Mapping of soil properties at high resolution in Switzerland using boosted geoadditive models

Abstract

Abstract. High-resolution maps of soil properties are a prerequisite for assessing soil threats and soil functions and to foster sustainable use of soil resources. For many regions in the world precise maps of soil properties are missing, but often sparsely sampled and discontinuous (legacy) soil data are available. Soil property data (response) can then be related by digital soil mapping (DSM) to spatially exhaustive environmental data that describe soil forming factors (covariates) to create spatially continuous maps. With air- and spaceborne remote sensing data and multi-scale terrain analysis large sets of covariates have become common. Building parsimonious models, amenable to pedological interpretation, is then a challenging task. We propose a new boosted geoadditive modelling framework (geoGAM) for DSM. A geoGAM models smooth nonlinear relations between responses and single covariates and combines these model terms additively. Residual spatial autocorrelation is captured by a smooth function of spatial coordinates and nonstationary effects are included by interactions between covariates and smooth spatial functions. The core of fully automated model building for geoGAM is componentwise gradient boosting. We illustrate the application of the geoGAM framework by using soil data from the Canton of Zurich, Switzerland. We modelled effective cation exchange capacity (ECEC) in forest topsoils as continuous response. For agricultural land we predicted the presence of waterlogged horizons in given soil depth layers as binary and drainage classes as ordinal responses. For the latter we used proportional odds geoGAM taking the ordering of the response properly into account. Fitted geoGAM contained only few covariates (7 to 17) selected from large sets (333 covariates for forests, 498 for agricultural land). Model sparsity allowed covariate interpretation by partial effects plots. Prediction intervals were computed by model-based bootstrapping for ECEC. Predictive performance of the fitted geoGAM, tested with independent validation data and specific skill scores (SS) for continuous, binary and ordinal responses, compared well with other studies that modelled similar soil properties. SS of 0.23 up to 0.53 (with SS = 1 for perfect predictions and SS = 0 for zero explained variance) were achieved depending on response and type of score. geoGAM combines efficient model building from large sets of covariates with ease of effect interpretation and therefore likely raises the acceptance of DSM products by end-users.

Evaluation of digital soil mapping approaches with large sets of environmental covariates

“…Predictions of log-transformed data were unbiasedly backtransformed according to Cressie (2006, Eq. (20); see also Nussbaum et al, 2017) and for sqrttransformed data we used…”

“…The optimal numbers of boosting iterations m stop and parameters for further model reduction were found by minimizing crossvalidation RMSE. For more details on the model-building procedure, see Nussbaum et al (2017) and the R package geoGAM .…”

“…Non-linear additive modelling through GAM also relies on covariate selection for stable trend estimation. Poggio et al (2013) used a covariate selection procedure with a random component, and Nussbaum et al (2017) applied component-wise gradient boosting to preselect relevant covariates. Unless combined with either lasso or boosting, large sets of covariates are difficult to process through LM, EDK or GAM.…”

Barest Pixel Composite for Agricultural Areas Using Landsat Time Series

“…For sand and silt, we have to be very careful using these prediction models, since there is no clear pattern in the regression line. Figure 10 shows the comparison of the Barest Soil Composite prediction map with the available digital soil map [55,56] for predicted clay values in the area southeast of Greifensee (top) and soil organic matter (SOM) properties in northeast of the Grand Marais (bottom). For canton Zurich, the texture classes were available of the conventional soil map [53].…”

“…For the canton of Zurich, we had a 1:5000 conventional soil map available for the agricultural area [53], and, based on the texture triangle used for this map [54], we were able to classify the polygons in five different clay classes. Additionally, for the area around the Grand Marais and the area around Zurich, we had digital soil maps available of the topsoil properties clay and soil organic matter (SOM) [55,56]. Based on this information, we did a visual comparison between the soil property prediction based on the Bare Soil Composite and the conventional and digital soil maps.…”