Approach for Long-Term Forecasting of Frosts and Droughts in Smart Agriculture

Due to the trend of a global increase in average daily temperatures and the occurrence of extreme weather events, the task of long-term forecasting of agrometeorological risks is becoming increasingly actual. The work considers frost and drought as the main meteorological risks. In each individual case, based on a preliminary analysis of the initial information, researchers select the most effective method; there is no comprehensive methods’ comparison. In this regard, the goal of the work was to formulate the concept of a unified intelligent system for long-term forecasting of drought and frost. The proposed approach involves generating traditional, well-studied models for each source dataset in real time and selecting the most accurate result. For the computational experiment, three datasets were prepared: Datasets 1 and 2 for the period February 1, 2005–February 26, 2024, with minimum daily temperatures and average humidity, respectively; Dataset 3 is a part of Dataset 1 for the period July 1, 2023– February 26, 2024. The results showed that different methods turn out to be the most accurate for different source data: for Dataset 1—SNaive method, for Dataset 2—ARMA and SNaive, and for Dataset 3—SES and FIT_ARIMA. This confirms the validity of the proposed approach.