Assessment of automated evapotranspiration estimates obtained using the GP-SEBAL algorithm for dry forest vegetation (Caatinga) and agricultural areas in the Brazilian semiarid region

Abstract

Real evapotranspiration (ETr) plays a key role in water balance, especially for dry forest vegetation (Caatinga) and irrigated agricultural areas, as is the case in the Petrolina region located in the Brazilian semiarid region (BSAR). Population growth increases the need for food and the use of water resources, which are scarcer in semiarid regions. Thus, knowledge of the energy balance (EB) and radiative balance are indispensable for the management of water resources, especially in irrigated agricultural areas. Currently, one of the biggest challenges in the application of algorithms for estimating ETr based on satellite images is the calibration of the EB. The goal of this study was to evaluate the applicability of Geographic Resources Analysis Support System (GRASS) Python surface EB algorithm for land (GP-SEBAL) to calculate ETr for Caatinga vegetation and agricultural areas in the BSAR. GP-SEBAL was applied to two images from Landsat 8 obtained in 2013 and compared with manual SEBAL and observed data. In this study, kappa coefficient accuracy was used to compare the ground truth data and predicted land use and land cover. The performance of algorithms in determining the EB components, land-surface temperature (LST), net radiation (Rn), soil heat flux (G), latent heat flux (LE), sensible heat flux (H), and ETr were compared with statistical indices and uncertainty percentages (PUs) were calculated. In addition, a sensitivity analysis using a deviation of 30 W/m2 with an interval of 5 W/m2 was used to analyze the sensitivity of ETr. The results showed a strong correlation between GP-SEBAL and manual SEBAL for ETr, LST, Rn, LH, and H. The values suggest the regularity of estimating the EB in an automated manner. The results indicate a small PU difference in the estimated ETr values calculated using GP-SEBAL and SEBAL for two images of 10.87% and 20.11% for Caatinga vegetation and 15.01% and 35.16% for agriculture, respectively. We conclude that GP-SEBAL is an efficient automated algorithm for estimating ETr that uses a less complex process with a lower incidence of errors and considerably faster execution time than classical applications of SEBAL.