Long-Term (1986–2015) Crop Water Use Characterization over the Upper Rio Grande Basin of United States and Mexico Using Landsat-Based Evapotranspiration

Abstract

The evaluation of historical water use in the Upper Rio Grande Basin (URGB), United States and Mexico, using Landsat-derived actual evapotranspiration (ETa) from 1986 to 2015 is presented here as the first study of its kind to apply satellite observations to quantify long-term, basin-wide crop consumptive use in a large basin. The rich archive of Landsat imagery combined with the Operational Simplified Surface Energy Balance (SSEBop) model was used to estimate and map ETa across the basin and over irrigated fields for historical characterization of water-use dynamics. Monthly ETa estimates were evaluated using six eddy-covariance (EC) flux towers showing strong correspondence (r2 > 0.80) with reasonable error rates (root mean square error between 6 and 19 mm/month). Detailed spatiotemporal analysis using peak growing season (June–August) ETa over irrigated areas revealed declining regional crop water-use patterns throughout the basin, a trend reinforced through comparisons with gridded ETa from the Max Planck Institute (MPI). The interrelationships among seven agro-hydroclimatic variables (ETa, Normalized Difference Vegetation Index (NDVI), land surface temperature (LST), maximum air temperature (Ta), potential ET (ETo), precipitation, and runoff) are all summarized to support the assessment and context of historical water-use dynamics over 30 years in the URGB.

Characterizing Crop Water Use Dynamics in the Central Valley of California Using Landsat-Derived Evapotranspiration

“…The relative accuracy of SSEBop ETa has been evaluated multiple times and shown to match well with Ameriflux eddy-covariance flux towers as well as Max Planck Institute monthly ETa, demonstrating that SSEBop can detect spatial variability and monthly and annual trends with reasonable accuracy [20,22,[24][25][26][27]. In Senay, et al [5], we compared monthly SSEBop ETa to monthly MPI ETa from 1984 to 2011 for eight HUC-8 sub-basins in the middle and lower Central Valley and found an average r 2 of 0.76 and average root mean square error (RMSE) of 11.7 mm.…”

“…All Landsat pixels containing the Central Valley were processed in GEE using the SSEBop model as well as new cloud-based interpolation and aggregation algorithms for all scenes with 60% cloud cover or less. This includes masking clouds using the Landsat Quality Assessment band and Landsat 7 scan-line errors and filling these pixels through linear interpolation from contemporaneous scenes (within 48 days before/after an image) [22].…”

“…The GridMET ETr has been bias-corrected with a coefficient of 0.85 derived from a comparison with California Irrigation Management Information System (CIMIS) station data (https://cimis.water.ca.gov/). Similar to Senay, et al [22], before ETa was computed for each scene, cloud-masked pixels were filled using per-pixel linear interpolation from images 48 days before and after the scene date. Daily ETa was calculated using Equation (1) in GEE along with daily GridMET ETr and then aggregated to the annual total.…”

“…In Senay, et al [5], we compared monthly SSEBop ETa to monthly MPI ETa from 1984 to 2011 for eight HUC-8 sub-basins in the middle and lower Central Valley and found an average r 2 of 0.76 and average root mean square error (RMSE) of 11.7 mm. More recently, we compared monthly SSEBop ETa to Ameriflux eddy-covariance flux tower monthly ETa in the Upper Rio Grande Basin for non-cropland environments such as forest, shrubland, and grassland sites from 2007 to 2014 and found an average r 2 of 0.85 and average normalized RMSE of less than 10% [22].…”

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

“…The number of Landsat imagery and sensor types used in this study is presented in Table . 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 gap‐filled with simple linear interpolation using Landsat images in a 48‐day window (Senay et al, ).…”