Decision tree algorithms, such as random forest, have become a widely adapted method for mapping soil properties in geographic space. However, implementing explicit spatial trends into these algorithms has proven problematic. Using x and y coordinates as covariates gives orthogonal artifacts in the maps, and alternative methods using distances as covariates can be inflexible and difficult to interpret. We propose instead the use of coordinates along several axes tilted at oblique angles to provide an easily interpretable method for obtaining a realistic prediction surface. We test the method on four spatial datasets and compare it to similar methods. The results show that the method provides accuracies better than or on par with the most reliable alternative methods, namely kriging and distance-based covariates. Furthermore, the proposed method is highly flexible, scalable and easily interpretable. This makes it a promising tool for mapping soil properties with complex spatial variation.

Predicting wheat yield is crucial due to the importance of wheat across the world. When modeling yield, the difference between potential and actual yield consistently changes because of advances in technology. Considering historical yield potential would help determine spatiotemporal trends in agricultural development. Comparing current and historical yields in Denmark is possible because yield potential has been documented throughout history. However, the current national winter wheat yield map solely uses soil properties within the model. The aim of this study was to generate a new Danish winter wheat yield map and compare the results to historical yield potential. Utilizing random forest with soil, climate, and topography variables, a winter wheat yield map was generated from 876 field trials carried out from 1992 to 2018. The random forest model performed better than the model based only on soil. The updated national yield map was then compared to yield potential maps from 1688 and 1844. While historical time periods are characterized by numerous low yield potential areas and few highly productive areas, current yield is evenly distributed between low and high yields. Advances in technology and farm practices have exceeded historical yield predictions, mainly due to the use of fertilizer, irrigation, and drainage. Thus, modeling yield projections could be unreliable in the future as technology progresses.

Convolutional neural networks (CNNs) have been originally used for computer vision tasks, such as image classification. While several digital soil mapping studies have been assessing these deep learning algorithms for the prediction of soil properties, their potential for soil classification has not been explored yet. Moreover, the use of deep learning and neural networks in general has often raised concerns because of their presumed low interpretability (i.e., the black box pitfall). However, a recent and fast-developing sub-field of Artificial Intelligence (AI) called explainable AI (XAI) aims to clarify complex models such as CNNs in a systematic and interpretable manner. For example, it is possible to apply model-agnostic interpretation methods to extract interpretations from any machine learning model. In particular, SHAP (SHapley Additive exPlanations) is a method to explain individual predictions: SHAP values represent the contribution of a covariate to the final model predictions. The present study aimed at, first, evaluating the use of CNNs for the classification of potential acid sulfate soils located in the wetland areas of Jutland, Denmark (c. 6,500 km²), and second and most importantly, applying a model-agnostic interpretation method on the resulting CNN model. About 5,900 soil observations and 14 environmental covariates, including a digital elevation model and derived terrain attributes, were utilized as input data. The selected CNN model yielded slightly higher prediction accuracy than the random forest models which were using original or scaled covariates. These results can be explained by the use of a common variable selection method, namely recursive feature elimination, which was based on random forest and thus optimized the selection for this method. Notably, the SHAP method results enabled to clarify the CNN model predictions, in particular through the spatial interpretation of the most important covariates, which constitutes a crucial development for digital soil mapping.