How can we quantify, explain, and apply the uncertainty of complex soil maps predicted with neural networks?

Artificial neural networks (ANNs) have proven to be a useful tool for complex questions that involve large amounts of data, for example, predicting soil classes on various scales. Our use case of predicting soil maps with ANNs is in high demand by government agencies, construction companies, or farmers, given cost and time intensive field work.
However, there are two main challenges when applying ANNs. In their most common form, deep learning algorithms do not provide interpretable predictive uncertainty. This means that properties of an ANN such as the certainty and plausibility of the predicted variables, rely on the interpretation by experts rather than being quantified by evaluation metrics validating the ANNs. This leads to the second challenge: these algorithms have shown a high confidence in their predictions in areas geographically distant from the training area or areas only sparsely covered by training data.

To tackle these challenges, we use the Bayesian deep learning approach “last-layer Laplace approximation”, which is specifically designed to quantify uncertainty into deep networks, in our explorative study on soil classification. It corrects the overconfident areas without reducing the accuracy of the predictions, giving us a more realistic uncertainty expression of the model's prediction.  In our study area in southern Germany we divide the soils into typical soils of valleys, the Swabian Jura and the Black Forest. As a test case, we then explicitly exclude the soil types of Swabian Jura and Black Forest in the training area but include these regions in the prediction. These two regions are characterized by very different soil types compared to the rest of the study area due to their considerably different geology, climate, and terrain.

Our findings emphasize the need to address the issue of overconfidence in ANNs, particularly for distant regions from the training area. Moreover, the insights gained from this research are not only limited to addressing overconfidence in ANNs, but also offer valuable information on the predictability of soil types and identifying knowledge gaps. By analysing regions where the model has limited data support and, consequently, high uncertainty, stakeholders can recognize the areas that require more data collection efforts.