Evaluation of hydrologic impact of an irrigation curtailment program using Landsat satellite data

Upper Klamath Lake (UKL) is the source of the Klamath River that flows through southern Oregon and northern California. The UKL Basin provides water for 81,000+ ha (200,000+ acres) of irrigation on the U.S. Bureau of Reclamation Klamath Project located downstream of the UKL Basin. Irrigated agriculture also occurs along the tributaries to UKL. During 2013–2016, water rights calls resulted in various levels of curtailment of irrigation diversions from the tributaries to UKL. However, information on the extent of curtailment, how much irrigation water was saved, and its impact on the UKL is unknown. In this study, we combined Landsat‐based actual evapotranspiration (ETa) data obtained from the Operational Simplified Surface Energy Balance model with gridded precipitation and U.S. Geological Survey station discharge data to evaluate the hydrologic impact of the curtailment program. Analysis was performed for 2004, 2006, 2008–2010 (base years), and 2013–2016 (target years) over irrigated areas above UKL. Our results indicate that the savings from the curtailment program over the June to September time period were highest during 2013 and declined in each of the following years. The total on‐field water savings was approximately 60 hm3 in 2013 and 2014, 44 hm3 in 2015, and 32 hm3 in 2016 (1 hm3 = 10,000 m3 or 810.7 ac‐ft). The instream water flow changes or extra water available were 92, 68, 45, and 26 hm3, respectively, for 2013, 2014, 2015, and 2016. Highest water savings came from pasture and wetlands. Alfalfa showed the most decline in water use among grain crops. The resulting extra water available from the curtailment contributed to a maximum of 19% of the lake inflows and 50% of the lake volume. The Landsat‐based ETa and other remote sensing datasets used in this study can be used to monitor crop water use at the irrigation district scale and to quantify water savings as a result of land‐water management changes.

K E Y W O R D S agriculture, evapotranspiration, irrigation curtailment, Landsat, satellite data, SSEBop model, upper Klamath Lake, water saving

| INTRODUCTION
The increasing pressure to meet human demand for food for continuously growing populations is depleting water resources and leading to global water scarcity. It is estimated that over two-thirds of the global population is currently suffering from lack of water at least 1 month out of the year (Mekonnen & Hoekstra, 2016), and up to 700 million people could be displaced by intense freshwater scarcity by 2030 (Hameeteman, 2013). Agriculture consumes up to 70% of global human water use. Within the agricultural sector, water used for irrigation accounts for nearly 65% of the world's freshwater withdrawals excluding thermoelectric power (Hutson et al., 2004). Over the last several decades, increases in global food production were to a large extent possible due to (a) doubling of the irrigation area, (b) corresponding increases of water withdrawal for irrigation, and (c) increased usage of fertilizer (Oki & Kanae, 2006). The increase in water use for irrigation is depleting surface water resources (Rodell et al., 2018;William Darwall William Darwall et al., 2018), depleting groundwater resources (Wada et al., 2010), reducing water quality (Gallardo et al., 2018), and impacting ecosystem biodiversity (Clausen & York, 2008;Falkenmark, 2003;Gallardo et al., 2018;Garcia-Moreno et al., 2014).
Within the United States, agriculture is a major consumer of groundwater and surface water. About 80% of the Nation's consumptive water use (> 90% in the western United States) is attributed to crop irrigation.
The arid/semiarid climate of the western United States with low rainfall distribution has resulted in increased pressure for freshwater for use in irrigated agriculture. Recently, water management programs have been implemented to reduce agricultural water use in the western United States. These programs are often supported by the government and water management authorities, which are implementing policies to promote water conservation rather than water use (Peck, 2015). A well-studied example is the crop fallowing program in the Palo Verde Irrigation District (PVID) on the border between California and Arizona, where a water management agreement between the PVID and the Metropolitan Water District of Southern California left a fraction of PVID fields intentionally fallow, so water could be transferred to meet domestic water demand in municipal areas (Senay, Schauer, Friedrichs, Velpuri, & Singh, 2017). Similarly, a farmland fallowing and forbearance project in Arizona was implemented in 2014 to increase water in the Colorado River system by saving about 25 to 74 cubic hectometers (hm 3 , 1 hm 3 = 10,000 m 3 or 810.7 ac-ft) of water and improving Lake Mead water levels (https://wrrc. arizona.edu/arizona-land-fallowing, last accessed: Nov 12, 2019). Other examples of irrigation water curtailment include reducing water permits for extraction of surface and groundwater, termination of water rights, or economic payouts to compensate the forbearance of water rights.
This study focuses on quantifying the effects of curtailment of irrigation diversions in the Upper Klamath Lake (UKL) Basin in southern Oregon. Since 2013, when water claims were determined in the basin, water users with senior water rights (who do not receive the water they are entitled to) may request that state water regulators shut off or curtail diversions to junior water rights users. Regulators and water users need to understand the effect of curtailing diversions, specifically how curtailment affects streamflow and inflows to UKL. The impact of the curtailment in the UKL Basin and quantifying the additional water that remained in the stream from curtailment were first studied by Hess and Stonewall (2014), where they compared the measured U.S. Geological Survey (USGS) 2013 streamflow data with curtailments with those data from hydrologically similar years without curtailments. That study found that streamflow from the Williamson and Wood Rivers to UKL increased by approximately 17 hm 3 (14,100 ac-ft) and 7 hm 3 (5,500 acft), respectively, between July and September. However, the Hess and Stonewall study was limited to analysis of instream river flows only in 2013. Furthermore, information on temporal and spatial variability in the water savings (2013-2016) across the UKL Basin and from which crop types/landscapes savings originated were not addressed.
The goal of this current study was to use Landsat-derived ETa to quantify the hydrologic impact of the 2013-2016 curtailment program. The primary objectives of this study were to (a) quantify the onfield water savings and how these water savings affected streamflow due to irrigation curtailment, (b) understand the spatial and temporal variability in water savings, (c) estimate and quantify water savings by crop type during the curtailment period, and (d) understand the impact of water savings on the UKL Basin water budget.
2 | STUDY AREA AND DATA USED

| Study area
The UKL Basin is on the eastern slopes of the Cascade mountains in southern Oregon and is characterized by low rainfall and semiarid conditions. UKL is the largest freshwater body in the state of Oregon. It is the source of the Klamath River that flows through southern Oregon and northern California and is a crucial source of water for irrigation up to 202,000 ha (~500,000 ac) including 81,000+ ha (200,000+ ac) of the U.S. Bureau of Reclamation irrigation project located downstream of the lake (Marshall, Robles, Majka, & Haney, 2010). UKL is a shallow lake with an average depth of 2 m and an average volume of approximately 546 hm 3 (Walker, Walker, & Kann, 2012). Hubbard (1970) provided a detailed account of the UKL primary water budget components. Principal inflows into UKL are contributed by streams, irrigation canals, agricultural drainage, springs, seeps, and direct precipitation on the lake.
The Williamson and the Wood Rivers contribute nearly 49% and 16% of the UKL inflows, respectively. These two rivers together contribute more than 80% (66% of the total 80% surface inflows) of the surface inflows to the lake. The streams that drain from the Cascade ranges on the west side of the lake (Cherry, Sevenmile, Fourmile, Threemile Creeks, and Central Canal) together contribute up to 8% of the lake inflows. Other minor creeks, such as Rock, Varney, Moss, and Denny, contribute up to another 2% of the lake inflows. It is reported that about 4% of the total inflows comes from agricultural drainage, and the remaining inflows (up to 14%) are contributed by springs and seeps around the lake; precipitation directly contributes to the remaining 7% of inflows to the lake (Hubbard, 1970). Because it is hard to delineate watersheds for these small to minor creeks at the scale presented in Figure 1, we combined their watersheds with the Wood River basin.

| Data used
The top-of-atmosphere reflectance Landsat data (Collection 1) were obtained from https://earthexplorer.usgs.gov (last accessed: Nov 12, 2019). In this study, a total of 1,066 Landsat 5/7/8 images with 60% cloud cover or less from paths 44-45, rows 30-31, which covered the basin, were processed for the base (2004, 2006, and 2008-2010) and target years (2013)(2014)(2015)(2016). Base and target years were selected based on recommendations from the USGS Oregon Water Science Center. The number of Landsat imagery and sensor types used in this study is presented in Table 1. The Landsat quality assessment band was used to flag and mask out clouds, cloud shadows, and Landsat 7 scan-line errors, and these pixels were gapfilled with simple linear interpolation using Landsat images in a 48-day window (Senay et al., 2019).
Landsat-based ETa data used in this study were modelled using the Operational Simplified Surface Energy Balance (SSEBop) algorithm  that uses model-assimilated weather datasets and Landsat thermal data to produce ETa for the UKL Basin. Monthly total precipitation for the study area over the base years and target years was summarized from 4-km monthly Parameter-elevation Regressions on Independent Slopes Model (PRISM) precipitation datasets (Daly et al., 2000;Daly, Smith, & McKane, 2007)

| Modeling ET data using SSEBop model
Landsat ET is modelled using the SSEBop approach, which leverages the thermal band from Landsat along with weather datasets to generate a daily total of ETa . The Landsat thermal band is converted into land surface temperature using emissivity derived from the normalized difference vegetation index. The primary product of the SSEBop model is the daily ET fraction, which is an index between 0 (dry, bare surface with no ET) to 1 (wet, green, irrigated land with maximum ET). The ET fraction is then used to reduce the grass-reference potential ET dataset derived from GridMET (Abatzoglou, 2013). GridMET is a gridded 4-km dataset of daily weather variables for the continental United States provided from 1979 to the present and available from the University of Idaho http://www.climatologylab.org/gridmet.html.
SSEBop actual ET can be summarized with the following equation: where ETa is the daily total actual ET in millimetre for the Landsat image; γ s is the surface psychrometric constant (K) derived from a radiation budget; T s is the land surface temperature derived from the Landsat thermal band (K); T c is cold/wet reference limit, derived from the daily maximum air temperature (K); k is a scaling factor (fixed at 1.25) to convert grass reference to alfalfa reference; and ETo is the gridded grass-reference (short crop) potential ET from GridMET (Abatzoglou, 2013). Equation (1)  T A B L E 1 Number of Landsat imagery and sensor types used for the baseline and target years Landsat sensor Target years   2004  2006  2008  2009  2010  2013  2014  2015  2016  Total   Landsat 5 TIRS  68  67  34  73  39  281   Landsat 7 TIRS  59  60  44  66  39  67  64  75  68  542   Landsat 8 TIRS  54  58  69  62  243   Total  127  127  78  139  78  121  122  144  130 1,066

Baseline years
Note: Landsat 5 data are not available over target years, and Landsat 8 data were not available for the baseline years. Abbreviation: TIRS, thermal infrared sensor.

| Quantifying the on-field water savings
The modelled seasonal SSEBop ETa provides a direct measure of onfield water use per pixel over base years (2004, 2006, and 2008-2010) and target years (2013)(2014)(2015)(2016). As the curtailment in irrigation is expected to reduce the Landsat ETa, any reduction in ETa over target years with respect to the base years would indicate water savings. Hence, using ETa, we computed water savings (the deviation in ETa or ΔETa) for each target year with respect to the base year as where i ranges from target years 2013-2016, and j ranges from base years (2004, 2006, and 2008-2010). Based on Equation (2), we get a set of 5 values for each target year. Mean ΔETa (ΔETa mean ) was also computed for each target year using the mean of all the base years as shown below: where ETa BY mean is the mean estimate of SSEBop ETa for all the base years.

| Computation of extra water available
The deviation in ETa does not necessarily reflect the extra water available (EWA) in the streams resulting from the curtailment because any changes in ETa over the target year can be partially attributed to the differences in precipitation between target and base years. Hence, to compensate the impact of precipitation on the ΔETa estimates, we used gridded estimates of precipitation over base year and target year to compute the deviation in precipitation (ΔPPT) and average deviation in precipitation (ΔPPT mean ) between base and target years as shown below. and EWA for each of the target year (with respect to the base year [s]) and average EWA (with respect to the average of the base years) are then computed as The parameters ΔETa and EWA are computed per pixel and are summarized as polygon averages over irrigated areas in a river basin. The data are also presented for the river basin irrigation (Sprague River, Williamson River, Wood River, and Upper Klamath Agency Lakes irrigation) and irrigation type (surface water, groundwater, and conjunctive use). 3.6 | Impact of water savings on Upper Klamath Lake water budget Curtailment of surface water should result in more flow in streams and inflow to UKL than if curtailment did not occur. Hence, the impact of savings from the irrigation curtailment on the UKL was investigated by comparing the savings computed from Equations 6 and 7 with surface inflows (Q s_in ), surface outflows (Q s_out ),and lake volume (L v ). We computed lake inflows (Q s_in ) as shown below:

| Seasonal SSEBop ETa estimates
Basin-wide seasonal SSEBop ETa estimates were produced for the June-September months corresponding to the irrigation curtailment during the peak growing season. Figure 2 shows the total seasonal ETa for the base and target years. In the figure, the areas of low ETa < 100 mm (in yellow) are concentrated in the northwestern and eastern regions of the basin. The areas of high ETa > 800 mm (in dark blue) represent water bodies or marsh lands. During June to September, the UKL Basin was mostly dry as it received average rainfall of about 50-60 mm. However, the seasonal SSEBop average ETa was roughly four times higher than the rainfall received (up to 260 mm) during June-September.

| Validation results
Validation was carried out at two scales. First, we compared daily summaries of ETa obtained from the eddy covariance tower sites with overpass ETa modelled using SSEBop model. Comparison between F I G U R E 3 Validation of overpass Operational Simplified Surface Energy Balance actual evapotranspiration data (in red) against daily flux tower data (in grey) from a) Bulrush site (top panel) and b) Mixed site (bottom panel) F I G U R E 2 Operational Simplified Surface Energy Balance seasonal (June-September) actual evapotranspiration totals (mm) for baseline (2004, 2006, and 2008-2010) and target years (2013-2016) the estimated daily ETa and measured ETa showed decent correlation at both sites (Pearson's r, Figure 3). SSEBop ETa was found to capture the temporal variability seen at both the sites.

| Basin-scale ETa and ΔETa
The deviation in ETa (ΔETa), or water savings for each target year, was computed using Equations 2 and 3. Figure 5 shows the compari-  To further understand the changes in on-field irrigation water savings (defined as water savings over the irrigated areas) during target years, we plotted histograms of the images corresponding to the irrigated area pixels. Figure 7 shows histograms of the irrigated area pixels

| Variability in area under irrigation water savings
To quantify the water savings from irrigated lands during the target years, we first computed histograms from the ΔETa images for all irri- 4.6 | Summary of on-field water savings (ΔETa) The savings in irrigation due to a curtailment estimated using Equation 2 are presented in the top row of Figure 9 for Sprague, Upper Klamath Agency Lakes, Williamson, and Wood River irrigation regions.
Estimates of irrigation water savings were also computed for different irrigation types (surface water, groundwater, and conjunctive use) and are shown in the bottom row of Figure 9. Results indicate that estimates of water savings showed an overall decreasing trend, but for  Figure 10 presents the EWA estimated over target years using

| Summary of EWA
Equation 6 for each irrigation region and type. Unlike irrigation water savings, the year to year variability in EWA is unidirectional as EWA gradually reduced over the irrigation curtailment period.
Out of four irrigation regions, the Williamson River Basin irrigation area showed the highest interquartile range in EWA, whereas the Wood River irrigation area showed the lowest. Similarly, surface water irrigation area showed the highest interquartile range in EWA, whereas groundwater irrigation area showed the lowest. The EWA estimates from all irrigation (sum of EWA from four irrigation regions is equal to the sum of EWA from the three irrigation types) indicate a similar declining trend from 2013 to 2016. Of all the irrigation areas, the Sprague River Basin irrigation showed the highest EWA, 48 hm 3 (38,877 ac-ft) of water in 2013 and declined to 18 hm 3 (14,536 ac-ft) of irrigation water savings in 2016 ( Table 2).
The next highest EWA was in the Williamson River irrigation area with 31 hm 3 (24,988 ac-ft) of EWA in 2013, which declined to 9 hm 3 (7,027 ac-ft) of EWA in 2016. The Wood River irrigation area showed EWA up to 13 hm 3 (10,286 ac-ft) in 2013 but then declined to nearly negligible EWA (40 ac-ft) in 2016. The total EWA for 2013 was about 92 hm 3 (74,500 ac-ft); however, the EWA reduced dramatically to 26 hm 3 (21,421 ac-ft) in 2016.

| Crop type water savings
We used USDA-NASS CDL data to understand crop type water savings. The results in Table 3 Table 3 indicate no savings (i.e., more water was consumed by these crop types during the target years). For example, other hay crop types did not show any savings over the target years; instead, their crop water use increased. The four non-grain crop classes account for more than 50% of the total irrigated area because the classification of areas as irrigated for this study was guided by the water rights and included areas such as grassland, pastures, and herbaceous wetlands, which are widely used for grazing cattle.
F I G U R E 1 0 Box plots showing the variability in extra water available (EWA) over the target years for Sprague, Upper Klamath agency lakes, Williamson, and Wood river irrigation. The mean of the distribution is represented by the horizontal black line, the box indicates upper 75th and lower 25th percentiles, and the whiskers above and below the boxes indicate the maximum and the minimum estimates of EWA. Absence of whiskers for Williamson river irrigation indicates that the minimum and maximum estimates of EWA are well above the 1.5 times the 25th and 75th percentile estimates 4.9 | Impact of water savings and EWA on UKL water levels Water savings from the curtailment are assumed to add to streamflow and result in higher inflows into the UKL Basin. Hence, we obtained river inflows from Equation 8 and computed the percentage of irrigation water savings and EWA with respect to lake surface inflows, lake outflows, and average lake volume. Table 4 indicates that the savings generated during the curtailment program contributed to a maximum of 19% of the lake inflows, 27% of the lake outflows, and 50% of the lake volume.

| DISCUSSION
The possibility of using remotely sensed data for routinely monitoring agricultural water use is a priority for water managers and planners around the world (Senay et al., 2019). This study presents an application of Landsat-based ETa for solving real-world problems. The benefit of using remote sensing is its capability to produce spatially and temporally explicit information on agricultural water use. This study provides insights into the hydrologic impacts of an irrigation curtailment in the UKL Basin. The value of remote sensing estimates often depends on the accuracy and uncertainty of the data and models used. In this study, the errors in the estimation of water savings from the curtailment program arise from (a) error in the ETa estimates and (b) error in the precipitation estimates. The accuracy of seasonal ETa was quantified to have uncertainty up to 10-15% at basin scale . However, validation of seasonal ETa estimates using two eddy covariance sites used in this study indicates an underestimation up to −16%. Uncertainties in the PRISM rainfall data used in this study are thoroughly investigated by (Daly et al., 2008) who found that uncertainties reduce considerably when averaged over spatial domains with R 2 as high as .85 (with very low bias error approximately ± 0-5%) when compared with other gridded rainfall datasets. Based on these estimates, we expect a bias in the range of ±0-22% in our estimates for absolute magnitudes, but water savings are based on relative differences from base years and thus less affected by bias errors.
In this study, we estimated ETa over irrigated areas within the UKL Basin and computed the ΔETa in the target years compared with the ΔETa from the base years. The ΔETa represents direct savings in water use over the fields that are irrigated. To normalize the impact of precipitation on changes in ETa, we computed the EWA parameter, which is more important in predicting the flow in streams; however, the direct impact of curtailment may be better quantified by a decrease in ETa alone, as the EWA includes assumptions on the transport of excess precipitation/irrigation to the river system. Generally, some years have more precipitation, which could increase ETa over T A B L E 4 The impact of water savings and EWA on the Upper Klamath Lake summarized using lake inflows, lake outflows, and mean lake volume
Thus, the water savings in 2013 and 2014 are comparable in terms of ETa, but EWA in 2013 was much higher than EWA in 2014. One way of interpreting the 2013 and 2014 ETa and EWA disparity is that although fallow lands used much of the increased precipitation in 2013, the crop water demand on irrigated areas was partially met by the precipitation, and thus, farmers did not need to apply as much water on the fields or excess precipitation/irrigation found its way to the river system, thus increasing the streamflow (higher EWA).
Irrigation water savings could be a result of (a) the direct curtailment of irrigation, (b) changes in the water management practices, (c) changes in crop type, (d) changes in the agronomic practices such as irrigation scheduling, or (e) a combination of all. The water savings estimated in this study (Table 2) quantifies the total savings and does not provide the information on the actual cause of the savings. If we assume minimal change in management and agronomic practices between base and target years, then we can conclude that most of the savings are attributable to the curtailment program.
The estimation of impact of the curtailment on the lake inflows, outflows, and lake volume (Table 4) is illustrated using the data available at the time of the study. We used pre-curtailment data (2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012) for the estimation of inflows from the Williamson River. However, because pre-curtailment data for Wood River inflows were not available, data from the post-curtailment period (2014-2018) were used, which could report a higher flow due to reduced irrigation. Hence, the real impact of curtailment on total lake inflows may be slightly underestimated. Similarly, the surface outflow estimated using data obtained from the Link River at Klamath Falls constitutes about 85% of the outflow from the lake (Hubbard, 1970). Direct pump irrigation from the lake and direct evaporation losses from the lake were not considered.
Since the irrigation curtailment is scheduled mostly during the June-September months, the estimates of EWA correspond to the seasonal savings because EWA is calculated using June-September precipitation and ETa variability only. However, the antecedent soil moisture and the end of season precipitation are not accounted for in the EWA estimation. Because the basin is arid/semiarid, the monthly ETa is higher than precipitation for most months. Hence, the impact of antecedent rainfall would be low at the beginning of the cropping season. Also, high precipitation occurring towards the end of the season may not be completely captured in the seasonal ETa estimate.
Generally, off-season antecedent soil moisture and end of season precipitation equate to about 10-15% of the seasonal ETa. Hence, the direct use of seasonal EWA (June-September) would be different from annual (water year) EWA estimates. Moreover, the impact of the off-season (October-May) hydrologic regime outside of the irrigated areas and change in storage should be considered to accurately predict basin-scale EWA estimates. Further studies are needed to understand (a) the water budget dynamics of the off season and its impact during the curtailment, (b) impact of seasonal curtailment on the offseason hydrologic regime, and c) the overall impact of the curtailment on the ecohydrology of the basin.

| CONCLUSIONS
The goal of this study was to use Landsat-derived ETa to quantify the hydrologic impact of the 2013-2016 curtailment program in the Upper Klamath Lake Basin. The following four objectives were addressed to meet this goal: (1) We quantified the on-field water savings by comparing remotely sensed ETa during base (2004, 2006, and 2008-2010) and target years (2013)(2014)(2015)(2016). The study results show that the target (curtailment) years had lower ETa than the base years, confirming the implementation of the curtailment program and demonstrating the capability of remote sensing in monitoring and assessing water management programs in a spatially explicit manner.
Precipitation differences between base and target years indicated that the target years were slightly wetter than the base years. A small increase in precipitation may have resulted in increased flows (EWA) during the target years.
(2) We analysed satellite-based ETa and precipitation data to understand the spatial variability in water savings.
Both the ΔETa savings and EWA showed that the curtailment pro-