Open Access

ARTICLE

Statistical and Visual Evaluation of Artificial Neural Networks and Multiple Linear Regression Performances in Estimating Reference Crop Evapotranspiration for Mersin

Fatma Bunyan Unel^1,*, Lutfiye Kusak¹, Murat Yakar¹, Abdullah Sahin², Hakan Dogan³, Fikret Demir⁴

1 Department of Geomatics Engineering, Engineering Faculty, Ciftlikkoy Campus, Mersin University, Mersin, 33343, Turkiye
2 T.C. Governorship of Adana, Adana, 01060, Turkiye
3 Turkish State Meteorological Service, Ankara, 06120, Turkiye
4 Mersin Provincial Directorate of Meteorology Department, Mersin, 33130, Turkiye

* Corresponding Author: Fatma Bunyan Unel. Email:

(This article belongs to the Special Issue: Resource and Environmental Information Modeling)

Revue Internationale de Géomatique 2025, 34, 433-460. https://doi.org/10.32604/rig.2025.065502

Received 14 March 2025; Accepted 23 June 2025; Issue published 29 July 2025

Abstract

This study aimed to create a model for calculating the total reference crop evapotranspiration (ETo) in Mersin Province from May 2015 to 2020 and to generate maps using spatial analysis. Lemons from citrus play a significant role in Mersin’s agriculture, and because of lemons’ sensitivity to temperature, ETo is essential for them. It was observed that the ETo value () calculated using the Penman-Monteith (PM) method increased over the years. A model was developed using data from 36 Automated Weather Observing Systems (AWOS) in Mersin, Türkiye, which is located in a semi-arid climate zone. The model was created using Multiple Linear Regression (MLR) and artificial neural network (ANN) methods. The station climate data were divided into training and test datasets separately and collectively, and ETo values were estimated with different combinations using three scenarios and six model constructs. The dataset was divided into training (2015–2018) and testing (2019–2020). ANN1 and MLR1 are analyses of individual AWOS, while the other models are analyses of all AWOS together. The statistical performance analysis involved a comparison of the R2, Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Root Mean Square Error (RMSE) values. The analysis results indicated that ANN1 (0.9997, 0.0105, 0.2718%, and 0.0162, respectively) and ANN2 (0.9958, 0.0678, 1.5341%, and 0.0864, respectively) models successfully predicted as statistical with both single and all AWOS. The models were visually evaluated using the Inverse Distance Weighting (IDW) interpolation method, and maps of plant water consumption were generated. The relationships between both models and years in the monthly total ETo maps allowed for a clearer comparison.

---

Keywords

---