Mareike Ließ

Improving the Spatial Prediction of Soil Organic Carbon Stocks in a Complex Tropical Mountain Landscape by Methodological Specifications in Machine Learning Approaches

Tropical forests are significant carbon sinks and their soils’ carbon storage potential is immense. However, little is known about the soil organic carbon (SOC) stocks of tropical mountain areas whose complex soil-landscape and difficult accessibility pose a challenge to spatial analysis. The choice of methodology for spatial prediction is of high importance to improve the expected poor model results in case of low predictor-response correlations. Four aspects were considered to improve model performance in predicting SOC stocks of the organic layer of a tropical mountain forest landscape: Different spatial predictor settings, predictor selection strategies, various machine learning algorithms and model tuning. Five machine learning algorithms: random forests, artificial neural networks, multivariate adaptive regression splines, boosted regression trees and support vector machines were trained and tuned to predict SOC stocks from predictors derived from a digital elevation model and satellite image. Topographical predictors were calculated with a GIS search radius of 45 to 615 m. Finally, three predictor selection strategies were applied to the total set of 236 predictors. All machine learning algorithms—including the model tuning and predictor selection—were compared via five repetitions of a tenfold cross-validation. The boosted regression tree algorithm resulted in the overall best model. SOC stocks ranged between 0.2 to 17.7 kg m-2, displaying a huge variability with diffuse insolation and curvatures of different scale guiding the spatial pattern. Predictor selection and model tuning improved the models’ predictive performance in all five machine learning algorithms. The rather low number of selected predictors favours forward compared to backward selection procedures. Choosing predictors due to their indiviual performance was vanquished by the two procedures which accounted for predictor interaction.

Comparison of Three Supervised Learning Methods for Digital Soil Mapping: Application to a Complex Terrain in the Ecuadorian Andes

A digital soil mapping approach is applied to a complex, mountainous terrain in the Ecuadorian Andes. Relief features are derived from a digital elevation model and used as predictors for topsoil texture classes sand, silt, and clay. The performance of three statistical learning methods is compared: linear regression, random forest, and stochastic gradient boosting of regression trees. In linear regression, a stepwise backward variable selection procedure is applied and overfitting is controlled by minimizing Mallow’s Cp. For random forest and boosting, the effect of predictor selection and tuning procedures is assessed. 100-fold repetitions of a 5-fold cross-validation of the selected modelling procedures are employed for validation, uncertainty assessment, and method comparison. Absolute assessment of model performance is achieved by comparing the prediction error of the selected method and the mean. Boosting performs best, providing predictions that are reliably better than the mean. The median reduction of the root mean square error is around 5%. Elevation is the most important predictor. All models clearly distinguish ridges and slopes. The predicted texture patterns are interpreted as result of catena sequences (eluviation of fine particles on slope shoulders) and landslides (mixing up mineral soil horizons on slopes).

The Sloping Mire Soil-Landscape of Southern Ecuador - Influence of predictor resolution and model tuning on random forest predictions

The sloping mire landscape of the investigation area, in the southern Andes of Ecuador, is dominated by stagnic soils with thick organic layers. The recursive partitioning algorithm Random Forest was used to predict the spatial water stagnation pattern and the thickness of the organic layer from terrain attributes. Terrain smoothing from 10 to 30 m raster resolution was applied in order to obtain the best possible model. For the same purpose, several model tuning parameters were tested and a prepredictor selection with the R-package Boruta was applied. Model versions were evaluated and compared by 100 repetitions of the calculation of the residual mean square error of a five-fold cross-validation. Position specific density functions of the predicted soil parameters were then used to display prediction uncertainty. Prepredictor selection and tuning of the Random Forest algorithm in some cases resulted in an improved model performance. We therefore recommend testing prepredictor selection and tuning to make sure that the best possible model is chosen. This needs particular emphasis in complex tropical mountain soil-landscapes which provide a real challenge to any soil mapping approach but where Random Forest has proven to be successful due to the testing of model tuning and prepredictor selection.

Environmental drivers of spatial patterns of topsoil nitrogen and phosphorus under monsoon conditions in a complex terrain of South Korea

Nitrogen (N) and phosphorus (P) in topsoils are critical for plant nutrition. Relatively little is known about the spatial patterns of N and P in the organic layer of mountainous landscapes. Therefore, the spatial distributions of N and P in both the organic layer and the A horizon were analyzed using a light detection and ranging (LiDAR) digital elevation model and vegetation metrics. The objective of the study was to analyze the effect of vegetation and topography on the spatial patterns of N and P in a small watershed covered by forest in South Korea. Soil samples were collected using the conditioned latin hypercube method. LiDAR vegetation metrics, the normalized difference vegetation index (NDVI), and terrain parameters were derived as predictors. Spatial explicit predictions of N/P ratios were obtained using a random forest with uncertainty analysis. We tested different strategies of model validation (repeated 2-fold to 20-fold and leave-one-out cross validation). Repeated 10-fold cross validation was selected for model validation due to the comparatively high accuracy and low variance of prediction. Surface curvature was the best predictor of P contents in the organic layer and in the A horizon, while LiDAR vegetation metrics and NDVI were important predictors of N in the organic layer. N/P ratios increased with surface curvature and were higher on the convex upper slope than on the concave lower slope. This was due to P enrichment of the soil on the lower slope and a more even spatial distribution of N. Our digital soil maps showed that the topsoils on the upper slopes contained relatively little P. These findings are critical for understanding N and P dynamics in mountainous ecosystems.

Natural Landslides Which Impact Current Regulating Services: Environmental Preconditions and Modeling

Recurrent landslide activity in the natural mountain forest is assumed to be a major factor for maintaining its high biodiversity. It is hypothesized that abiotic–biotic interactions are a prerequisite for natural landslides. A statistical model solely driven by topographic predictors can explain areas prone to landslides but also shows that other factors (e.g., geology, soil, climate, vegetation) than topography might play an important role to improve model performance. Thus, the chapter also shows approaches to derive spatial information on soil properties and wind stress as potential driving predictors for the model. Furthermore, it can be shown that even changes in the biogeochemical cycle and the regulation between nutrient input and biomass production might influence the risk of landslides.