Gaussian process regression as an alternative to kriging and SVM for spatial yield prediction

Detecting spatial yield variability is essential for precision agriculture because it minimizes environmental impact and enhances economic returns. This study proposes Gaussian Process Regression (GPR) as a predictive model for yield estimation, particularly in cultivated areas where the highest yields appear in the central region of the field, while the edges exhibit lower productivity. The study was conducted in Patos de Minas, Brazil, using 795 georeferenced soybean yield samples over 3.7 hectares. The analysis evaluates GPR across different sample sizes and compares it with Ordinary Kriging (OK) and Support Vector Machine (SVM). The results indicate that GPR and OK perform similarly under high sampling densities, but GPR achieves higher predictive accuracy under low-sampling conditions. A sample size of at least 60% of the full dataset is necessary to maintain reliable spatial prediction, as smaller sample sizes lead to greater prediction errors and less defined spatial patterns. SVM, in contrast, produces a smoothing effect across all sampling densities, which reduces its ability to capture local variations. These findings highlight GPR as a robust alternative for yield mapping, particularly in scenarios with limited data availability. From a practical perspective, GPR and OK remain strong candidates for yield prediction, reinforcing the importance of model selection based on data availability and spatial variability.