Crop yield forecasting with climate data using PCA and Machine Learning

Abstract

Accurately forecasting annual crop production is crucial for countries as it enables them to formulate import and export policies and estimate the economic benefits of their agricultural planning. The crop growth is significantly influenced by weather conditions throughout the year, and climate conditions during different stages of plant development can greatly affect crop yield. The availability of historical climate data has greatly benefited the agricultural sciences and food sector, particularly with the application of Artificial Intelligence methods in big data analysis, enabling the extraction of practical information and actions. The objective of this research is to develop a predictive Machine Learning (ML) model that utilizes climate data from a specific time frame to forecast the wheat yield in the Pelagonia valley, a crucial region for wheat cultivation in North Macedonia. Principal Component Analysis (PCA) was employed as a dimensionality-reduction method to reduce the input data set's dimensionality. A least-squares boosting regression model was implemented as the ML method to estimate wheat yield from climate data. The results indicate a high accuracy of wheat yield prediction, even with limited dataset, on both the training and testing datasets. The study demonstrates the feasibility of utilizing ML methods as complementary to existing models for accurate wheat yield forecasting, offering significant advantages due to the ease of calibrating the ML model parameters.