The annual ETo could only explain 4.5% of the variation in annual evapotranspiration rate among GSAs

We estimated a total of 19.9 and 21.8 tera-liter of water consumption via evapotranspiration over the agricultural land in California’s Central Valley in 2014 and 2016 water years , respectively . In the water year 2014, the top eight crop types accounted for 75% of total crop consumptive water use in the valley, including almond , rice , grapes , alfalfa , corn , walnuts , pistachios , and tomatoes . Overall, rice was the second largest water consumer after almonds, although it only used 7.6% of cropland, due to its highest annual evapotranspiration rate of 1,109 mm yr−1, on a per unit area basis . Pasture, walnut, almonds, citrus, and alfalfa also had relatively high evapotranspiration rate  . In contrast, wheat consumed the least amount of water per area among major crops, whereas pistachio, tomato, corn, cotton, and grapes had a moderate evapotranspiration rate between 600 and 800 mm yr−1. As the second and third largest cropland use in the valley, grapes, and corn used less total water than rice but similar water with alfalfa, walnuts, and fruits.Across the valley, the mean annual evapotranspiration rate varied by 35% over all agricultural pixels in 2014, mostly due to the diversity of the crop types. We also found high ariability of evapotranspiration rate within each crop type, especially over orchards such as almond, pistachios, and walnut, with a CV higher than 20% , most likely due to differences in planting density, age, canopy structures, and stressors among orchards . For example, the almond evapotranspiration rate varied by 34% , and the rate for pistachio varied by 59% in 2014. Wheat also had a very high variability , different from other annual crops, which typically had a much lower variation of evapotranspiration rate than perennial crops. For all major crop categories, the difference in CV between 2014 and 2016 was <7.2%. Compared to 2014, total crop consumptive water use increased by 9.6% in 2016 , with an evapotranspiration rate of 856 mm yr−1,containers size for raspberries although the reference evapotranspiration from Spatial-CIMIS decreased by 4%. This increase in evapotranspiration was mostly caused by land-use changes with higher irrigated areas and crops with higher averaged water consumptive use .

Total irrigated agriculture land use increased by 7.0% in 2016, partly due to a 2,370 km2 land-use conversion from fallow/ idle lands in 2014 to cropland in 2016. A large portion of fallow land conversion grew rice , wheat , and perennial crops in 2016, leading to an increase of total water use by 1.3 tera-liters. Another major land-use change was the conversion from annual crops to high water demand orchards, including almonds, walnuts, citrus, or grape in 2016, accounting for 1.5% of 2016 cropland and decreasing water use by 0.07 tera-liters due to the low evapotranspiration rate of young orchards.Variability of evapotranspiration rate among GSAs was primarily driven by non-meteorological drivers. Across GSAs, we found that the evapotranspiration rate highly correlated with net radiation and actual Priestley-Taylor coefficient . Many of these factors were regulated by land-use types, vegetation cover, and plant water stress status. In contrast, EToF was the dominant driver of evapotranspiration rate variability among GSAs , mostly driven by crop types, e.g., rice with EToF of 0.61 , tomato 0.33 , almond 0.52 , and pistachio 0.37 summarized at the GSA scale. Even for the same crop type, EToF varied significantly among GSAs for some tree crops and wheat 0.36 . The average almond EToF , e.g., ranged from 0.25 in the City of Tracy GSA in Tracy county to 0.75 in Rock Creek Reclamation District GSA in Chico county. Pistachio’s EToF was much lower in the majority of the western San Joaquin Valley areas , probably due to the plant stress caused by salinity . Citrus EToF had an IQR of 0.16 at the GSA scale. In contrast, the mean EToF showed much smaller variation among GSAs for the majority of annual crops such as alfalfa 0.54 , rice 0.61 , pasture 0.57 , and cotton 0.35 . Within each GSAs, the annual EToF also showed large spatial variation, with a mean CV of 31% across all agricultural fields; Some GSAs with a lower evapotranspiration rate had the highest variability , mostly located at and around the Westlands Water District region in the western-Fresno and Kings county.

In addition to crop diversity within each GSA, significant variation of EToF was also found for each tree crop type, such as almonds and pistachios , with IQRs of CVs among GSAs greater than 15%. For examples, the CVs of pistachio EToF within each GSAs had a mean of 34% and an IQR of 33% across GSAs, with the largest within-GSA variation found in the Central Delta-Mendota GSA; In the Southeast Kings GSA, CV of Pistachio EToF is 29%, much lower than its neighbor, Tri-County Water Authority GSA . Other types with highly variable EToF included almond, citrus, walnut, and wheat. In contrast, EToF was more homogeneous within GSA for alfalfa with a mean CV of 17% and IQR of 6% , and rice . About 39 GSAs had >60% of agricultural land areas planted with perennial crops including almond, pistachio, citrus, walnut, and grape in 2014, which accounted for 76% of total agricultural water use by these GSAs and 27% of Central Valley’s total agricultural water use in 2014 . These GSAs will likely face greater vulnerability to prolonged drought due to the high cost of fallowing productive orchards. When dividing the total consumptive use of perennial crops by the GSA area , we found that some small and medium-size GSAs, such as Delano-Earlimart Irrigation District GSA, Madera Water District GSA, and New Stone Water District GSA, will need to reserve a much greater depth of groundwater storage to maintain the orchards during drought.Our study showed that the semiempirical Priestley-Taylor algorithm, when calibrated with ground measurement data over diverse crop types and driven by Landsat Analysis Ready Data, improved the accuracy of the older 1 km MODIS-driven PT-0 model . The crop-specific Priestley-Taylor optimization performed consistently between the testing and independent data sets, and slightly better than the PT-JPL method . The generalized Priestley-Taylor optimization had a similar overall performance with PT-JPL when driven by the same input data. However, relatively larger uncertainties were found during nongrowing seasons, from November to March, when the evapotranspiration rate was relatively low. This was partly due to the limited field measurements data during winter and early spring for optimizing the sensitivity of actual Priestley-Taylor coefficients to the moisture content.

Moreover, our Priestley-Taylor approach does not separate soil evaporation and plant transpiration. This introduces uncertainty in evapotranspiration estimates during non-growing seasons when evapotranspiration is mainly driven by evaporation from the soil due to minimal canopy coverage or leaf area. For example, we did find that PT-JPL better captured the peak of the actual Priestley-Taylor coefficient for the corn site during the dormant season , when PT-JPL’s estimates showed that soil evaporation was the most significant component. The uncertainty of our refined Priestley-Taylor approach here is similar to the DisALEXI model, as shown by the report from the Sacramento-San Joaquin Delta intercomparison project . For additional reference, Anderson et al. reported that DisALEXI had an RMSE of 1.09 mm day−1 at site number 1 and 1.24 mm day−1 at site number 24 when compared to daily measurements. Being a process-based model, DisALEXI does not depend on land-use maps and field measurements for calibration once validated. The semiempirical Priestley-Taylor approach, however, has the advantage of easy implementation,big plastic pots compared to other more sophisticated and computationally more expensive approaches.At a regional scale, the annual mean values of per-area water use of major crop types in the Central Valley estimated here are generally within the ranges reported in the literature . For example, DWR’s water portfolio and balances data set, as part of DWR’s 2018 Water Plan, reports that water requirement by corn ranges from 390 to 835 mm yr−1 in 2014 across sub-regions of all planning areas in the Central Valley . Burt et al. estimated that corn in the Central Valley conventionally used 813 mm yr−1 in a typical precipitation year. Our regional average of corn evapotranspiration was 16% more than DWR’s average corn water requirement over planning areas . Larger differences were found for alfalfa, pasture, wheat, almonds, pistachio, and vineyard, for which our regional averages were 30%–65% lower or higher than DWR’s values. Over all 30-m pixels of agricultural lands in Central Valley , the average annual evapotranspiration rate, estimated here, is higher than the estimates over the whole Central Valley by the BESS biophysical process-based model forced with 1 km satellite observations . The discrepancy is likely due to the scale effect and differences in land cover maps. Larger pixels likely contain other land-use areas such as fallow, urban, water, and natural vegetation. Nonetheless, our estimates in 2014 align with the values reported in Schauer and Senay based on the SSEBop remote sensing evapotranspiration model driven by Landsat thermal data. Our estimation of 19.6 tera-liter water consumption in 2014 was equivalent to 74% of DWR’s estimate over all planning areas within the Central Valley, which was derived from CalSIMETAW . Among the planning areas, the discrepancies ranged from –53.4% to −18.5%, with the most significant disagreement occurring in the southern and center-east of the San Joaquin Valley .

Similarly, previous studies in the Sacramento-San Joaquin Delta, a subset of the Central Valley, showed that remote sensing estimates are lower than CALSIMETAW’s estimates by 6–24% . Over this Delta area, our crop-specific Priestley-Taylor method in this study estimated 1.20 tera-liter in 2016, very similar to the DISALEXI’s estimate of 1.16 tera-liter in the water year 2016; both were about 80% of CalSIMETAW’s estimates of 1.49 tera-liters, based on the published data summary table in Medellín-Azuara et al. . Two factors may have caused the discrepancy in regional estimates between PT-UCD and CalSIMETAW. First, CalSIMETAW’s crop-coefficient approach implemented at a regional scale may overestimate actual evapotranspiration, because it did not account for the impacts of planting variabilities such as orchard age distribution and planting density, field conditions such as salinity and disease, and crop management like deficit irrigation. Second, the land-use map used by CALSIMETAW was different from the DWR’s land-use map that we used here. For example, CALSIMETAW estimated 13.7, 23.9, and 23.9 km2 of corn, alfalfa, and pasture in PA 704 in 2014, in contrast to our DWR’s map-based estimates of 18.2, 21.4, and 9.3 km2 of corn, alfalfa, and pasture.Currently, California’s GSAs employ various approaches to estimate evapotranspiration in their water budget accounting and management plan development, causing systematic inconsistencies among GSAs. For example, the Olcese GSA near Bakersfield estimates monthly evapotranspiration from 1993 to 2015 using the METRIC method version by the Irrigation Training & Research Center at the California Polytechnic State University; North Kings GSA uses CA DWR’s crop coefficients to estimate annual evapotranspiration rate over detailed analysis units from 1998 to 2010, while the Delano and Yuba GSAs use crop coefficients published by ITRC in 2003 and derived from an SEBAL-based evapotranspiration map in 2009, respectively. Our study shows that the fractional of reference ET , or similarly crop coefficients, for most crops, varies spatially across and even within GSAs, and for some crops, EToF changes considerably between years. More consistent estimates with known uncertainty from a calibrated or thoroughly evaluated approach are needed to ensure consistent quantitative information for data-driven decisions for water planning. Our optimized Priestley-Taylor approach driven by remote sensing observations provides an efficient way to capture both spatial heterogeneity and temporal dynamics of water balance. In particular, we found that orchards and wheat generally had a greater spatial variability of evapotranspiration and crop coefficients than other major crop types, across the Central Valley, within, and among GSAs. Age distribution and other stressors such as salinity likely contributed to such evapotranspiration variability for tree crops . Among three major nut tree crops, pistachio had the lowest mean annual evapotranspiration rate , followed by walnut and almond . Coincidentally, 26% of pistachio acreages in 2014%, 18% of walnut in 2015%, and 15% of almond in 2014 across California were non-bearing orchards . The high variability of wheat water use is likely due to cultivar and end-use for the crop .