Automated Multi-Horizon Forecasting of Hourly Reference Evapotranspiration Using Statistical, Tabular, Ensemble, Deep-learning and Foundation Models

Hourly reference evapotranspiration (ET0) forecasting plays a critical role in irrigation scheduling, water resource planning, and climate-resilient agricultural management. Short-term multi-horizon forecasting of hourly ET0 remains relatively underexplored, particularly using automated forecasting frameworks and emerging time-series foundation models. This study proposes an automated multi-horizon forecasting framework for hourly ET0 prediction using statistical, tabular, ensemble, deep-learning, and foundation forecasting paradigms implemented through the AutoGluon TimeSeries (AG-TS) framework. Hourly ET₀ values were computed using the ASCE Penman–Monteith equation from meteorological observations collected between October 2015 and December 2025. Forecasting experiments were conducted for four operational horizons (1-h, 6-h, 24-h, and 168-h) using SeasonalNaive, Theta, ETS, RecursiveTabular, DirectTabular, WeightedEnsemble, Temporal Fusion Transformer (TFT), and Chronos2 models. Results revealed horizon-dependent model behaviour, with Chronos2 achieving superior performance at the 1-h and 168-h forecasting horizons, while WeightedEnsemble demonstrated better predictive capability at the 6-h and 24-h horizons. Comparative analysis further indicated that Chronos2 maintained competitive forecasting accuracy with comparatively moderate computational demand, whereas TFT incurred substantially higher training costs despite rapid inference capability. These findings demonstrate the potential of automated forecasting frameworks for robust multi-horizon ET0 prediction and highlight the emerging applicability of pretrained time-series foundation models for operational hydrological and agro-environmental forecasting tasks.

Authors

Publication Details

Journal
Water Resources Management
Published
2026-11-19
DOI
https://doi.org/10.1007/s11269-026-04908-5
Citations
1
Primary Topic
Hydrological Forecasting Using AI
Type
article
Field-Weighted Citation Impact
1.30

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Automated Multi-Horizon Forecasting of Hourly Reference Evapotranspiration Using Statistical, Tabular, Ensemble, Deep-learning and Foundation Models

Santosh Kumar M B, Sowmya Mangalath Ravindran
1 citations
Water Resources Management
Hydrological Forecasting Using AI
1.30
article

Automated Multi-Horizon Forecasting of Hourly Reference Evapotranspiration Using Statistical, Tabular, Ensemble, Deep-learning and Foundation Models

Santosh Kumar M B, Sowmya Mangalath Ravindran
article en
1 citations

Abstract

Hourly reference evapotranspiration (ET0) forecasting plays a critical role in irrigation scheduling, water resource planning, and climate-resilient agricultural management. Short-term multi-horizon forecasting of hourly ET0 remains relatively underexplored, particularly using automated forecasting frameworks and emerging time-series foundation models. This study proposes an automated multi-horizon forecasting framework for hourly ET0 prediction using statistical, tabular, ensemble, deep-learning, and foundation forecasting paradigms implemented through the AutoGluon TimeSeries (AG-TS) framework. Hourly ET₀ values were computed using the ASCE Penman–Monteith equation from meteorological observations collected between October 2015 and December 2025. Forecasting experiments were conducted for four operational horizons (1-h, 6-h, 24-h, and 168-h) using SeasonalNaive, Theta, ETS, RecursiveTabular, DirectTabular, WeightedEnsemble, Temporal Fusion Transformer (TFT), and Chronos2 models. Results revealed horizon-dependent model behaviour, with Chronos2 achieving superior performance at the 1-h and 168-h forecasting horizons, while WeightedEnsemble demonstrated better predictive capability at the 6-h and 24-h horizons. Comparative analysis further indicated that Chronos2 maintained competitive forecasting accuracy with comparatively moderate computational demand, whereas TFT incurred substantially higher training costs despite rapid inference capability. These findings demonstrate the potential of automated forecasting frameworks for robust multi-horizon ET0 prediction and highlight the emerging applicability of pretrained time-series foundation models for operational hydrological and agro-environmental forecasting tasks.

Water Resources ManagementVol. 40(15)
Ministry of Earth Sciences, Cochin University of Science and Technology
Climate action
Openalex Percentile: Top 29%
Hydrological Forecasting Using AI
1.30
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon