Multicriteria Decision-Making Approach to Enhance Automated Anchor Pixel Selection Algorithm for Arid and Semi-Arid Regions

Abstract

Finding the precise value of pixel-scale evapotranspiration (ET) for an entire basin is a major challenge to hydrologists. Many efforts have been made to conquer this challenge, among which remote sensing methods are the most promising ones. The surface energy balance algorithm for land (SEBAL) is one of the several established remote sensing methods to estimate ET. The anchor pixel selection process (e.g., hot and cold pixels) is one of the critical steps in the SEBAL model that also determines the accuracy of the model outputs. Several researchers have improved anchor pixel selection by an automated fashion. In the current study, a new simple method has been proposed to seek the best anchor pixels. Then, the daily ET outputs were assessed using observed data from the Eddy covariance (EC) tower data at Santa Cruz River Watershed for 2014–2015. The results showed that the daily ET with measured data at the study site confirmed our automated anchor pixel selection method to be reliable in terms of selecting appropriate hot and cold pixels under dry conditions by producing ET maps with reasonable accuracies (R2=0.78 and RMSE=0.46 mm day−1). Our study suggests that considerations of simple image-derived parameters, such as temperature difference between hot and cold pixels, distance from a representative station, and elevation differences could improve the automatic selection of anchor pixels and the implementation of the SEBAL model under dry conditions. The machine learning technique could be combined with the proposed automated algorithm to map ET in different climates (not only arid and semiarid) faster and with more accuracy.