ISPRS Int. J. Geo-Inf. 2025, 14(6), 224; https://doi.org/10.3390/ijgi14060224 - 5 Jun 2025
Abstract
►
Show Figures
Integrating spatio-temporal kriging with machine learning improves estimation accuracy by addressing complex spatial and temporal variations in spatio-temporal phenomena. The improvement can be attributed to the enhanced flexibility of machine learning in capturing non-linear global trends, which traditional methods struggle to model, while
[...] Read more.
Integrating spatio-temporal kriging with machine learning improves estimation accuracy by addressing complex spatial and temporal variations in spatio-temporal phenomena. The improvement can be attributed to the enhanced flexibility of machine learning in capturing non-linear global trends, which traditional methods struggle to model, while kriging remains effective in representing spatio-temporal interactions. However, differences in the estimated global trends and spatio-temporal interactions resulting from applying machine learning may influence the spatio-temporal variation patterns of the kriging results. Therefore, this study evaluates the effectiveness of machine learning in spatio-temporal kriging using NO2 concentrations in Seoul, focusing on its impact on overall accuracy and the contributions to global trends and spatio-temporal interactions. The results show that integrating machine learning enhances overall accuracy relative to ordinary spatio-temporal kriging. Global trend estimates differ by the models, with polynomial regression producing smoother patterns but larger errors, while random forest and boosting yield more abrupt patterns with smaller errors. These differences lead to smoother kriging outcomes in the polynomial model and more discrete patterns in the ensemble-based models. This study highlights the importance of considering both overall estimation accuracy and spatio-temporal patterns when selecting kriging methods.
Full article
Figure 1
