Answers are grounded in this paper and cite their sources.
Scalable reference evapotranspiration (ET₀) estimation across sparse meteorological networks remains critical for climate-smart irrigation in data-constrained regions. This study presents the first systematic evaluation combining spatial model transferability with progressively reduced input configurations for ET₀ prediction across contrasting agro-climatic zones in Bangladesh; prior cross-station studies never compared six modeling approaches under nine reduced input scenarios without recalibration. Six models — Multiple Linear Regression, Support Vector Regression, Random Forest, Light Gradient Boosting Machine, Artificial Neural Network (ANN), and Convolutional Neural Network (CNN) — were trained on 20-year daily data (2000–2019) from subtropical humid Mymensingh and transferred without recalibration to semi-arid Bogura, with complete transferability evaluated across all nine configurations. Under full inputs, CNN achieved optimal performance (RMSE = 0.2670 mm/day, R² = 0.9518), while ANN demonstrated superior transferability (RMSE = 0.3161 mm/day, R² = 0.9339; 18% degradation). Excluding relative humidity incurred minimal transferability cost (CNN: RMSE = 0.3728 mm/day, R² = 0.9080), while temperature-sunshine combinations offered optimal accuracy-parsimony balance. Temperature-only configurations produced catastrophic transferability failure (RMSE = 0.9051 mm/day, R² = 0.4579), falling below hydrological thresholds. Principal component analysis confirmed consistent deep learning superiority across eight of nine configurations, providing an actionable framework for ET₀ estimation across data-sparse South Asian agricultural systems.
No takes yet. Share an insight, caveat, or question.
Islam et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: