Analysis and Prediction of Precipitation in Smart Agriculture by Utilizing Multiple ML Models

This research study applies five ML models (ARIMA, Prophet, Random Forest, XGBoost and an Optimal Hybrid ensemble) for forecasting monthly precipitation in the Južno-Bački district, Autonomous Province of Vojvodina, Republic of Serbia, by implementing the 73 years of meteorological data (1951–2024). The Optimal Hybrid machine learning model achieves the best performance of all models (R² = 0.4577, MAE = 17.85 mm, 83.33% CI coverage). Extreme variabilities recorded in August– September (with errors larger than 50%) emphasise the need for supplementation to meteorological data for achieving more optimal results with machine learning model processing. The research results are interpreted not only as statistical performance, but as operational, field, practical potential for smart agriculture implementation. The Optimal Hybrid model is suitable for three scenarios of irrigation (drought/normal/surplus) for the period of 12 months. The August and September Machine Learning models’ forecast errors indicate the fact that climate variables in the form of indices and expert seasonal long term projections still need to be implemented and combined for effective and reliable drought risk minimization and allocation of key resources as drinking water. Only that way system forecasts can be used in smart agriculture irrigation planning with acceptable risk. Additionally, forecast errors in August and September should be treated as diagnostic and trend related forecasts and not as exact point forecasts. The reason for this is high volatility and variability of climate values. Nevertheless the big data forecasts of this type should represent a preventive system for alarming farmers. It should also help to mitigate weaknesses of the systems currently in use that treat the cosequencies (curative methods) and not act preventively. In addition to that, sustainable development, climate change mitigation and adaptation, local alarming systems for farmers, as well as the climate change government strategies and policies in the most climate-sensitive domains such as agriculture can be improved with new technologies, innovations and particularly AI-based systems.