Improvement of evapotranspiration estimates for grasslands in the southern Great Plains: Comparing a biophysical model (SWAT) and remote sensing (MODIS)

Study region: Mixed- and tall-grass prairies in the southern Great Plains, USA Study focus: Estimates of evapotranspiration (ET) are widely available from remote sensing and commonly used for water management. However, this approach is limited by prolonged satellite revisit periods and prominent algorithm-induced bias. We compared ET estimates from MODIS (Moderate Resolution Imaging Spectroradiometer) products and those based on the biophysically-based Soil and Water Assessment Tool (SWAT) to measurements using the eddy covariance (EC) technique for a subhumid, tall-grass prairie site near Stillwater, Oklahoma (OK) and a semiarid, mixed-grass prairie site near Clinton, OK in the southern Great Plains, USA. New hydrological insights for the region: SWAT and MODIS produced ET estimates closer to EC measurements for the tall-grass prairie (calibration site) than for the mixed-grass prairie (validation site), with a better performance from SWAT. For the tall-grass prairie, the R2 values were relatively high and comparable (0.77 and 0.87), and the biases were relatively small (−0.40% and 5.04%) for the SWAT and MODIS comparisons to EC. SWAT performed much better than MODIS for the mixed-grass prairie with R2 values of 0.68 vs. 0.13 and bias of − 1.87% vs. − 45.71%, respectively. The SWAT simulation also reproduced better estimates of aboveground net primary productivity than the MODIS products. This study suggests that site-specific SWAT simulations can produce better ET estimates than MODIS products, especially in the water-limited mixed-grass prairie.