Future Forecasting of Temperature in Nineveh Governorate Using the Hybrid AR–ANN Model.

Abstract

This research aims to predict future temperatures in Nineveh Governorate using the AR-ANN hybrid model by combining linear statistical analysis with artificial intelligence techniques to address the temporal complexity of climate series. The study relied on maximum and minimum temperature data recorded at seven selected climate stations for the historical period 1994-2024 , Issued by the General Authority for Meteorology and Seismic Monitoring, for the purpose of analyzing climate trends and building a predictive model for the future period (2025-2034), in the first stage, an autoregressive (AR) model was built to determine the linear structure of the time series based on the analyses of (ACF) and (PACF) and the criteria of (AIC, BIC), The resulting time values ​​were then used as inputs to a multi-layer artificial neural network (MLP) to process the remaining nonlinear components, after which the two models were combined within a hybrid framework (ANN–AR). The efficiency of the models was evaluated using statistical accuracy measures, the most important of which are RMSE and MAPE.

The results showed the hybrid model's superiority over the traditional linear model in terms of reducing error values ​​and improving forecast accuracy at almost all stations, with spatial variation in performance reflecting the local characteristics of each station. Future projections also indicated a slight to moderate upward trend in maximum temperatures , A more consistent rise in minimum temperatures, especially during the winter months, indicates a trend toward a narrowing of the daily temperature range and an increase in nighttime warming.

The study confirms that adopting hybrid models represents a more suitable methodological framework for representing the complex temporal behavior of climatic elements in semi-arid environments, and provides a scientific basis for supporting climate planning and resource management in Nineveh Governorate.