a
39
flowmeter database and hundreds of farmer irrigation application records at three spatial scales
40
(management area, water right group, and field). We found that ET-based calculations of
41
irrigation exhibited similar temporal patterns as flowmeter data, but tended to be positively
42
biased with substantially more interannual variability than reported pumping rate. Disagreement
43
between ET-based irrigation calculations and reported irrigation was strongly correlated with
44
annual precipitation. Agreement between calculated and observed ET was better for multi-year
45
averages than for individual years across all spatial scales. The selection of an ET model was
46
also an important consideration, as variability in calculated irrigation across an ensemble of
47
satellite-driven ET models was larger than the potential impacts of conservation measures
48
employed in the region. Linking individual wells to specific fields was challenging, but
49
uncertainties in calculating irrigation depths due to the above-mentioned factors exceeded
50
potential uncertainty from irrigation status and field boundary mapping. From these results, we
51
suggest key practices for working with ET-based irrigation data that include accurately
52
accounting for changes in root zone soil moisture for within-season applications, such as
53
irrigation scheduling, and conducting an application-specific evaluation of sources of
54
uncertainty. Remotely-sensed approaches have a high potential for improving scientific research
55
and water resource management through improved spatial and temporal characterization of
56
irrigation, but uncertainties must be resolved to fully realize this potential.
57
58
Keywords: OpenET, remote sensing, evapotranspiration, water management, High Plains
59
Aquifer, uncertainty
60
61
Graphical Abstract:
62
63
Zipper et al. | Irrigation OpenET | 3 of 52
1. Introduction
64
Irrigated agriculture is the dominant global user of water. Groundwater supplies an
65
estimated 40% of global irrigation, with this figure rising even higher in semi-arid/arid regions or
66
in drought years when surface water availability is limited (Gleeson et al., 2020). As such,
67
groundwater use plays a critical role in global food production and trade (Dalin et al., 2017) and
68
sustaining local and regional economies (Deines et al., 2020). However, groundwater use can
69
also lead to detrimental outcomes, such as the depletion of interconnected surface water
70
resources (de Graaf et al., 2019; Zipper et al., 2022), declining water levels and storage capacity
71
in regionally- and globally-important aquifers (Hasan et al., 2023; Jasechko et al., 2024), and
72
associated water scarcity and insecurity (D’Odorico et al., 2019; Marston et al., 2020). In many
73
agricultural settings without alternative water sources, pumping reductions are the only currently
74
viable tool available to reduce water abstraction and slow water table decline rates (Butler et al.,
75
2020).
76
Making informed management decisions requires information about pumping rates and
77
the anticipated impacts on the environment (Foster et al., 2020). However, management is
78
challenging because data on the locations, schedules, and volumes of groundwater withdrawals
79
are rarely available, even in data-rich countries like the United States (Marston, Abdallah, et al.,
80
2022). Given the paucity of groundwater pumping data, emerging application-ready remote
81
sensing products may be a valuable tool to fill this data gap (Melton et al., 2022). While
82
flowmeters on pumping wells directly monitor the amount of water coming out of the ground,
83
which we refer to here as ‘irrigation water withdrawals’, remotely sensed approaches typically
84
provide data for spatially distributed evapotranspiration (ET) rates. Satellite-based ET data can
85
then be incorporated into a water balance or statistical model to infer ‘irrigation water
86
applications’, or the amount of water that is applied to a field after accounting for losses
87
(Dhungel et al., 2020; Folhes et al., 2009; Foster et al., 2019). Like all modeled quantities,
88
however, these ET-based calculations of irrigation applications are subject to numerous
89
uncertainties, which can lead to inefficient or inequitable water management decisions if not
90
well-characterized (Foster et al., 2020).
91
Unfortunately, due to the lack of reliable irrigation water withdrawal and application data
92
for ground-truthing, there have been limited opportunities to evaluate the ability of ET-based
93
approaches to calculate irrigation withdrawals and applications. While many past studies have
94
sought to estimate irrigation water use using satellite-based ET data and other hydrological
95
variables such as soil moisture (Brocca et al., 2018; Dari et al., 2020; Ketchum et al., 2023),
96
these estimates have typically been evaluated against aggregated statistics or synthetic model
97
estimates of water use. Other studies use statistical or machine learning approaches to relate ET
98
to observed water use, but these approaches are limited in terms of their applicability outside of
99
the model training region (Filippelli et al., 2022; Majumdar et al., 2022; Wei et al., 2022). As a
100
result, there is a lack of knowledge about how effectively ET data can be translated into
101
irrigation water withdrawals and applications across different spatial scales, from an individual
102
field to a region, which are relevant to regulatory and management purposes.
103
Zipper et al. | Irrigation OpenET | 4 of 52
Here, we address this gap by comparing calculations of ET-based irrigation application
104
and reported irrigation at multiple spatial scales (management area, water right group, field).
105
Reported irrigation data is from both a high-quality flowmeter database of irrigation water
106
withdrawals and direct farmer-provided records of irrigation water applications (Figure 1).
107
Specifically, we ask:
108
(1) How well do irrigation calculations derived from remotely sensed data and other spatial
109
datasets agree with water withdrawal and application data from flowmeters and farmer
110
records?
111
(2) What are the major sources of uncertainty in calculating irrigation withdrawals and
112
applications using remotely-sensed ET data?
113
Addressing these questions provides insights into the potential for remotely sensed ET products
114
to address critical water management issues and highlights key future research needed to
115
operationalize these tools for irrigation mapping and water conservation assessment.
116
117
118
Figure 1. Overview of study including key input datasets (OpenET: Melton et al., 2022; gridMET:
119
Abatzoglou, 2013; AIM: Deines, Kendall, Crowley, et al., 2019), spatial scales, and study objectives. The
120
images show the area in and around the Sheridan-6 Local Enhanced Management Area (blue outline), the
121
location of which is shown in Figure 2.
122
123
2. Methods
124
125
2.1 Study areas and irrigation ground data
126
We conducted our comparison of ET-based irrigation calculations to in-situ measurements of
127
groundwater withdrawals and applications at three spatial scales that address different potential
128
use cases for remotely sensed irrigation data:
129
Zipper et al. | Irrigation OpenET | 5 of 52
(1) At the management area scale (Section 2.1.1), we compared ET-based volumes to total
130
irrigation water withdrawals within a 255 km2 groundwater management area, the
131
Sheridan-6 Local Enhanced Management Area (SD-6 LEMA; blue area in Figure 1 and
132
Figure 2).
133
(2) At the water right scale (Section 2.1.1), we subdivided the SD-6 LEMA into water right
134
groups (WRGs) made up of non-overlapping combinations of pumping wells, fields, and
135
authorized places of use and compared ET-based irrigation volumes to total water
136
withdrawals within each WRG.
137
(3) At the field scale (Section 2.1.2), we compared ET-based calculated irrigation depths to
138
field-resolution irrigation water application data from fields where farmers voluntarily
139
shared irrigation records (field-years of data by region shown in Figure 2 in parenthesis).
140
Conducting our analysis at these three spatial scales allowed us to leverage independent data
141
sources for comparison (a state database at the management area and water right scale, farmer
142
records at the field scale) and address different aspects of uncertainty (i.e., linking locations of
143
water withdrawals to locations of applications was required at the water right scale).
144
145
146
Figure 2. Map of the state of Kansas subdivided into agricultural reporting districts. The location of the
147
Sheridan-6 (SD-6) Local Enhanced Management Area is shown in blue. The number of field-years of data
148
at the field-resolution scale are shown in parentheses for the northwest (NW), north-central (NC), west-
149
central (WC), and southwest (SW) reporting districts within the state. The Kansas portion of the High
150
Plains Aquifer is shown in gray.
151
152
2.1.1 Sheridan-6 Local Enhanced Management Area
153
The Sheridan-6 Local Enhanced Management Area (SD-6 LEMA) covers 255 km2 in
154
northwest Kansas, much of which is used to grow irrigated corn, soybeans, sorghum, and wheat
155
(Figure 2; Figure S1). The SD-6 LEMA was formed when local irrigators, concerned about
156
declining groundwater levels, proposed an allocation of 1397 mm (55”) of water over a five-year
157
period, which represented an approximate 20% reduction in pumping rates compared to
158
historical averages (Drysdale & Hendricks, 2018). After approval by the state’s chief engineer,
159
this allocation was codified in law for a five year period beginning in 2013. The irrigators within
160
the SD-6 LEMA have since renewed for two additional five year periods (2018-2022 and 2023-
161
2027). To date, the SD-6 LEMA exceeded the original conservation goals and reduced irrigation
162
Zipper et al. | Irrigation OpenET | 6 of 52
water withdrawals by 26-31% (Deines, Kendall, Butler, et al., 2019; Drysdale & Hendricks,
163
2018) and slowed water table decline rates (Butler et al., 2020; Whittemore et al., 2023) with
164
only minor negative impacts on yield and none on profitability (Golden, 2018). As such, the SD-
165
6 LEMA is a successful example of irrigator-driven groundwater conservation (Marston, Zipper,
166
et al., 2022) and has motivated the development of additional conservation approaches around
167
the state (Steiner et al., 2021).
168
We selected the SD-6 LEMA as the focus of our management area and water right scale
169
comparison because conservation practices have led to high irrigation efficiencies of producers
170
in the SD-6 LEMA with relatively little wasted irrigation water (e.g., deep percolation from
171
return flows or major fluxes of soil evaporation caused by excessive irrigation; Deines et al.,
172
2021). High irrigation efficiency suggests that irrigation water withdrawals and applications
173
should be approximately equal to each other and ET-based approaches may be particularly
174
effective for calculating irrigation volumes in this setting. Additionally, due to numerous past
175
studies of groundwater use in the SD-6 LEMA (Deines et al., 2021; Deines, Kendall, Butler, et
176
al., 2019; Dhungel et al., 2020; Drysdale & Hendricks, 2018; Glose et al., 2022; Whittemore et
177
al., 2023), we have a high degree of confidence in the accuracy of the irrigation withdrawal data
178
for the SD-6 LEMA.
179
Irrigation withdrawal data were aggregated from the Water Information Management and
180
Analysis System (WIMAS; https://geohydro.kgs.ku.edu/geohydro/wimas/) database maintained
181
by the Kansas Department of Agriculture - Division of Water Resources and the Kansas
182
Geological Survey. Withdrawal data are at the resolution of points of diversion, which in the SD-
183
6 region correspond exclusively to pumping wells since there are no surface water resources used
184
for irrigation. The data are high quality, as all non-domestic pumping wells in the state of Kansas
185
are required to use a totalizing flow meter subject to accuracy checks from the Kansas
186
Department of Agriculture (Butler et al., 2016). We also used reported irrigated acreage from the
187
WIMAS database, though unlike water use, the reported irrigated acreage is not subject to
188
verification and therefore the accuracy is unknown. In the SD-6 LEMA, we conducted our
189
comparison at two spatial scales:
190
● For the management area scale comparison, we summed the total annual withdrawals
191
from all irrigation wells within the SD-6 LEMA boundaries. For any water rights that had
192
authorized places of use both inside and outside the LEMA (n = 9, or 6% of the total
193
water right groups), we scaled the total water use based on the proportion of total
194
estimated irrigated area that was within the LEMA for that well. This is the approach
195
used in Brookfield et al. (2023) and is extended here through additional analyses of
196
uncertainty, the use of effective precipitation for estimating irrigation depths, and
197
comparison to other spatial scales.
198
● For the water right group (WRG) scale comparison, we established non-overlapping
199
groups of water withdrawals and applications by combining wells, water rights, and
200
authorized places of use as in Earnhart & Hendricks (2023). This aggregation was
201
necessary due to the complexities of agricultural water management that make it
202
Zipper et al. | Irrigation OpenET | 7 of 52
impossible to quantify the water use for a specific field from the WIMAS data alone: (i) a
203
single well may provide water to multiple fields; (ii) a single field may receive water
204
from multiple wells; (iii) a single water right may cover multiple wells and fields; and
205
(iv) irrigators are only required to report the authorized place of use and the total number
206
of acres irrigated, not the specific locations where water was used within the authorized
207
area in a specific year. For each WRG, we then summed the total reported annual water
208
withdrawals for all wells within the WRG. To evaluate potential errors associated with
209
defining WRGs, we also compared reported irrigated acreage for all the wells in the
210
WRG (from WIMAS) to estimated irrigated acreage based on the fields mapped as
211
irrigated within the authorized place of use for each WRG. Irrigated fields were identified
212
based on previously published remotely-sensed irrigation maps for each year in the
213
Annual Irrigation Maps (AIM) dataset (Deines, Kendall, Crowley, et al., 2019).
214
The SD-6 LEMA comparisons were conducted for the period 2016-2020, as that is the extent
215
covered by all necessary input datasets (described in Section 2.2).
216
217
2.1.2 Individual fields
218
For an independent additional comparison, we also collected field-resolution irrigation
219
application information from four farmers willing to share this information with us. Farmers
220
were contacted directly based on existing personal relationships and through regional
221
organizations such as groundwater management districts and asked to provide applied irrigation
222
volumes for as many fields as they were willing to share at the finest possible temporal
223
resolution. We also requested either data files or annotated pictures showing the irrigated extent
224
for each field so we could extract satellite-based ET data for each field. Unlike the management
225
area and WRG scale comparisons, therefore, for the field-scale comparison we had reported data
226
specifying actual places of use and irrigated extent. Irrigation data we received included minute-
227
resolution water use from irrigation control software, irregularly timed sub-annual water use
228
based on periodic visits to flowmeters, and annual values based on flowmeter data that farmers
229
associated with specific fields. For this study, all data were aggregated to the annual total depth
230
of applied irrigation. In total, we received data for 43 fields between 2016 and 2022, totaling 239
231
field-years of data. To protect the privacy of the farmers involved (Zipper, Stack Whitney, et al.,
232
2019), the locations of the fields are only shown here at the resolution of federal agricultural
233
reporting districts (Figure 2). The data span three of the five reporting districts that overlie the
234
High Plains Aquifer, with the most fields in west-central and northwest Kansas (note: one field,
235
just across the border in Nebraska, is included with the NW Kansas district). None of the fields
236
included within this dataset are within the SD-6 LEMA.
237
238
2.2 Calculating irrigation from ET data
239
We integrated ET data with several other geospatial datasets to calculate irrigation
240
volumes and/or depths (Figure 1). We extracted OpenET data from Google Earth Engine at a
241
monthly time step for 2016-2022 (Melton et al., 2022). OpenET includes ET data from six
242
Zipper et al. | Irrigation OpenET | 8 of 52
different satellite-driven models, as well as an ensemble mean. The models included are
243
DisALEXI (Anderson et al., 2007, 2018), eeMETRIC (Allen et al., 2005, 2007, 2011),
244
geeSEBAL (Bastiaanssen et al., 1998; Laipelt et al., 2021), PT-JPL (Fisher et al., 2008), SIMS
245
(Melton et al., 2012; Pereira, Paredes, Melton, et al., 2020), and SSEBop (Senay et al., 2022).
246
The ensemble mean was calculated as the mean of all models, with outlier values from the
247
ensemble identified based on median absolute deviations and removed prior to calculation of the
248
ensemble mean (Volk et al., 2024). The OpenET products were validated against 70 eddy
249
covariance towers deployed at agricultural sites spanning a range of climate and land cover
250
conditions across the western US and generally had a strong agreement, with all models within
251
+/- 15% of growing season mean flux tower ET averaged across all sites (Melton et al., 2022). A
252
subsequent evaluation affirmed the accuracy of the ET data from OpenET via comparison to a
253
total of 141 sites with eddy covariance towers, along with seven sites with Bowen ratio systems
254
and four weighing lysimeters, finding that the growing season ensemble ET values for cropland
255
had a mean absolute error of 78.1 mm (13.0%) and a mean bias error of -11.9 mm (2.0%). The
256
overall accuracy for cropland sites was the best of any land cover type evaluated, and
257
performance for annual crops, including corn, soybeans and wheat, was particularly strong (Volk
258
et al., 2024). However, there were no eddy covariance towers near our study area - the closest
259
irrigated fields with eddy covariance towers were in Mead, NE, where annual precipitation is
260
~50% greater than western Kansas - and therefore OpenET’s accuracy for irrigated agriculture in
261
semi-arid conditions typical of the western High Plains Aquifer has not been locally assessed.
262
OpenET data and precipitation data (from the 4 km gridMET data; Abatzoglou, 2013)
263
were averaged for each field. For the field-resolution comparison, field boundaries, crop type,
264
and irrigation status were defined based on information provided by farmers. For the
265
management area and WRG comparisons, field boundaries were defined based on a Kansas-
266
specific modification of the US Department of Agriculture (USDA) Common Land Unit dataset
267
(Gao et al., 2017; MardanDoost et al., 2019), annual crop type from the USDA Cropland Data
268
Layer (USDA, 2022), and irrigation status from AIM (Deines, Kendall, Crowley, et al., 2019).
269
For crop type and irrigation status, we summarized the rasterized input data to a single
270
categorical value for each field based on the most common raster value. To evaluate potential
271
impacts of this approach, we evaluated confidence in the field-resolution irrigation classifications
272
by evaluating the area of fields with a mixture of irrigated and non-irrigated pixels in the AIM
273
dataset. The irrigation confidence results suggested that this irrigation status mapping approach
274
was more likely to overestimate, rather than underestimate, irrigated area (Figure S3, Figure S4)
275
due to field boundaries not perfectly aligning with on-the-ground management divisions.
276
To calculate irrigation using our ET data (Figure 1), we calculated the precipitation
277
deficit (ET - effective precipitation) for each field (Figure S5) and masked it to only fields
278
mapped as irrigated (Figure S6). Effective precipitation was calculated as precipitation from
279
gridMET minus deep percolation out of the bottom of the root zone, which we estimated as a
280
function of precipitation based on 2013-2017 deep percolation estimates from Deines et al.
281
(2021) (regressions shown in Figure S9). This method does not account for soil moisture storage
282
Zipper et al. | Irrigation OpenET | 9 of 52
from year-to-year, so we carried out these calculations at three timescales: the growing season
283
(April-October), the calendar year (January-December), and the water year (October-September).
284
This allowed us to test the degree to which the timescale of aggregation influenced agreement
285
with reported irrigation withdrawal data. Since negative irrigation depths are not physically
286
possible, for any irrigated fields with a negative precipitation deficit we set the irrigation depth to
287
0 mm, though this was rare (Figure S5). Irrigation depth was calculated separately for each year
288
and each model (six ET models, as well as the ensemble mean). Since there are no surface water
289
rights in this region, we assumed that all irrigation was sourced from groundwater. Our approach
290
to estimating irrigation adopts several assumptions, including that there is minimal runoff or
291
fluxes of water apart from precipitation, irrigation, deep percolation and evaporation. While past
292
work has suggested that there is virtually no runoff under conservation practices in the SD-6
293
LEMA (Deines et al., 2021), these assumptions may be less appropriate in other parts of the
294
state, in particular the 4 field-years of data in the north-central region (Figure 2). Additionally,
295
there may be differences in the relationship between precipitation and deep percolation in other
296
regions given that irrigation efficiency is particularly high in the SD-6 LEMA. To assess the
297
potential impacts of our effective precipitation estimates on our findings, we repeated all
298
analyses using actual precipitation (instead of effective precipitation) in our precipitation deficit
299
calculations, and these results are shown in the Supplemental Information Section SI3.
300
301
3. Results
302
3.1 Management area comparison
303
At the scale of the SD-6 LEMA, the ET-based irrigation volumes are the same order of
304
magnitude as the reported withdrawal values but have a positive bias and greater interannual
305
variability (Figure 3, Table 1). Agreement between estimated irrigation and reported water
306
withdrawals is fairly similar regardless of whether irrigation is estimated based on the calendar
307
year, growing season, or water year. The best-performing model and timescale depend on the fit
308
metric being used (Table 1). For instance, the average mean absolute error (MAE) value across
309
all models was lowest for the water year-based irrigation volumes, but the growing season
310
irrigation volumes based on geeSEBAL had the lowest MAE of any model or timescale.
311
Broadly, we interpret the performance of satellite-based irrigation water withdrawals to be best
312
when the growing season is used as the temporal unit of aggregation as, averaged across all
313
models, this timescale has close to the lowest MAE, the slope closest to 1.0, and intermediate
314
bias and R2 compared to other timescales.
315
The relatively high R2 values we observe across all both the calendar year and growing
316
season timescales of aggregation (generally R2 > 0.9), combined with the relatively high MAEs
317
(~1-2 x107 m3, which is approximately equal to typical irrigation withdrawals for the
318
management area) and a slope lower than one (Table 1) collectively support our interpretation
319
that the ET-based irrigation calculations capture appropriate temporal patterns of variability in
320
estimated irrigation, but tend to overestimate both the average magnitude and degree of
321
interannual variability in irrigation volumes. As a result, when averaged across the full time
322
Zipper et al. | Irrigation OpenET | 10 of 52
period (2016-2020) the ET-based approaches tend to fall above the range of reported variability,
323
with the lowest bias from geeSEBAL and the highest bias from SIMS. The high calculated
324
irrigation volumes from SIMS make sense due to the formulation of this model, which assumes
325
well-watered conditions sufficient to meet the needs of the satellite-observed crop density
326
(Melton et al., 2012). Even irrigated crops in this region likely experience periodic water stress
327
during the growing season, as evidenced by the narrow distribution of SIMS ET data with
328
respect to other models (Figure S7). Since the overestimates we observed suggest that our
329
estimated effective precipitation values may be too low, we repeated our analyses using actual
330
precipitation in our irrigation calculations (Section SI3). In this case, we found that the ET-based
331
irrigation calculations better captured the central tendency of the reported data but have greater
332
year-to-year variability characterized by underestimates in wet years and overestimates in dry
333
years (Figure S10a-c, Figure S12). As a result, the agreement between the reported and
334
calculated irrigation volumes based on actual precipitation was substantially better when
335
averaged across multiple years because this averaged out opposing positive and negative errors
336
in dry and wet years, respectively (Figure S10d-f).
337
At the annual resolution, the differences between the ET-based irrigation volumes and the
338
reported groundwater withdrawals are strongly responsive to growing season weather conditions,
339
whether irrigation was calculated using effective precipitation (Figure 4a) or actual precipitation
340
(Figure S12). The ET-based approaches overestimated the metered irrigation volumes the most
341
in dry years, such as 2020, and the least in wet years, such as 2019 (Figure 4a). Based on this, we
342
tested whether precipitation could be used to statistically adjust ET-based irrigation calculations
343
to better match reported irrigation volumes (Figure 4b). For each ET model, we developed a
344
linear regression between the irrigation volume residual (shown for the ensemble mean in Figure
345
4a) and used this linear relationship to adjust ET-based irrigation calculations. The resulting
346
precipitation-adjusted irrigation calculations, shown in Figure 4b, had a substantially better
347
agreement with reported irrigation values, with reductions in MAE by one to two orders of
348
magnitude, and four of the models and the ensemble mean had slopes between 0.9 and 1.1 after
349
adjustment (full fit statistics in Table S1). Relationships were similarly strong when using actual,
350
instead of effective, precipitation for irrigation calculations (Figure S12).
351
352
Zipper et al. | Irrigation OpenET | 11 of 52
353
Figure 3. Comparison between reported WIMAS pumping and ET-based irrigation volumes from (a, d)
354
annual, (b, e) growing season, and (c, f) water year precipitation deficit over the entire SD-6 LEMA. The
355
left column (a-c) shows a comparison at annual resolution, and the right column (d-f) shows the five-year
356
average as a point with the five-year range as error bars. The gray shading in the background shows the
357
range of the reported values over the period.
358
359
Zipper et al. | Irrigation OpenET | 12 of 52
360
Table 1. Fit statistics for annual-resolution LEMA-scale OpenET-WIMAS comparison for each timescale
361
of aggregation and model. ‘C.Y.’ = Calendar Year, ‘G.S.’ = Growing Season, ‘W.Y.’ = Water Year.
362
Yellow shading indicates model with the best performance for that statistic and timescale. Blue shading
363
indicates best performance for that fit metric across all timescales. The red shading in the ‘Average’ row
364
indicates the best performing timescale across all models. Results shown in bar-chart form in Figure S8.
365
MAE [x107 m3]
Bias [%]
Slope
R2
Model
C.Y.
G.S.
W.Y.
C.Y.
G.S.
W.Y.
C.Y.
G.S.
W.Y.
C.Y.
G.S.
W.Y.
DisALEXI
1.33
0.95
1.19
65%
47%
51%
0.41
0.43
0.23
0.71
0.69
0.32
eeMETRIC
2.20
2.00
1.86
109%
99%
92%
0.38
0.39
0.29
0.97
0.88
0.64
Ensemble
1.52
1.37
1.22
75%
68%
60%
0.42
0.47
0.34
0.98
0.93
0.67
geeSEBAL
0.84
0.60
0.89
30%
25%
19%
0.36
0.40
0.31
0.96
0.89
0.71
PT-JPL
1.28
1.11
1.00
63%
55%
49%
0.42
0.46
0.33
0.98
0.92
0.64
SIMS
3.36
2.93
3.04
166%
144%
150%
0.55
0.61
0.42
0.98
0.93
0.57
SSEBop
0.99
1.23
0.86
49%
61%
34%
0.44
0.49
0.37
0.97
0.91
0.69
Average
1.65
1.46
1.44
79%
71%
65%
0.42
0.46
0.32
0.94
0.88
0.61
366
367
Zipper et al. | Irrigation OpenET | 13 of 52
368
Figure 4. Impacts of precipitation on ET-based irrigation calculations. (a) Difference between ET-based
369
calculated irrigation volume (from the OpenET ensemble) and reported water withdrawals for the SD-6
370
LEMA as a function of total growing season precipitation. A positive value means that the ET-based
371
irrigation volume was higher than the reported total. The red line indicates a linear best-fit with a shaded
372
standard error confidence interval (R2 = 0.96) and points are labeled by year. (b) Precipitation-adjusted
373
irrigation volumes for each model compared to reported irrigation volumes. In this plot, each ET-based
374
irrigation calculation was statistically adjusted as a function of precipitation, for example as shown in
375
Figure 4a for the Ensemble. Axis limits in Figure 4b are the same as Figure 3e for comparison.
376
377
3.2 SD-6 LEMA water right group comparison
378
The WRG comparison, like the SD-6 LEMA total comparison, revealed that there was a
379
positive correlation between calculated and reported irrigation, but a general positive bias of
380
calculated irrigation and improved agreement when averaged over multiple years. For the WRG-
381
scale comparison, the growing season-based irrigation volumes from the ensemble ET were
382
used. As with the management area comparison, the estimated irrigation volumes showed
383
substantially more interannual variability than reported irrigation volumes at the WRG scale,
384
with ET-based irrigation volumes higher than reported volumes for most WRGs and years, and
385
the greatest positive bias during dry years such as 2020 (Figure 5a). When averaged across all
386
five years, the scatter in the agreement between estimated and reported irrigation volumes was
387
dramatically reduced (Figure 5c), though the calculated irrigation volumes tended to be
388
positively biased related to reported irrigation as in the management area analysis (Figure 3).
389
Zipper et al. | Irrigation OpenET | 14 of 52
The correlation between estimated and reported irrigation was worse for irrigation depths
390
(Figure 5b, Figure 5d) than volumes (Figure 5a, Figure 5c), though irrigation volumes were more
391
consistently positively biased than depths. Overall, our results indicate that uncertainty in
392
estimated irrigation depth is greater than uncertainty in estimated irrigated area, which is further
393
supported by the field-scale comparison in Section 3.3 and has been observed in other ET-based
394
irrigation comparisons in Nevada and Oregon (Ott et al., 2024). Nevertheless, place of use and
395
irrigation status are potential drivers of some disagreement between calculated and reported
396
irrigation volumes. The irrigated area within WRGs based on annual irrigation maps (Deines,
397
Kendall, Crowley, et al., 2019) very rarely closely matched the reported irrigated area in the
398
WIMAS database. While there was a positive correlation between reported and estimated
399
irrigated area, differences between these two numbers exceeded 10% in 634 of 685 WRG-years,
400
and estimated irrigated area was often higher than reported irrigated area (Figure 6). On average,
401
the estimated irrigated area was 36% higher than the reported irrigated area (median = 28.4%).
402
This indicates that overestimated irrigated area may contribute to overestimated irrigation
403
volumes at both the management area scale and the WRG scale
404
While irrigated area is required for annual water use reports, water use reports do not
405
include spatial information specifying where the water was actually used, and total irrigated area
406
is not subject to verification or enforcement penalties (unlike reported water use). Therefore, it is
407
unknown how accurate the reported data are, but one plausible explanation for the disagreement
408
in estimated and reported irrigated area is uncertainty in field or parcel boundaries, particularly
409
related to corners of parcels that are irrigated with center-pivot systems. Since the field boundary
410
dataset we are using was originally based on 2007 common land units (CLUs) mapped by the
411
USDA with some refinements (Gao et al., 2017), it may not accurately delineate fields that
412
harbor differently managed component areas. For example, a square quarter section containing a
413
center pivot might consist of separate CLUs for the irrigated circle and the non-irrigated corners,
414
or it might simply be the quarter section boundary with multiple records for differently managed
415
subfields used when the farmer signs up for federal government programs such as crop
416
insurance. In the latter case, the entire field would be classified as irrigated based on our
417
assignment of irrigation by majority, even though the ~20% of the field in the corners would not
418
be reported as irrigated by the farmer. This is consistent with our observation that there tended to
419
be more low-confidence classifications for irrigated fields than non-irrigated fields (Figure S3).
420
Areas of low-confidence classifications were often field corners (Figure S4), suggesting that the
421
misclassification of non-irrigated corners as irrigated due to insufficiently refined field
422
boundaries may inflate our irrigated area estimates.
423
To assess the potential impacts of errors in irrigated area classification, we repeated the
424
analysis using only WRGs where the reported and estimated irrigated area agreed within 10%
425
(Figure S23). The results of this comparison had a smaller positive bias for both irrigation
426
volumes and depths, with overall the best agreement observed for multi-year average volumes
427
(Figure S23c). While the annual-resolution irrigation depths had a similar overall correlation
428
(Figure 5b and Figure S23b), the correlation between five-year average calculated and reported
429
Zipper et al. | Irrigation OpenET | 15 of 52
irrigation depth improved when only using WRGs with strong irrigated area agreement (R2 =
430
0.25, Figure S23d) compared to using all WRGs within the LEMA (R2 = 0.01, Figure 5d).
431
432
433
Figure 5. Comparison of reported irrigation for each water right group (WRG) to ET-based irrigation
434
calculation using the ensemble ET. (a) Annual irrigation volume for each WRG; (b) Annual irrigation
435
depth for each WRG; (c) Average irrigation volume for each WRG; (d) Average irrigation depth for each
436
WRG. In each plot, the gray line shows a 1:1 agreement between reported and estimated irrigation.
437
438
Zipper et al. | Irrigation OpenET | 16 of 52
439
Figure 6. Comparison between reported irrigated area (from WIMAS) and estimated irrigated area (from
440
AIM and authorized places of use) within each water right group in the SD-6 LEMA. Points colored
441
orange have an agreement within 10% and the orange line shows 1:1 agreement.
442
443
3.3 Field-scale comparison
444
At the field scale, we again observed better agreement between calculated and reported
445
irrigation when averaging across multiple years than when looking at individual years (Figure 7,
446
Table 2). At the annual resolution, there was not a strong correlation between calculated and
447
reported irrigation (average R2 across models = 0.16; Table 2). However, the range of calculated
448
irrigation depths matched the reported depths fairly well, unlike the management area and WRG
449
scales where we observed more consistent overestimates by the ET-based approaches, especially
450
during dry years (Figure 4, Figure 5). Since the fields included here were not part of the SD-6
451
LEMA, this may reflect lower irrigation efficiencies and increased non-evaporative losses (such
452
as deep percolation or runoff), particularly since our effective precipitation relationship was
453
based on the data from the SD-6 LEMA (Figure S9). Agreement for individual years did not
454
appear to vary systematically as a function of the region within the state, though the dataset was
455
not evenly distributed among regions with the large majority of the fields in either west-central
456
or north-west Kansas (71.5% and 21.8% of total field-years, respectively; Figure 2) which are
457
climatically very similar.
458
The choice of model also contributed to variability for both individual years and multi-
459
year averages. As we observed for the management area comparison, at the field scale we found
460
a consistent rank ordering, with the lowest calculated irrigation depths by geeSEBAL and the
461
highest by SIMS at both the scale of individual years and multi-year average (Figure 7). When
462
averaged across multiple years, the error in each model was substantially reduced (Table 2). For
463
the multi-year average, we observed the best overall performance by SSEBop, which had the
464
lowest MAE (90 mm), smaller bias than most other models (16.7%), a slope close to one (1.06),
465
Zipper et al. | Irrigation OpenET | 17 of 52
and the highest R2 (0.56) of all models. The DisALEXI and Ensemble irrigation depth
466
calculations also agreed more closely with the reported data than other models, with comparable
467
MAE (99 and 100 mm, respectively) and a slope relatively close to one for the multi-year
468
average.
469
470
471
Figure 7. Comparison between reported and calculated irrigation for individual fields. The top row shows
472
annual reported irrigation and the bottom row shows the multi-year average, both colored by the region
473
within the state. In each panel, the gray line indicates 1:1 agreement and the blue lines in the bottom
474
panels show a linear best-fit with a shaded standard error confidence interval.
475
476
Table 2. Fit statistics for field-resolution comparison between calculated and reported irrigation
477
application depths.
478
MAE [mm]
Bias [%]
Slope
R2
Model
Annual
Multi-Year
Annual
Multi-Year
Annual
Multi-Year
Annual
Multi-Year
DisALEXI
146
99
14.7
12.2
0.47
0.87
0.17
0.36
eeMETRIC
223
190
53.0
49.6
0.33
0.72
0.16
0.42
Ensemble
143
100
16.9
14.4
0.52
0.95
0.20
0.40
geeSEBAL
157
124
-29.5
-31.3
0.54
0.89
0.21
0.37
PT-JPL
140
102
-0.9
-3.3
0.47
0.65
0.11
0.17
SIMS
280
259
72.4
68.2
0.28
0.51
0.06
0.13
SSEBop
144
90
16.7
14.4
0.49
1.06
0.24
0.56
Average
176
138
20.5
17.7
0.44
0.81
0.16
0.34
479
480
Zipper et al. | Irrigation OpenET | 18 of 52
4. Discussion
481
We found that there was generally a positive correlation between calculated and reported
482
irrigation at the management area, WRG, and field scales. However, there was substantially
483
more variability in the ET-based irrigation calculations than reported irrigation, with calculated
484
irrigation often higher than reported irrigation, particularly in dry years. As a result, agreement
485
between reported and calculated irrigation tended to improve when averaged across multiple
486
years. Here, we discuss key sources of uncertainty that may have contributed to differences
487
between reported and calculated irrigation and how those may affect the utility of ET-based
488
irrigation products for research and management.
489
490
4.1 Sources of uncertainty in estimating irrigation from ET data
491
We identified and evaluated several sources of uncertainty that may explain differences
492
between satellite ET-based and reported irrigation water withdrawals and applications, including
493
(i) accounting for non-evaporative water balance components such as changes in soil moisture
494
storage; (ii) variability among models; (iii) linking fields to wells; and (iv) uncertainty in the
495
input datasets that are integrated with ET data to calculate irrigation.
496
Year-to-year changes in soil moisture appeared to be a primary driver of disagreements
497
between the estimated and reported irrigation at all three spatial scales. Since our approach relies
498
on a relatively simple water balance (ET - effective precipitation) to estimate applied irrigation,
499
this suggests that irrigation is being overestimated by the ET-based approaches in dry years such
500
as 2020 because soil moisture storage in the root zone is being drawn down (Figure 4a). Holding
501
all other aspects of the water balance constant, if soil moisture storage decreased during the dry
502
2020 growing season, this would cause an overestimate of irrigation since some of the ET in
503
2020 was using soil moisture that fell in previous years, such as the relatively wet 2019. One
504
contributing factor to our observed overestimates of irrigation may be the relatively simple
505
approach we used to estimate effective precipitation, which was based on a regional regression
506
for deep percolation (Figure SI9), and assumes there is no runoff in the region based on past
507
work (Deines et al., 2021). Since repeating our analysis using irrigation calculated from
508
precipitation (shown in Section SI3) generally showed less bias but greater interannual
509
variability than the effective precipitation approach we used, it suggests that our effective
510
precipitation calculation may be overestimating deep percolation losses, and as a result
511
underestimating the total volume of irrigation applied. The consistent positive precipitation
512
deficit for rainfed corn (Figure 8) further suggests that effective precipitation is being
513
underestimated by our approach, and calculating effective precipitation using a field-specific soil
514
water balance model approach such as ETDemands (Allen et al., 2020) could help to improve
515
overall agreement.
516
Variability in individual producer irrigation behavior across years may also contribute to
517
the increased interannual variability in the ET-based irrigation volumes observed in Figure 3
518
compared to the reported irrigation volumes. For example, previous research in the neighboring
519
state of Nebraska has shown that metered groundwater use typically exceeds crop water
520
Zipper et al. | Irrigation OpenET | 19 of 52
requirements in wetter or average rainfall years while farmers are observed to adopt more water-
521
efficient irrigation practices in drier years to reduce non-consumptive water losses, likely
522
motivated by a combination of the higher costs of irrigation and greater likelihood of
523
experiencing irrigation system capacity constraints in drought years (Foster et al., 2019). ET-
524
based approaches to calculating irrigation are also unable to capture other water fluxes such as
525
surface runoff. While runoff may be a source of error in our simple water balance approach for
526
some locations (e.g. fields with larger slopes), it is regionally a small component of the water
527
balance and is unlikely to explain systematic patterns of model errors observed across our study
528
area (Deines et al., 2021). Furthermore, our ET-based irrigation volumes did not account for
529
leakage in irrigation systems and other losses of water between where it is pumped from the
530
ground but before it reaches the field, though based on the high efficiency in the SD-6 LEMA
531
area we expect that these losses are minimal. These findings suggest that, for annual or finer
532
temporal resolutions, the use of more complex water balance approaches, such as soil water
533
balance models (Dhungel et al., 2020; Kharrou et al., 2021; Pereira, Paredes, & Jovanovic,
534
2020), will be necessary to accurately disentangle the rates, locations, and timing of irrigation
535
applications, and there may be promise through the assimilation of additional data sets such as in
536
situ or remotely sensed soil moisture (Dari et al., 2020; Filippelli et al., 2022; Jalilvand et al.,
537
2019).
538
The selection of ET model also led to substantial variability in the estimated irrigation
539
depths, with a relatively consistent ordering across models (from lowest to highest): geeSEBAL,
540
DisALEXI, PT-JPL, SSEBop, Ensemble, eeMETRIC, SIMS (Figure 3, Figure 7). Since the
541
effective precipitation input data used to estimate irrigation was the same for all models, this
542
variability in estimated irrigation among the models can be attributed to entirely differences in
543
the approaches used by each ET model, and variability can be quite substantial. For example, for
544
irrigated corn in the SD-6 LEMA, the medians span ~156-270 mm across ET models in a given
545
year (Figure 9), which approaches the magnitude of total applied irrigation water and greatly
546
exceeds the magnitude of the conservation actions put in place in this region (Whittemore et al.,
547
2023). The variability among models may be due to differences in the approaches to computation
548
of the sensible heat flux used in each of the five energy balance models, differences in the spatial
549
scale of key meteorological inputs for the DisALEXI, PT-JPL and geeSEBAL models, and
550
model assumptions, especially for SIMS, which assumes well-watered conditions. This
551
underscores the importance of local model accuracy assessments to identify the models that
552
perform best for the crop types and irrigation management practices that are most prevalent in
553
the region. In the absence of suitable independent dataset for use in a local or regional accuracy
554
assessment, OpenET recommends use of the ensemble ET value, which has been shown to
555
perform best overall for the western U.S. across most accuracy metrics (Melton et al., 2022;
556
Volk et al., 2024). We found that the model ensemble was generally among the best-performing
557
approaches to estimating irrigation (as observed in Volk et al., 2024 through comparison with
558
eddy covariance data), particularly after statistically adjusting to account for potential errors in
559
effective precipitation calculations (Figure 4b), suggesting that the ensemble would be a
560
Zipper et al. | Irrigation OpenET | 20 of 52
reasonable approach to use across our study region until additional local accuracy assessments
561
can be conducted.
562
Accurately linking the point of water diversion with the place where that water is applied
563
was a major challenge in our analysis and has been identified as a key source of uncertainty in
564
other domains (Ott et al., 2024). While developing these links may not be needed for many
565
applications, such as estimating regional water use (Figure 3) or field-based water management
566
(Figure 7), connecting the point of diversion with place of use is critical to evaluate irrigation
567
application depths and to assess the effectiveness of conservation measures and the ultimate
568
impacts of pumping on other aspects of regional agrohydrological systems such as streamflow
569
(Kniffin et al., 2020; Zipper, Carah, et al., 2019; Zipper et al., 2021), aquifer dynamics (Feinstein
570
et al., 2016; Peterson & Fulton, 2019; Wilson et al., 2021), or groundwater-dependent
571
ecosystems (Tolley et al., 2019). At the WRG scale, our ET-based calculations of irrigation
572
volume had better agreement than calculations of irrigation depth (Figure 5), consistent with
573
results from the nearby Colorado portion of the Republican River Basin (Filippelli et al., 2022).
574
The weaker relationship between calculated and reported irrigation depth, compared to irrigation
575
volume, reflects the importance of irrigated area as a determinant of overall irrigation volumes
576
(Lamb et al., 2021; Wei et al., 2022). Because irrigated area (which, despite challenges noted
577
elsewhere in the manuscript, e.g. Figure 6, is relatively easier to estimate than irrigation depth)
578
had a strong correlation between estimated and reported data, dividing out this term to convert
579
from irrigation volume to irrigation depth reduced the overall correlation between calculated and
580
reported irrigation applications. Despite exceptionally high-quality water use data for the state of
581
Kansas, the limited linkages between the point of diversion and actual place of use highlights a
582
key data gap for the application of remotely sensed irrigation data for hydrogeological research
583
and management, and a necessary improvement for field-level operationalization.
584
Estimating irrigation from satellite-based ET requires a variety of datasets in addition to
585
ET, such as irrigation status and precipitation (Figure 1), and each input dataset is subject to its
586
own uncertainties. For ET, we would expect errors in calculated irrigation to increase for periods
587
or regions with increased cloud cover that affect the optical and thermal bands of satellites used
588
by ET models. Since cloud cover is associated with precipitation events, this may have an
589
outsized effect on estimating ET during times when soil moisture is being replenished. The
590
gridded meteorological datasets we use here are also unable to capture the fine-scale variation in
591
precipitation dynamics that can occur during typical convective summer storms in semi-arid
592
regions (Gibson et al., 2017; Mourtzinis et al., 2017), which would have a stronger influence on
593
the field-scale comparisons included in this analysis. To assess this potential uncertainty, we
594
replicated our management area-scale analysis using National Weather Service Advanced
595
Hydrologic Prediction Service radar precipitation data, which is thought to better capture spatial
596
patterns in precipitation in western Kansas (Whittemore et al., 2023), instead of gridMET
597
precipitation. However, the results of the gridMET-based analysis and radar precipitation
598
analysis were very similar to approaches using gridMET precipitation (see Supplemental
599
Information, Section SI4 compared to Section SI3), indicating that precipitation uncertainty was
600
Zipper et al. | Irrigation OpenET | 21 of 52
not likely a major driver of differences between reported and calculated irrigation. Additionally,
601
irrigation mapping can be particularly challenging during wet years where the differences in
602
canopy cover and greenness between irrigated and non-irrigated fields are smaller (Xu et al.,
603
2019). While the irrigation extent dataset we used is the best-available for this region and
604
consistently shows differences in precipitation deficit between irrigated and rainfed corn, there is
605
also substantial overlap between their distributions, suggesting that some degree of
606
misclassification is practically assured (Figure 8). This may be particularly challenging in
607
relatively small unirrigated portions of otherwise irrigated fields, such as the non-irrigated
608
corners of center-pivot systems (Figure S4).
609
610
611
Figure 8. Distribution of field-resolution growing season ensemble ET - Effective Precipitation for corn
612
fields in the LEMA, separated by year and colored by irrigation status. The gray shaded interval shows
613
the average annual reported irrigation depth (182 mm) +/- one standard deviation (46 mm) over the 2016-
614
2020 period.
615
616
Zipper et al. | Irrigation OpenET | 22 of 52
617
Figure 9. Distribution of ET - precipitation for all irrigated corn fields in the LEMA, colored by model.
618
The gray shaded interval shows the average annual reported irrigation depth (182 mm) +/- one standard
619
deviation (46 mm) over the 2016-2020 period.
620
621
4.2 Utility for research and management purposes
622
As water becomes increasingly scarce, the importance of accurate accounting of how,
623
where, when, and how much water is being used is becoming more critical. In the US, each state
624
is responsible for administering water rights and regulating water use within their jurisdictional
625
boundaries. Water use metering and reporting requirements vary significantly between states.
626
Satellite-based ET data could provide a nationally consistent approach to computing
627
consumptive use of water applied for irrigation, and potentially for estimating the volume of
628
water applied for crop irrigation, which is the largest source of consumptive water use in the US
629
(Marston et al., 2018). However, these satellite-based irrigation calculations need to be
630
comparable to what is actually happening on the ground, demonstrating the importance of high-
631
fidelity in situ measurements of irrigation. This study was made possible by metered
632
groundwater pumping records detailing the location, amount, and timing of irrigation. Outside of
633
Kansas, metered records of irrigation are rare, with many states not requiring flowmeters on
634
agricultural water uses (Marston, Abdallah, et al., 2022). This gap is increasingly being filled
635
with reanalysis and ET-based water use products (Haynes et al., 2023; Martin et al., 2023). For
636
ET-based irrigation data to become more useful to researchers, irrigators, regulators, and
637
policymakers, metered irrigation records are needed for other areas with different soils, climate,
638
irrigation practices, and cropping patterns to evaluate the performance of ET-based irrigation
639
calculations under these different conditions.
640
The sources of uncertainty we discuss in Section 4.1 contributed to variable levels of
641
agreement between ET-based and reported water withdrawals and applications across the
642
Zipper et al. | Irrigation OpenET | 23 of 52
comparisons we conducted. At the management area scale, we found a generally strong positive
643
correlation (e.g., R2 generally above 0.85; Table 1), comparable to other studies using remotely
644
sensed data to estimate irrigation depths with statistical models (Filippelli et al., 2022; Majumdar
645
et al., 2022; Wei et al., 2022). However, we observed a general positive bias and substantially
646
more year-to-year variability in ET-based irrigation than in the reported data. The WRG and
647
field scale comparisons had weaker correlations than the management area scale, in particular for
648
irrigation depths, potentially indicating that irrigation depth predictions have greater uncertainty
649
than irrigation volumes. However, across all scales, the agreement between ET-based irrigation
650
calculations and reported data improved substantially when averaged over multiple years (Figure
651
3, Figure 5, Figure 7).
652
Since errors in estimated irrigation can lead to significant economic and hydrological
653
impacts if used for management purposes (Foster et al., 2020), continued methodological
654
development to overcome the uncertainties described above will be important to advance these
655
tools for some applications. For instance, for purposes that require estimating long-term average
656
consumptive use, such as calculating the water balance for a large (10s to 100s of km) region, the
657
precipitation-adjusted spatially- and temporally-aggregated results we show in Figure 4 might be
658
sufficient. In contrast, using these data for other purposes, such as monitoring within-season
659
irrigation timing and volume from a specific well, would require significant improvements in the
660
accuracy of calculated irrigation at these finer spatial and temporal scales and careful selection of
661
an appropriate ET model. We found that statistical adjustments to ET-based irrigation
662
calculations can substantially improve agreement with reported values at annual resolution
663
(Figure 4b, Table S1), potentially suggesting a path towards greater local accuracy, and
664
highlighting the critical importance of accurate effective precipitation values and ground-based
665
data for comparison. While the approach we used required reported irrigation data, and therefore
666
would not be tractable in locations without existing withdrawal monitoring, it may be possible to
667
use a limited subset of reporting locations to develop relationships that can be applied more
668
broadly (Bohling et al., 2021). Additional products, such as high-resolution soil moisture data
669
from remote sensing-model integration (Vergopolan et al., 2021), may also provide a pathway
670
for bias-correction and/or temporal disaggregation when integrated with field-specific water
671
balance modeling tools (Hoekstra, 2019). Given that OpenET is a relatively new product (Melton
672
et al., 2022), continued work on specific research and management applications will provide
673
useful targets for prioritizing efforts to reduce existing uncertainties.
674
675
5. Conclusions
676
We evaluated the agreement between ET-based calculations of irrigation using a simple water
677
balance approach and reported irrigation from a statewide database and farmer information. We
678
found that there were generally positive correlations between the ET-based approaches and
679
reported data, but that the ET-based approaches typically demonstrated more variability than
680
reported values and overestimated irrigation, particularly during dry years. This may be partially
681
attributed to changes in soil moisture storage and the approach used to calculate effective
682
Zipper et al. | Irrigation OpenET | 24 of 52
precipitation. We also found that agreement improved substantially when irrigation is averaged
683
over multiple years, particularly at the field resolution where irrigated area was well-constrained.
684
Key uncertainties were identified related to the choice of ET model and the approach used to
685
map irrigation status and link wells to fields where irrigation water is used. The uncertainties in
686
ET-based irrigation calculations likely exceed the signal of management activities in this region,
687
suggesting further methodological refinement is needed for applications requiring precise
688
quantification of irrigation depth for a given location and/or single year. However, for
689
applications focused on relative differences in irrigation intensity across space and/or multi-year
690
average irrigation applications, some of these uncertainties may safely be ignored. This work
691
suggests that ET-based approaches to calculating irrigation are a potentially valuable approach
692
for developing improved spatial and temporal water use data, and will likely require application-
693
specific targeted improvements to reduce key uncertainties.
694
695
6. Acknowledgments
696
We appreciate assistance from Will Carrara and John Woods with data acquisition/processing
697
and feedback on the manuscript from Sayantan Majumdar (Desert Research Institute). Ashley
698
Grinstead’s participation was supported by the Kansas Geological Survey Geohydrology
699
Internship Program (https://www.kgs.ku.edu/Hydro/gipIndex.html). We greatly appreciate the
700
farmers who were willing to share their water use data with us. This work was supported by
701
National Aeronautics and Space Administration (NASA) [grant number 80NSSC22K1276] and
702
National Science Foundation (NSF) [grant number RISE-2108196]. TF was also supported by
703
Innovate UK [award number 10044695], as part of the UK Research and Innovation and
704
European Commission funded project ‘TRANSCEND: Transformational and robust adaptation
705
to water scarcity and climate change under deep uncertainty’. Data and code used in the study
706
are available at https://github.com/samzipper/SD-6_MapWaterConservation during the review
707
process and will be posted to a repository with a DOI at the time of manuscript acceptance.
708
709
7. References
710
Abatzoglou, J. T. (2013). Development of gridded surface meteorological data for ecological
711
applications and modelling. International Journal of Climatology, 33(1), 121–131.
712
https://doi.org/10.1002/joc.3413
713
Allen, R. G., Tasumi, M., Morse, A., & Trezza, R. (2005). A Landsat-based energy balance and
714
evapotranspiration model in Western US water rights regulation and planning. Irrigation
715
and Drainage Systems, 19(3), 251–268. https://doi.org/10.1007/s10795-005-5187-z
716
Allen, R. G., Tasumi, M., Morse, A., Trezza, R., Wright, J. L., Bastiaanssen, W., et al. (2007).
717
Satellite-based energy balance for mapping evapotranspiration with internalized
718
calibration (METRIC)—Applications. Journal of Irrigation and Drainage Engineering,
719
133(4), 395–406.
720
Allen, R. G., Pereira, L. S., Howell, T. A., & Jensen, M. E. (2011). Evapotranspiration
721
information reporting: I. Factors governing measurement accuracy. Agricultural Water
722
Management, 98(6), 899–920. https://doi.org/10.1016/j.agwat.2010.12.015
723
Allen, R. G., Robison, C. W., Huntington, J., Wright, J. L., & Kilic, A. (2020). Applying the
724
Zipper et al. | Irrigation OpenET | 25 of 52
FAO-56 dual Kc method for irrigation water requirements over large areas of the
725
Western US. Transactions of the ASABE, 63(6), 2059–2081.
726
https://doi.org/10.13031/trans.13933
727
Anderson, M. C., Norman, J. M., Mecikalski, J. R., Otkin, J. A., & Kustas, W. P. (2007). A
728
climatological study of evapotranspiration and moisture stress across the continental
729
United States based on thermal remote sensing: 1. Model formulation. Journal of
730
Geophysical Research: Atmospheres, 112(D10). https://doi.org/10.1029/2006JD007506
731
Anderson, M. C., Gao, F., Knipper, K., Hain, C., Dulaney, W., Baldocchi, D., et al. (2018).
732
Field-Scale Assessment of Land and Water Use Change over the California Delta Using
733
Remote Sensing. Remote Sensing, 10(6), 889. https://doi.org/10.3390/rs10060889
734
Bastiaanssen, W. G. M., Menenti, M., Feddes, R. A., & Holtslag, A. A. M. (1998). A remote
735
sensing surface energy balance algorithm for land (SEBAL). 1. Formulation. Journal of
736
Hydrology, 212, 198–212.
737
Bohling, G. C., Butler, J. J., Whittemore, D. O., & Wilson, B. B. (2021). Evaluation of Data
738
Needs for Assessments of Aquifers Supporting Irrigated Agriculture. Water Resources
739
Research, 57(4), e2020WR028320. https://doi.org/10.1029/2020WR028320
740
Brocca, L., Tarpanelli, A., Filippucci, P., Dorigo, W., Zaussinger, F., Gruber, A., & Fernández-
741
Prieto, D. (2018). How much water is used for irrigation? A new approach exploiting
742
coarse resolution satellite soil moisture products. International Journal of Applied Earth
743
Observation and Geoinformation, 73, 752–766. https://doi.org/10.1016/j.jag.2018.08.023
744
Brookfield, A. E., Zipper, S., Kendall, A. D., Ajami, H., & Deines, J. M. (2024). Estimating
745
Groundwater Pumping for Irrigation: A Method Comparison. Groundwater, 62(1), 15–
746
33. https://doi.org/10.1111/gwat.13336
747
Butler, J. J., Whittemore, D. O., Wilson, B. B., & Bohling, G. C. (2016). A new approach for
748
assessing the future of aquifers supporting irrigated agriculture. Geophysical Research
749
Letters, 43(5), 2004–2010. https://doi.org/10.1002/2016GL067879
750
Butler, J. J., Bohling, G. C., Whittemore, D. O., & Wilson, B. B. (2020). Charting Pathways
751
Toward Sustainability for Aquifers Supporting Irrigated Agriculture. Water Resources
752
Research, 56(10), e2020WR027961. https://doi.org/10.1029/2020WR027961
753
Dalin, C., Wada, Y., Kastner, T., & Puma, M. J. (2017). Groundwater depletion embedded in
754
international food trade. Nature, 543(7647), 700–704.
755
https://doi.org/10.1038/nature21403
756
Dari, J., Brocca, L., Quintana-Seguí, P., Escorihuela, M. J., Stefan, V., & Morbidelli, R. (2020).
757
Exploiting High-Resolution Remote Sensing Soil Moisture to Estimate Irrigation Water
758
Amounts over a Mediterranean Region. Remote Sensing, 12(16), 2593.
759
https://doi.org/10.3390/rs12162593
760
Deines, J. M., Kendall, A. D., Crowley, M. A., Rapp, J., Cardille, J. A., & Hyndman, D. W.
761
(2019). Mapping three decades of annual irrigation across the US High Plains Aquifer
762
using Landsat and Google Earth Engine. Remote Sensing of Environment, 233, 111400.
763
https://doi.org/10.1016/j.rse.2019.111400
764
Deines, J. M., Kendall, A. D., Butler, J. J., & Hyndman, D. W. (2019). Quantifying irrigation
765
adaptation strategies in response to stakeholder-driven groundwater management in the
766
US High Plains Aquifer. Environmental Research Letters, 14(4), 044014.
767
https://doi.org/10.1088/1748-9326/aafe39
768
Deines, J. M., Schipanski, M. E., Golden, B., Zipper, S. C., Nozari, S., Rottler, C., et al. (2020).
769
Transitions from irrigated to dryland agriculture in the Ogallala Aquifer: Land use
770
Zipper et al. | Irrigation OpenET | 26 of 52
suitability and regional economic impacts. Agricultural Water Management, 233,
771
106061. https://doi.org/10.1016/j.agwat.2020.106061
772
Deines, J. M., Kendall, A. D., Butler, J. J., Basso, B., & Hyndman, D. W. (2021). Combining
773
Remote Sensing and Crop Models to Assess the Sustainability of Stakeholder-Driven
774
Groundwater Management in the US High Plains Aquifer. Water Resources Research,
775
e2020WR027756. https://doi.org/10.1029/2020WR027756
776
Dhungel, R., Aiken, R., Lin, X., Kenyon, S., Colaizzi, P. D., Luhman, R., et al. (2020).
777
Restricted water allocations: Landscape-scale energy balance simulations and
778
adjustments in agricultural water applications. Agricultural Water Management, 227,
779
105854. https://doi.org/10.1016/j.agwat.2019.105854
780
D’Odorico, P., Carr, J., Dalin, C., Dell’Angelo, J., Konar, M., Laio, F., et al. (2019). Global
781
virtual water trade and the hydrological cycle: patterns, drivers, and socio-environmental
782
impacts. Environmental Research Letters, 14(5), 053001. https://doi.org/10.1088/1748-
783
9326/ab05f4
784
Drysdale, K. M., & Hendricks, N. P. (2018). Adaptation to an irrigation water restriction
785
imposed through local governance. Journal of Environmental Economics and
786
Management, 91, 150–165. https://doi.org/10.1016/j.jeem.2018.08.002
787
Earnhart, D., & Hendricks, N. P. (2023). Adapting to water restrictions: Intensive versus
788
extensive adaptation over time differentiated by water right seniority. American Journal
789
of Agricultural Economics. https://doi.org/10.1111/ajae.12361
790
Feinstein, D. T., Fienen, M. N., Reeves, H. W., & Langevin, C. D. (2016). A Semi-Structured
791
MODFLOW-USG Model to Evaluate Local Water Sources to Wells for Decision
792
Support. Groundwater, 54(4), 532–544. https://doi.org/10.1111/gwat.12389
793
Filippelli, S. K., Sloggy, M. R., Vogeler, J. C., Manning, D. T., Goemans, C., & Senay, G. B.
794
(2022). Remote sensing of field-scale irrigation withdrawals in the central Ogallala
795
aquifer region. Agricultural Water Management, 271, 107764.
796
https://doi.org/10.1016/j.agwat.2022.107764
797
Fisher, J. B., Tu, K. P., & Baldocchi, D. D. (2008). Global estimates of the land–atmosphere
798
water flux based on monthly AVHRR and ISLSCP-II data, validated at 16 FLUXNET
799
sites. Remote Sensing of Environment, 112(3), 901–919.
800
https://doi.org/10.1016/j.rse.2007.06.025
801
Folhes, M. T., Rennó, C. D., & Soares, J. V. (2009). Remote sensing for irrigation water
802
management in the semi-arid Northeast of Brazil. Agricultural Water Management,
803
96(10), 1398–1408. https://doi.org/10.1016/j.agwat.2009.04.021
804
Foster, T., Gonçalves, I. Z., Campos, I., Neale, C. M. U., & Brozović, N. (2019). Assessing
805
landscape scale heterogeneity in irrigation water use with remote sensing and in situ
806
monitoring. Environmental Research Letters, 14(2), 024004.
807
https://doi.org/10.1088/1748-9326/aaf2be
808
Foster, T., Mieno, T., & Brozović, N. (2020). Satellite-Based Monitoring of Irrigation Water
809
Use: Assessing Measurement Errors and Their Implications for Agricultural Water
810
Management Policy. Water Resources Research, 56(11), e2020WR028378.
811
https://doi.org/10.1029/2020WR028378
812
Gao, J., Sheshukov, A. Y., Yen, H., Kastens, J. H., & Peterson, D. L. (2017). Impacts of
813
incorporating dominant crop rotation patterns as primary land use change on hydrologic
814
model performance. Agriculture, Ecosystems & Environment, 247, 33–42.
815
https://doi.org/10.1016/j.agee.2017.06.019
816
Zipper et al. | Irrigation OpenET | 27 of 52
Gibson, J., Franz, T. E., Wang, T., Gates, J., Grassini, P., Yang, H., & Eisenhauer, D. (2017). A
817
case study of field-scale maize irrigation patterns in western Nebraska: implications for
818
water managers and recommendations for hyper-resolution land surface modeling.
819
Hydrology and Earth System Sciences, 21(2), 1051–1062. https://doi.org/10.5194/hess-
820
21-1051-2017
821
Gleeson, T., Cuthbert, M., Ferguson, G., & Perrone, D. (2020). Global Groundwater
822
Sustainability, Resources, and Systems in the Anthropocene. Annual Review of Earth and
823
Planetary Sciences, 48(1). https://doi.org/10.1146/annurev-earth-071719-055251
824
Glose, T. J., Zipper, S., Hyndman, D. W., Kendall, A. D., Deines, J. M., & Butler, J. J. (2022).
825
Quantifying the impact of lagged hydrological responses on the effectiveness of
826
groundwater conservation. Water Resources Research, 58(7), e2022WR032295.
827
https://doi.org/10.1029/2022WR032295
828
Golden, B. (2018). Monitoring the Impacts of Sheridan County 6 Local Enhanced Management
829
Area. Kansas State University. Retrieved from https://agriculture.ks.gov/docs/default-
830
source/dwr-water-appropriation-documents/sheridancounty6_lema_goldenreport_2013-
831
2017.pdf?sfvrsn=dac48ac1_0
832
de Graaf, I. E. M., Gleeson, T., Beek, L. P. H. (Rens) van, Sutanudjaja, E. H., & Bierkens, M. F.
833
P. (2019). Environmental flow limits to global groundwater pumping. Nature, 574(7776),
834
90–94. https://doi.org/10.1038/s41586-019-1594-4
835
Hasan, M. F., Smith, R., Vajedian, S., Pommerenke, R., & Majumdar, S. (2023). Global land
836
subsidence mapping reveals widespread loss of aquifer storage capacity. Nature
837
Communications, 14(1), 6180. https://doi.org/10.1038/s41467-023-41933-z
838
Haynes, J. V., Read, A. L., Chan, A. Y., Martin, D. J., Regan, R. S., Henson, W., et al. (2023).
839
Monthly crop irrigation withdrawals and efficiencies by HUC12 watershed for years
840
2000-2020 within the conterminous United States [Data set]. U.S. Geological Survey.
841
https://doi.org/10.5066/P9LGISUM
842
Hoekstra, A. Y. (2019). Green-blue water accounting in a soil water balance. Advances in Water
843
Resources, 129, 112–117. https://doi.org/10.1016/j.advwatres.2019.05.012
844
Jalilvand, E., Tajrishy, M., Ghazi Zadeh Hashemi, S. A., & Brocca, L. (2019). Quantification of
845
irrigation water using remote sensing of soil moisture in a semi-arid region. Remote
846
Sensing of Environment, 231, 111226. https://doi.org/10.1016/j.rse.2019.111226
847
Jasechko, S., Seybold, H., Perrone, D., Fan, Y., Shamsudduha, M., Taylor, R. G., et al. (2024).
848
Rapid groundwater decline and some cases of recovery in aquifers globally. Nature,
849
625(7996), 715–721. https://doi.org/10.1038/s41586-023-06879-8
850
Ketchum, D., Hoylman, Z. H., Huntington, J., Brinkerhoff, D., & Jencso, K. G. (2023). Irrigation
851
intensification impacts sustainability of streamflow in the Western United States.
852
Communications Earth & Environment, 4(1), 1–8. https://doi.org/10.1038/s43247-023-
853
01152-2
854
Kharrou, M. H., Simonneaux, V., Er-Raki, S., Le Page, M., Khabba, S., & Chehbouni, A. (2021).
855
Assessing Irrigation Water Use with Remote Sensing-Based Soil Water Balance at an
856
Irrigation Scheme Level in a Semi-Arid Region of Morocco. Remote Sensing, 13(6),
857
1133. https://doi.org/10.3390/rs13061133
858
Kniffin, M., Bradbury, K. R., Fienen, M., & Genskow, K. (2020). Groundwater Model
859
Simulations of Stakeholder-Identified Scenarios in a High-Conflict Irrigated Area.
860
Groundwater, 58(6), 973–986. https://doi.org/10.1111/gwat.12989
861
Laipelt, L., Henrique Bloedow Kayser, R., Santos Fleischmann, A., Ruhoff, A., Bastiaanssen,
862
Zipper et al. | Irrigation OpenET | 28 of 52
W., Erickson, T. A., & Melton, F. (2021). Long-term monitoring of evapotranspiration
863
using the SEBAL algorithm and Google Earth Engine cloud computing. ISPRS Journal
864
of Photogrammetry and Remote Sensing, 178, 81–96.
865
https://doi.org/10.1016/j.isprsjprs.2021.05.018
866
Lamb, S. E., Haacker, E. M. K., & Smidt, S. J. (2021). Influence of Irrigation Drivers Using
867
Boosted Regression Trees: Kansas High Plains. Water Resources Research, 57(5),
868
e2020WR028867. https://doi.org/10.1029/2020WR028867
869
Majumdar, S., Smith, R., Conway, B. D., & Lakshmi, V. (2022). Advancing remote sensing and
870
machine learning-driven frameworks for groundwater withdrawal estimation in Arizona:
871
Linking land subsidence to groundwater withdrawals. Hydrological Processes, 36(11),
872
e14757. https://doi.org/10.1002/hyp.14757
873
MardanDoost, B., Brookfield, A. E., Feddema, J., Sturm, B., Kastens, J., Peterson, D., & Bishop,
874
C. (2019). Estimating irrigation demand with geospatial and in-situ data: Application to
875
the high plains aquifer, Kansas, USA. Agricultural Water Management, 223, 105675.
876
https://doi.org/10.1016/j.agwat.2019.06.010
877
Marston, L. T., Ao, Y., Konar, M., Mekonnen, M. M., & Hoekstra, A. Y. (2018). High-
878
Resolution Water Footprints of Production of the United States. Water Resources
879
Research, 54(3), 2288–2316. https://doi.org/10.1002/2017WR021923
880
Marston, L. T., Lamsal, G., Ancona, Z. H., Caldwell, P., Richter, B. D., Ruddell, B. L., et al.
881
(2020). Reducing water scarcity by improving water productivity in the United States.
882
Environmental Research Letters, 15(9), 094033. https://doi.org/10.1088/1748-
883
9326/ab9d39
884
Marston, L. T., Zipper, S., Smith, S. M., Allen, J. J., Butler, J. J., Gautam, S., & Yu, D. J. (2022).
885
The importance of fit in groundwater self-governance. Environmental Research Letters,
886
17(11), 111001. https://doi.org/10.1088/1748-9326/ac9a5e
887
Marston, L. T., Abdallah, A. M., Bagstad, K. J., Dickson, K., Glynn, P., Larsen, S. G., et al.
888
(2022). Water-Use Data in the United States: Challenges and Future Directions. JAWRA
889
Journal of the American Water Resources Association, 58(4), 485–495.
890
https://doi.org/10.1111/1752-1688.13004
891
Martin, D., Regan, R. S., Haynes, J. V., Read, A. L., Henson, W., Stewart, J. S., et al. (2023).
892
Irrigation water use reanalysis for the 2000-20 period by HUC12, month, and year for the
893
conterminous United States [Data set]. U.S. Geological Survey.
894
https://doi.org/10.5066/P9YWR0OJ
895
Melton, F. S., Johnson, L. F., Lund, C. P., Pierce, L. L., Michaelis, A. R., Hiatt, S. H., et al.
896
(2012). Satellite Irrigation Management Support With the Terrestrial Observation and
897
Prediction System: A Framework for Integration of Satellite and Surface Observations to
898
Support Improvements in Agricultural Water Resource Management. IEEE Journal of
899
Selected Topics in Applied Earth Observations and Remote Sensing, 5(6), 1709–1721.
900
https://doi.org/10.1109/JSTARS.2012.2214474
901
Melton, F. S., Huntington, J., Grimm, R., Herring, J., Hall, M., Rollison, D., et al. (2022).
902
OpenET: Filling a Critical Data Gap in Water Management for the Western United
903
States. JAWRA Journal of the American Water Resources Association, 58(6), 971–994.
904
https://doi.org/10.1111/1752-1688.12956
905
Mourtzinis, S., Rattalino Edreira, J. I., Conley, S. P., & Grassini, P. (2017). From grid to field:
906
Assessing quality of gridded weather data for agricultural applications. European Journal
907
of Agronomy, 82, 163–172. https://doi.org/10.1016/j.eja.2016.10.013
908
Zipper et al. | Irrigation OpenET | 29 of 52
Ott, T. J., Majumdar, S., Huntington, J., Pearson, C., Bromley, M., Minor, B. A., et al. (2024).
909
Toward Sustainable Groundwater Management: Harnessing Remote Sensing and Climate
910
Data to Estimate Field-Scale Groundwater Pumping. ESS Open Archive.
911
https://doi.org/10.22541/essoar.170800918.86740881/v1
912
Pereira, L. S., Paredes, P., Melton, F., Johnson, L., Wang, T., López-Urrea, R., et al. (2020).
913
Prediction of crop coefficients from fraction of ground cover and height. Background and
914
validation using ground and remote sensing data. Agricultural Water Management, 241,
915
106197. https://doi.org/10.1016/j.agwat.2020.106197
916
Pereira, L. S., Paredes, P., & Jovanovic, N. (2020). Soil water balance models for determining
917
crop water and irrigation requirements and irrigation scheduling focusing on the FAO56
918
method and the dual Kc approach. Agricultural Water Management, 241, 106357.
919
https://doi.org/10.1016/j.agwat.2020.106357
920
Peterson, T. J., & Fulton, S. (2019). Joint Estimation of Gross Recharge, Groundwater Usage,
921
and Hydraulic Properties within HydroSight. Groundwater, 57(6), 860–876.
922
https://doi.org/10.1111/gwat.12946
923
Senay, G. B., Friedrichs, M., Morton, C., Parrish, G. E. L., Schauer, M., Khand, K., et al. (2022).
924
Mapping actual evapotranspiration using Landsat for the conterminous United States:
925
Google Earth Engine implementation and assessment of the SSEBop model. Remote
926
Sensing of Environment, 275, 113011. https://doi.org/10.1016/j.rse.2022.113011
927
Steiner, J. L., Devlin, D. L., Perkins, S., Aguilar, J. P., Golden, B., Santos, E. A., & Unruh, M.
928
(2021). Policy, Technology, and Management Options for Water Conservation in the
929
Ogallala Aquifer in Kansas, USA. Water, 13(23), 3406.
930
https://doi.org/10.3390/w13233406
931
Tolley, D., Foglia, L., & Harter, T. (2019). Sensitivity Analysis and Calibration of an Integrated
932
Hydrologic Model in an Irrigated Agricultural Basin With a Groundwater-Dependent
933
Ecosystem. Water Resources Research, 55(9), 7876–7901.
934
https://doi.org/10.1029/2018WR024209
935
USDA. (2022). Cropland Data Layer [Online]. Washington, D.C.: USDA National Agricultural
936
Statistics Service. Retrieved from https://nassgeodata.gmu.edu/CropScape/
937
Vergopolan, N., Chaney, N. W., Pan, M., Sheffield, J., Beck, H. E., Ferguson, C. R., et al.
938
(2021). SMAP-HydroBlocks, a 30-m satellite-based soil moisture dataset for the
939
conterminous US. Scientific Data, 8(1), 264. https://doi.org/10.1038/s41597-021-01050-2
940
Volk, J. M., Huntington, J. L., Melton, F. S., Allen, R., Anderson, M., Fisher, J. B., et al. (2024).
941
Assessing the accuracy of OpenET satellite-based evapotranspiration data to support
942
water resource and land management applications. Nature Water, 1–13.
943
https://doi.org/10.1038/s44221-023-00181-7
944
Wei, S., Xu, T., Niu, G.-Y., & Zeng, R. (2022). Estimating Irrigation Water Consumption Using
945
Machine Learning and Remote Sensing Data in Kansas High Plains. Remote Sensing,
946
14(13), 3004. https://doi.org/10.3390/rs14133004
947
Whittemore, D. O., Butler, J. J., Bohling, G. C., & Wilson, B. B. (2023). Are we saving water?
948
Simple methods for assessing the effectiveness of groundwater conservation measures.
949
Agricultural Water Management, 287, 108408.
950
https://doi.org/10.1016/j.agwat.2023.108408
951
Wilson, B. B., Liu, G., Bohling, G. C., & Butler, J. J. (2021). GMD4 Groundwater Flow Model:
952
High Plains Aquifer Modeling Maintenance Project (KGS Open File Report 2021-6).
953
Lawrence KS.
954
Zipper et al. | Irrigation OpenET | 30 of 52
Xu, T., Deines, J. M., Kendall, A. D., Basso, B., & Hyndman, D. W. (2019). Addressing
955
Challenges for Mapping Irrigated Fields in Subhumid Temperate Regions by Integrating
956
Remote Sensing and Hydroclimatic Data. Remote Sensing, 11(3), 370.
957
https://doi.org/10.3390/rs11030370
958
Zipper, S. C., Stack Whitney, K., Deines, J. M., Befus, K. M., Bhatia, U., Albers, S. J., et al.
959
(2019). Balancing Open Science and Data Privacy in the Water Sciences. Water
960
Resources Research, 55(7), 5202–5211. https://doi.org/10.1029/2019WR025080
961
Zipper, S. C., Carah, J. K., Dillis, C., Gleeson, T., Kerr, B., Rohde, M. M., et al. (2019).
962
Cannabis and residential groundwater pumping impacts on streamflow and ecosystems in
963
Northern California. Environmental Research Communications, 1(12), 125005.
964
https://doi.org/10.1088/2515-7620/ab534d
965
Zipper, S. C., Gleeson, T., Li, Q., & Kerr, B. (2021). Comparing Streamflow Depletion
966
Estimation Approaches in a Heavily Stressed, Conjunctively Managed Aquifer. Water
967
Resources Research, 57(2), e2020WR027591. https://doi.org/10.1029/2020WR027591
968
Zipper, S. C., Farmer, W. H., Brookfield, A., Ajami, H., Reeves, H. W., Wardropper, C., et al.
969
(2022). Quantifying Streamflow Depletion from Groundwater Pumping: A Practical
970
Review of Past and Emerging Approaches for Water Management. JAWRA Journal of
971
the American Water Resources Association, 58(2), 289–312.
972
https://doi.org/10.1111/1752-1688.12998
973
974
Zipper et al. | Irrigation OpenET | 31 of 52
Supplemental material for:
975
Estimating irrigation water use from remotely sensed evapotranspiration
976
data: Accuracy and uncertainties across spatial scales
977
978
Authors: Sam Zipper1,2,*, Jude Kastens3, Timothy Foster4, Brownie Wilson1, Forrest Melton5,
979
Ashley Grinstead1,6, Jillian M. Deines7, James J. Butler1, Landon T. Marston8
980
981
Affiliations:
982
1. Kansas Geological Survey, University of Kansas, Lawrence KS 66047
983
2. Department of Geology, University of Kansas, Lawrence KS 66045
984
3. Kansas Biological Survey & Center for Ecological Research, University of Kansas,
985
Lawrence KS 66047
986
4. School of Engineering, University of Manchester, Manchester, UK
987
5. Atmospheric Science Branch, Earth Science Division, NASA Ames Research Center,
988
Moffett Field, CA 94035
989
6. Department of Natural Resources and the Environment, University of Connecticut,
990
Storrs, CT 06269, United States
991
7. Earth Systems Predictability and Resiliency Group, Pacific Northwest National
992
Laboratory, Richland, WA 99354
993
8. Department of Civil and Environmental Engineering, Virginia Tech, Blacksburg VA
994
24061
995
*Correspondence to samzipper@ku.edu
996
997
Zipper et al. | Irrigation OpenET | 32 of 52
998
Section SI1. Additional information about the study area.
999
1000
1001
Figure S1. Annual area within LEMA for each land cover for all fields (left) and irrigated fields
1002
(right), from the USDA Cropland Data Layer. Irrigated fields were identified using AIM
1003
(Deines, Kendall, Crowley, et al., 2019).
1004
1005
1006
Figure S2. Annual irrigated area within the SD-6 LEMA from 2006 to 2021, based on AIM
1007
(Deines, Kendall, Crowley, et al., 2019). The dashed line shows the onset of the SD-6 LEMA.
1008
1009
Zipper et al. | Irrigation OpenET | 33 of 52
Section SI2. Additional plots related to irrigation calculations in the SD-6 LEMA area.
1010
1011
Table S1. Fit statistics for precipitation-corrected calculated irrigation shown in Figure 4b.
1012
Model
MAE [x107 m3]
Bias [%]
Slope
R2
Ensemble
0.093
0
1.01
0.95
DisALEXI
0.341
0
0.56
0.46
eeMETRIC
0.169
0
0.97
0.85
geeSEBAL
0.184
0
0.85
0.78
PT-JPL
0.122
0
0.96
0.90
SIMS
0.100
0
1.02
0.95
SSEBop
0.132
0
0.96
0.89
1013
1014
Zipper et al. | Irrigation OpenET | 34 of 52
1015
Figure S3. Characterization of confidence in irrigation classification and field boundary by year
1016
for the SD-6 LEMA area. The y-axis shows the total area in each of the 4 bins. The “High
1017
Confidence Rainfed” and “High Confidence Irrigated” fields have <10% and >90% of pixels
1018
within the field boundary mapped as irrigated by AIM, respectively. The “Low Confidence
1019
Rainfed” and “Low Confidence Irrigated” are classified as rainfed and irrigated, respectively, but
1020
have a larger >10% of the pixels within the field mapped as the opposite practice. Across all
1021
years, there is more Low Confidence Irrigated land than Low Confidence Rainfed land,
1022
suggesting that any errors in irrigation classification are likely to bias irrigated area high relative
1023
to reported data.
1024
1025
Zipper et al. | Irrigation OpenET | 35 of 52
1026
Figure S4. Map showing spatial distribution of irrigated classification and field boundary
1027
confidence data shown in Figure S3.
1028
1029
Zipper et al. | Irrigation OpenET | 36 of 52
1030
Figure S5. Maps of estimated field-resolution ET - effective precipitation for each year and
1031
model. The black outline corresponds to the SD-6 LEMA.
1032
1033
Zipper et al. | Irrigation OpenET | 37 of 52
1034
Figure S6. Maps of estimated field-resolution irrigation for each year and model. Fields that are
1035
classified as non-irrigated are not shown. The black outline corresponds to the SD-6 LEMA.
1036
1037
Zipper et al. | Irrigation OpenET | 38 of 52
1038
Figure S7. Density plots of field-resolution ET for irrigated fields in the SD-6 LEMA by year.
1039
1040
1041
Figure S8. Summary of fit statistics for comparison between each OpenET model and WIMAS
1042
data for total LEMA-scale pumping (i.e., data from Table 1 in graphical form).
1043
1044
Zipper et al. | Irrigation OpenET | 39 of 52
Section SI3. Additional plots related to effective precipitation use in irrigation calculations
1045
1046
1047
Figure S9. Relationship between deep percolation (from 2013-2017 simulated data from Deines
1048
et al., 2021) and calendar year, growing season, and water year total precipitation. Blue line
1049
shows linear best fit relationship and standard error in each plot.
1050
1051
Zipper et al. | Irrigation OpenET | 40 of 52
1052
Figure S10. Same as Figure 3, but using precipitation in irrigation calculations instead of
1053
effective precipitation.
1054
1055
Zipper et al. | Irrigation OpenET | 41 of 52
1056
Figure S11. Same as Figure S8, but using precipitation in irrigation calculations instead of
1057
effective precipitation.
1058
1059
Zipper et al. | Irrigation OpenET | 42 of 52
1060
Figure S12. Same as Figure 4, but using precipitation in irrigation calculations instead of
1061
effective precipitation.
1062
1063
Zipper et al. | Irrigation OpenET | 43 of 52
1064
Figure S13. Same as Figure S5, but using total growing season precipitation instead of effective
1065
precipitation.
1066
1067
Zipper et al. | Irrigation OpenET | 44 of 52
1068
Figure S14. Same as Figure S6, but irrigation calculated using total growing season precipitation
1069
instead of effective precipitation.
1070
1071
Zipper et al. | Irrigation OpenET | 45 of 52
1072
Figure S15. Same as Figure 5, but using precipitation in irrigation calculations instead of
1073
effective precipitation.
1074
1075
1076
1077
Zipper et al. | Irrigation OpenET | 46 of 52
1078
Figure S16. Same as Figure 7, but using precipitation in irrigation calculations instead of
1079
effective precipitation.
1080
1081
Table S2. Same as Table 2, but using precipitation in irrigation calculations instead of effective
1082
precipitation.
1083
MAE [mm]
Bias [%]
Slope
R2
Model
Annual
Multi-Year
Annual
Multi-Year
Annual
Multi-Year
Annual
Multi-Year
DisALEXI
159
87
-2.1
-3.7
0.36
0.86
0.15
0.40
eeMETRIC
209
140
36.2
33.6
0.28
0.69
0.15
0.45
Ensemble
154
85
0.1
-1.5
0.40
0.91
0.18
0.44
geeSEBAL
194
169
-44.3
-45.3
0.47
0.97
0.20
0.44
PT-JPL
160
99
-17.6
-19.2
0.37
0.68
0.11
0.21
SIMS
237
204
55.5
52.2
0.23
0.54
0.07
0.16
SSEBop
158
62
0.2
-1.3
0.39
1.04
0.21
0.61
Average
181
121
4.0
2.1
0.36
0.81
0.15
0.39
1084
Zipper et al. | Irrigation OpenET | 47 of 52
1085
Figure S17. Same as Figure 8, but using precipitation instead of effective precipitation.
1086
1087
Zipper et al. | Irrigation OpenET | 48 of 52
1088
Figure S18. Same as Figure 9, but using precipitation instead of effective precipitation.
1089
1090
1091
Zipper et al. | Irrigation OpenET | 49 of 52
Section SI4. Comparison between radar and gridMET precipitation data.
1092
1093
1094
Figure S19. Comparison between gridMET- and National Weather Service Advanced
1095
Hydrologic Prediction Service radar precipitation data for all fields within the SD-6 LEMA. ET-
1096
based irrigation calculations use the growing season as the timescale of aggregation.
1097
1098
Zipper et al. | Irrigation OpenET | 50 of 52
1099
Figure S20. Comparison between LEMA total irrigation estimated using gridMET data and radar
1100
precipitation data for each model and year. ET-based irrigation calculations use the growing
1101
season as the timescale of aggregation. The 1-1 line is included in each plot.
1102
1103
1104
Figure S21. Comparison to reported irrigation volumes for estimated irrigation using OpenET
1105
data and radar precipitation for the entire SD-6 LEMA.
1106
1107
Zipper et al. | Irrigation OpenET | 51 of 52
1108
Figure S22. Comparison to reported irrigation volumes for estimated irrigation using OpenET
1109
data and (top) gridMET precipitation and (bottom) radar precipitation for the entire SD-6 LEMA.
1110
ET-based irrigation calculations use the growing season as the timescale of aggregation.
1111
1112
Zipper et al. | Irrigation OpenET | 52 of 52
Section SI5. Additional figure related to WRG-scale WIMAS-OpenET comparison.
1113
1114
1115
Figure S23. Same as Figure 5, but only for WRGs where reported and calculated irrigated area
1116
agreed within 10% (i.e., orange points in Figure 6). Since there were relatively few WRGs with
1117
good irrigated area agreement within the SD-6 LEMA, we also included WRGs with agreement
1118
within 10% from a buffer area surrounding the SD-6 LEMA. Each panel shows: (a) Annual
1119
irrigation volume for each WRG; (b) Annual irrigation depth for each WRG; (c) Average
1120
irrigation volume for each WRG; (d) Average irrigation depth for each WRG. In each plot, the
1121
gray line shows a 1:1 agreement between reported and calculated irrigation.
1122
1123
Section SI6. References in supplemental material
1124
Deines, J. M., Kendall, A. D., Crowley, M. A., Rapp, J., Cardille, J. A., & Hyndman, D. W.
1125
(2019). Mapping three decades of annual irrigation across the US High Plains Aquifer
1126
using Landsat and Google Earth Engine. Remote Sensing of Environment, 233, 111400.
1127
https://doi.org/10.1016/j.rse.2019.111400
1128
Deines, J. M., Kendall, A. D., Butler, J. J., Basso, B., & Hyndman, D. W. (2021). Combining
1129
Remote Sensing and Crop Models to Assess the Sustainability of Stakeholder-Driven
1130
Groundwater Management in the US High Plains Aquifer. Water Resources Research,
1131
e2020WR027756. https://doi.org/10.1029/2020WR027756
1132
