- Slovak University of Technology Bratislava, Faculty of Civil Engineering, Department of Land and Water Resources Management, Bratislava, Slovakia (milan.cisty@stuba.sk)
Climate change impact assessments frequently rely on synthetic or downscaled meteorological datasets that may lack essential climatic variables such as humidity, radiation, and wind speed. This deficiency restricts the applicability of classical reference evapotranspiration (ETo) models, particularly the FAO-56 Penman-Monteith approach, which requires multiple climatic inputs and thus introduces additional uncertainty. This study aims to develop a data-efficient methodology for estimating ETo and irrigation water requirements using only temperature and precipitation, variables that are more readily available and less uncertain in climate generators. Two modelling approaches for ETo are evaluated: machine learning and optimised empirical equations. Machine learning models were trained on CarpatClim database. Although this database only contains data up to 2010, the authors of the study show the advantages of using such products for training models that can be used for future periods and for prospective studies of the impact of climate change. Daily ground-based meteorological records from the case study region also support calibration and validation. Model performance is assessed using a range of statistical measures.
Results indicate that machine learning models can accurately estimate ETo with minimal input data, outperforming empirical equations in both accuracy and predictive robustness. The estimated ETo values are incorporated into a water balance framework to determine irrigation water abstractions, accounting for soil moisture conditions, precipitation deficits, and plant water demand. For this purpose, a custom model based on the FAO CROPWAT model was created in the R language. These findings demonstrate the potential of a hybrid machine learning and water balance approach for assessing evapotranspiration and irrigation requirements under limited data conditions.
The proposed methodology offers significant benefits for climate change impact studies, agricultural water planning, and regions with incomplete meteorological observations. It also advances the practical implementation of data-light irrigation modelling, supporting broader applications in hydrological and environmental management.
Keywords: reference evapotranspiration, machine learning, CarpatClim database
How to cite: Cisty, M.: Robust Evapotranspiration Estimation Under Limited Data Conditions, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-17774, https://doi.org/10.5194/egusphere-egu26-17774, 2026.
