Long-Term Prediction of Climate Variables of the Lake Van Basin: A Comprehensive Analysis Using Statistical and Deep Learning Models

Abstract

The aim of the present study is to predict environmental conditions in the basin of Lake Van in the Eastern Anatolia, Türkiye. The datasets for the period between January 1950 and June 2023 that include critical parameters like sunshine hours, temperature, precipitation, and humidity were used for predicting. The research developed a two-way approach that combines statistical-based models which include Auto.Arima, SARIMA and TBATS models together with deep learning-based models like NNTAR, MLP and ELM. The model's accuracy and reliability are being evaluated by using the test lengths of 60, 72, and 84 months periods during the forecasting process. Evaluation metrics, especially the root mean square error (RMSE) and mean absolute error (MAE) were used to numerically show the capability of every model. This trial is a major contribution to environmental forecasting literature by focusing on a particular geographical zone, ensuring environment parameters diversity, employing diverse prediction models, and offering a long-term view. The conclusions have been proved meaningful for researchers and environmentalists who dedicate their time to climate science, environmental management, and sustainable development in the Eastern Anatolia region of Türkiye