Forecasting by means of a novel regression approach involving time-varying parameters: wheat yield projection based on climatic data

Farming systems worldwide have exhibited climatic adaptation and improved crop outputs over the last century. Nevertheless, interannual yield fluctuations still drive volatility in food commodity prices, challenging food security. Predicting short-term yield variations from observed climatic patterns can thus provide significant benefits. This paper enhances a transparent, computational efficient and closed-form probabilistic short-term yield forecasting method introduced in our previous research. The case of wheat yield for an Italian province, where the Standardized Precipitation-Evapotranspiration Index (SPEI) was identified as a key predictor, is considered for illustrative purposes. Forecasting is defined as evaluating the conditional probability distribution of the next yield value, given the recorded SPEI in a selected month of the current year. As a main challenge is that the yield–SPEI relationship exhibits parameters varying over time, here we propose a least-squares method adapted for interpolating curves with time-varying parameters. Combined with resampling techniques, this enables robust probabilistic forecasts. Validation through Monte Carlo simulations and resampling confirmed the method’s effectiveness. The framework can be potentially tailored to different crops, regions, and spatial scales. Such predictive capacity may prove valuable in preparing stakeholders across the crop production–consumption chain, including financial markets, in managing risks linked to unexpected production outcomes.