Enhancing daily reference evapotranspiration (ETref) prediction across diverse climatic zones: A pattern mining approach with DIRECTORS model

Abstract

Accurate evaluation of daily reference evapotranspiration (ET_ref) is essential for effective water resource management and drought mitigation, particularly in arid climates. However, developing countries frequently lack the necessary infrastructure for precise ET_ref assessment. Recent advancements have introduced various black box machine learning (ML) models, including the Adaptive Neuro-Fuzzy Inference System-Particle Swarm Optimization algorithm (ANF-PSO), Random Forest (RF), and Support Vector Machine (SVM), to predict daily ET_ref. Despite their effectiveness, these models suffer from a lack of interpretability, raising concerns about biases, fairness, and accountability in decision-making. Additionally, their performance varies significantly across different climatic conditions, limiting their general applicability. To address these challenges, this paper presents DIRECTORS, a novel daily ET_ref prediction model based on pattern mining. DIRECTORS leverages correlations among meteorological parameters and autonomously extracts climate-specific behavioral patterns without predefined pattern lengths. By utilizing these patterns and recent station behavior, DIRECTORS forecasts macroscopic daily ET_ref values and further refines these predictions using RF based on identified similar patterns. This innovative approach offers distinctive insights and solutions to the limitations of traditional ML models in daily ET_ref prediction. Extensive evaluation demonstrates DIRECTORS’ effectiveness and its potential to significantly enhance predictive accuracy, making it a valuable tool for water resource management and planning in varying environmental conditions.