A Scalable Genetic Algorithm Framework for Automated Anchor Pixel Selection to Improve Satellite-Based Evapotranspiration Monitoring Using SEBAL

Evapotranspiration (ET) governs the exchange of water and energy between the land surface and the atmosphere, accounting for over 70% of agricultural water consumption. Accurate ET estimation is therefore crucial for efficient irrigation management and sustainable water allocation. Traditional in situ methods, such as lysimeters and eddy covariance towers, provide precise measurements but are costly and spatially limited. In contrast, remote sensing–based energy balance models like the Surface Energy Balance Algorithm for Land (SEBAL) offer scalable, cost-effective solutions for large-scale ET monitoring.

A critical factor determining SEBAL’s accuracy is the appropriate selection of hot and cold anchor pixels, which represent the limiting conditions of no ET and maximum ET, respectively. However, this step remains one of the most challenging and subjective aspects of SEBAL implementation. Previous approaches, including visual selection, statistical filtering, and threshold-based rules (e.g., based on NDVI, albedo, and LST), have improved consistency but still suffer from regional dependency, random selection biases, and inconsistent parameter thresholds. Methods relying on proximity to meteorological stations or calibration with lysimeter data improve accuracy but are not universally applicable due to data limitations and landscape heterogeneity. Consequently, the same scene can yield different anchor pixels across methods, leading to divergent ET estimates and reduced reproducibility.

To address these limitations, this study proposes the development of an automated, reproducible framework for anchor pixel selection using a Genetic Algorithm (GA) optimization approach. The GA systematically identifies biophysically consistent anchor pixels by exploring multidimensional feature space (NDVI, LST) while minimizing uncertainty and eliminating subjective human bias. The method is implemented using daily MODIS (Moderate Resolution Imaging Spectroradiometer) imagery, aggregated to 250 m spatial resolution to capture field-scale variability better while maintaining high temporal fidelity.

This automated approach ensures scalability, portability, and reproducibility across diverse agro-ecological regions without heavy data requirements. It offers a simple yet robust workflow suitable for operational ET monitoring and can be integrated into regional irrigation and drought management systems