Comparative evaluation of machine learning techniques in predicting fundamental meteorological factors based on survey data from 1981 to 2021

Predicting the meteorological factors of the climate in the medium and long term is a significant challenge with socio-economic and environmental implications, given its complex and chaotic nature. The current short-term weather predictions the Iraqi meteorological organization offers are less valuable. As a result, this study introduces four machine-learning methods-artificial neural network, support vector machine, random forest (RF), and K nearest neighbors-to forecast six meteorological factors: total precipitation (TPRE), minimum temperature (MINT), maximum temperature (MAXT), relative humidity (RHUM), top-of-atmosphere radiation (TOAR), and wind speed (WIND) up to 1, 3, 6, and 12 months ahead in four Iraqi governorates. Data on these factors from 1981 to 2021 were extracted from the Modern-ERA Retrospective Analysis for Research and Applications version 2 dataset. The findings indicate that the RF algorithm outperformed other algorithms regarding prediction accuracy, while the SVR algorithm exhibited the least accuracy. Moreover, TPRE had the lowest performance with an average root-mean-square error (RMSE) of 19.002; conversely, RHUM and WIND showed much better performance with average RMSE values of 7.259 and 0.192 respectively. The highest performance was observed for MINT (MAXT and TOAR prediction with average-RMSE values of 2.346, 2.244, and 5.314, respectively). The present study's findings will bring significant advantages in safeguarding human lives and property and promoting health, security, and economic prosperity.