Data supplement: Spatial autocorrelation in machine learning for modelling soil organic carbon

<p>Spatial autocorrelation in machine learning for modelling soil organic carbon: Data supplement</p> <p><br>Alexander Kmoch, Clay Taylor Harrison, Jeonghwan Choi, Evelyn Uuemaa</p> <p>Spatial autocorrelation, the relationship between nearby samples of a spatial<br>random variable, is often overlooked in machine learning models, leading to<br>biased results. This study investigates various methods to account for spa-<br>tial autocorrelation when predicting soil organic carbon (SOC) using random<br>forest models. Five models incorporating spatial structure were compared<br>against baseline models that did not have any added spatial components.<br>Cross-validation showed slight improvements in accuracy for models consid-<br>ering spatial autocorrelation, while Shapley Additive Explanations confirmed<br>the importance of spatial variables. However, no decrease in spatial autocor-<br>relation of residuals was observed. Raster-based models exhibited enhanced<br>prediction detail, but high-resolution validation data availability limited thor-<br>ough validation. The findings emphasize the value of incorporating spatial<br>autocorrelation for improved SOC prediction in machine learning models.<br>Considerations such as the distribution of predictions and computational<br>complexity should help guide the selection of suitable approaches for specific<br>spatial modelling tasks.</p>