Evaluation of modelling approaches for predicting the spatial distribution of soil organic carbon stocks at the national scale

Abstract

Soil organic carbon (SOC) plays a major role in the global carbon budget. It can act as a source or a sink of atmospheric carbon, thereby possibly influencing the course of climate change. Improving the tools that model the spatial distributions of SOC stocks at national scales is a priority, both for monitoring changes in SOC and as an input for global carbon cycles studies. In this paper, we compare and evaluate two recent and promising modelling approaches. First, we considered several increasingly complex boosted regression trees (BRT), a convenient and efficient multiple regression model from the statistical learning field. Further, we considered a robust geostatistical approach coupled to the BRT models. Testing the different approaches was performed on the dataset from the French Soil Monitoring Network, with a consistent cross-validation procedure. We showed that when a limited number of predictors were included in the BRT model, the standalone BRT predictions were significantly improved by robust geostatistical modelling of the residuals. However, when data for several SOC drivers were included, the standalone BRT model predictions were not significantly improved by geostatistical modelling. Therefore, in this latter situation, the BRT predictions might be considered adequate without the need for geostatistical modelling, provided that i) care is exercised in model fitting and validating, and ii) the dataset does not allow for modelling of local spatial autocorrelations, as is the case for many national systematic sampling schemes.

Spatial variability of potassium in agricultural soils of the canton of Fribourg, Switzerland

“…This is to be expected due to the fact that both MLR and RF predictive maps are based on predictive values from models that are extrapolated to the grid, as opposed to OK, which uses the actual values for interpolation. Due to model predictions values from both MLR and RF lower values are overestimated and higher values underestimated, as observed by other researchers (Martin et al, 2014) which results overall in smaller ranges for both predicted values and 95% CI values. Increasing the sample size to capture the variability in our study area would have allowed for better predictions, however, this was not possible due to the low density of points in our study area (1 per 6.8 km 2 ).…”

Understanding large‐extent controls of soil organic carbon storage in relation to soil depth and soil‐landscape systems

“…Compared to previous work, our RT models resulted in similar accuracies (topsoil R 2 CV 0.52 and R 2 0.39 with external validation); e.g., Martin et al . [], using only exhaustive data sets, reported accuracies of R 2 CV 0.36 for modeled SOC stocks in France, and Lacoste et al [] reported accuracies up to R 2 0.43 (validation using the testing data set) for modeling the SOC stocks in the O horizon of French forest soils. Also, our findings show that RTs were capable to predict SOC at different depths but were limited in explaining the spatial variability in SOC which results from the complex biological, chemical, and physical processes.…”

“…Previously, research focused on estimating SOC levels and SOC stocks over vast areas using various approaches, including extrapolation of stocks based on soil profile data [ Batjes , , ; Mishra and Riley , ; Yu et al , ] or the estimation of stocks using proxies from, e.g., remote sensing data [ Chen et al , ; Stevens et al , ]. Others aimed at determining the controlling factors for topsoil SOC and modeling the spatial distribution at various scales [ de Brogniez et al , ; Jarmer et al , ; Martin et al , ; Meersmans et al , ; Minasny et al , ; Ungaro et al , ; Vasques et al , ]. Contrary to the topsoil SOC, the large subsoil SOC reservoir has received less attention and is, currently, not well understood [ Fontaine et al , ; Henderson et al , ; Jandl et al , ; Kempen et al , ; Rumpel and Kogel‐Knabner , ].…”

Future C loss in mid-latitude mineral soils: climate change exceeds land use mitigation potential in France

“…Nevertheless, recent research1819 as well as international treaties, such as the Kyoto Protocol20 and EU Soil Thematic Strategy21, underline the clear need for more accurate spatial and temporal explicit estimates of this pool in order to establish appropriate policy measures to combat climate change, land degradation and soil fertility decline threats. Although a significant attempt has been made to produce detailed maps of SOC at the national level, these studies mainly focused on current22232425 or past trends262728 and the potential influence of climate change on past SOC trends remains debatable2729. Since many scenarios clearly suggest that the contribution and amplitude of climate change will become increasingly stronger over the next decades, there exists a clear need to estimate the impact of these rapidly changing conditions on large scale SOC budgets.…”