Mapping of Soils and Land-Related Environmental Attributes in France: Analysis of End-Users’ Needs

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. 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. Further, these algorithms have shown a high confidence in their predictions in areas geographically distant from the training area or areas 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 subdivide the soils into soil regions and as a test case we explicitly exclude two soil regions in the training area but include these regions in the prediction. Our results emphasize the need for uncertainty measurement to obtain more reliable and interpretable results of ANNs, especially for regions far away from the training area. Moreover, the knowledge gained from this research addresses the problem of overconfidence of ANNs and provides valuable information on the predictability of soil types and the identification of knowledge gaps. By analyzing regions where the model has limited data support and, consequently, high uncertainty, stakeholders can recognize the areas that require more data collection efforts.

Assessing Climate Change Impact on Soil Salinity Dynamics between 1987–2017 in Arid Landscape Using Landsat TM, ETM+ and OLI Data

This paper examines the climate change impact on the spatiotemporal soil salinity dynamics during the last 30 years (1987–2017) in the arid landscape. The state of Kuwait, located at the northwest Arabian Peninsula, was selected as a pilot study area. To achieve this, a Landsat- Operational Land Imager (OLI) image acquired thereabouts simultaneously to a field survey was preprocessed and processed to derive a soil salinity map using a previously developed semi-empirical predictive model (SEPM). During the field survey, 100 geo-referenced soil samples were collected representing different soil salinity classes (non-saline, low, moderate, high, very high and extreme salinity). The laboratory analysis of soil samples was accomplished to measure the electrical conductivity (EC-Lab) to validate the selected and used SEPM. The results are statistically analyzed (p ˂ 0.05) to determine whether the differences are significant between the predicted salinity (EC-Predicted) and the measured ground truth (EC-Lab). Subsequently, the Landsat serial time’s datasets acquired over the study area with the Thematic Mapper (TM), Enhanced Thematic Mapper Plus (ETM+) and OLI sensors during the last three decades over the intervals (1987, 1992, 1998, 2000, 2002, 2006, 2009, 2013, 2016 and 2017) were radiometrically calibrated. Likewise, the datasets were atmospherically and spectrally normalized by applying a semi-empirical line approach (SELA) based on the pseudo-invariant targets. Afterwards, a series of soil salinity maps were derived through the application of the SEPM on the images sequence. The trend of salinity changes was statistically tested according to climatic variables (temperatures and precipitations). The results revealed that the EC-Predicted validation display a best fits in comparison to the EC-Lab by indicating a good index of agreement (D = 0.84), an excellent correlation coefficient (R2 = 0.97) and low overall root mean square error (RMSE) (13%). This also demonstrates the validity of SEPM to be applicable to the other images acquired multi-temporally. For cross-calibration among the Landsat serial time’s datasets, the SELA performed significantly with an RMSE ≤ ± 5% between all homologous spectral reflectances bands of the considered sensors. This accuracy is considered suitable and fits well the calibration standards of TM, ETM+ and OLI sensors for multi-temporal studies. Moreover, remarkable changes of soil salinity were observed in response to changes in climate that have warmed by more than 1.1 °C with a drastic decrease in precipitations during the last 30 years over the study area. Thus, salinized soils have expanded continuously in space and time and significantly correlated to precipitation rates (R2 = 0.73 and D = 0.85).