Articles
Enhancing water management in Morocco’s arid regions: using advanced models for accurate ETO estimation
Article number
1422_33
Pages
267 – 274
Language
English
Abstract
Addressing the need for effective water resource management in Morocco’s semi-arid and arid regions, this research enhances the accuracy of reference evapotranspiration (ETO) estimation.
ETO is crucial for understanding the hydrological cycle, assessing climate change impacts, and optimizing agricultural water use.
Estimating ETO is challenging due to the measurements of various climatic parameters.
Our approach integrates deep learning (DL) and machine learning (ML) with physics-based models, namely Hargreaves and Food and Agriculture Organization Penman-Monteith (FAO-PM), to predict ETO using limited meteorological data, including mean air temperature and global solar radiation.
The data, covering the period 2013 to 2020, were acquired from local weather stations and the ERA5-Land reanalysis.
Three DL models – long short-term memory neural network (LSTM), convolutional neural network (CNN), and gated recurrent unit (GRU) – along with four ML models: XGBoost, Random Forest, Support Vector Machine (SVM), and linear regression – were evaluated.
LSTM and XGBoost emerged as the most robust, with XGBoost achieving a determination coefficient (R2) of 0.93 and a root mean square error (RMSE) of 0.03 mm day‑1, highlighting the efficiency of using the IA model with limited data for ETO estimation.
Our model has been tested over another Moroccan region and it demonstrates effectiveness, and consistent accuracy across diverse geographical areas.
ETO is crucial for understanding the hydrological cycle, assessing climate change impacts, and optimizing agricultural water use.
Estimating ETO is challenging due to the measurements of various climatic parameters.
Our approach integrates deep learning (DL) and machine learning (ML) with physics-based models, namely Hargreaves and Food and Agriculture Organization Penman-Monteith (FAO-PM), to predict ETO using limited meteorological data, including mean air temperature and global solar radiation.
The data, covering the period 2013 to 2020, were acquired from local weather stations and the ERA5-Land reanalysis.
Three DL models – long short-term memory neural network (LSTM), convolutional neural network (CNN), and gated recurrent unit (GRU) – along with four ML models: XGBoost, Random Forest, Support Vector Machine (SVM), and linear regression – were evaluated.
LSTM and XGBoost emerged as the most robust, with XGBoost achieving a determination coefficient (R2) of 0.93 and a root mean square error (RMSE) of 0.03 mm day‑1, highlighting the efficiency of using the IA model with limited data for ETO estimation.
Our model has been tested over another Moroccan region and it demonstrates effectiveness, and consistent accuracy across diverse geographical areas.
Publication
Authors
O. Kaissi, S. Er-Raki, S. Belaqziz, M.H. Kharrou, A. Amazirh, E.H. Bouras, C. El Hachimi, A. Chehbouni
Keywords
reference evapotranspiration, deep learning, ERA5-Land reanalysis data, FAO-56 Penman-Monteith, Hargreaves
Online Articles (63)