Spatial distribution of soil organic carbon stocks in France

Abstract

Soil organic carbon plays a major role in the global carbon budget, and can act as a source or a sink of atmospheric carbon, whereby it can influence the course of climate change. Changes in soil organic soil stocks (SOCS) are now taken into account in international negotiations regarding climate change. Consequently, developing sampling schemes and models for estimating the spatial distribution of SOCS is a priority. The French soil monitoring network has been established on a 16 km × 16 km grid and the first sampling campaign has recently been completed, providing circa 2200 measurements of stocks of soil organic carbon, obtained through an in situ composite sampling, uniformly distributed over the French territory. <br><br> We calibrated a boosted regression tree model on the observed stocks, modelling SOCS as a function of other variables such as climatic parameters, vegetation net primary productivity, soil properties and land use. The calibrated model was evaluated through cross-validation and eventually used for estimating SOCS for the whole of metropolitan France. Two other models were calibrated on forest and agricultural soils separately, in order to assess more precisely the influence of pedo-climatic variables on soil organic carbon for such soils. <br><br> The boosted regression tree model showed good predictive ability, and enabled quantification of relationships between SOCS and pedo-climatic variables (plus their interactions) over the French territory. These relationship strongly depended on the land use, and more specifically differed between forest soils and cultivated soil. The total estimate of SOCS in France was 3.260 ± 0.872 PgC for the first 30 cm. It was compared to another estimate, based on the previously published European soil organic carbon and bulk density maps, of 5.303 PgC. We demonstrate that the present estimate might better represent the actual SOCS distributions of France, and consequently that the previously published approach at the European level greatly overestimates SOCS

Controls on fallen leaf chemistry and forest floor element masses in native and novel forests across a tropical island

“…Because spatial variation in forest soil properties is notoriously high (cf. Conant et al 2003), often requiring a logistically prohibitive number of samples, subsampling followed by compositing (bulking) is a common technique for soil analyses (e.g., Martin et al 2011). Despite the high variation for forest floor masses, the range of values is within published values (cf.…”

Patterns and Predictors of Recent Forest Conversion in New England

“…BRTs generally avoid overfitting [35], can be applied to a variety of spatial analyses, such as the distribution of species and vegetation types [36][37][38][39][40][41][42], hydrology [43,44], soil and landform properties [45][46][47] and natural disturbance [48], as well as quantification of land cover and land use change through human activities [34,49,50]. Across a wide variety of contexts, model comparisons have shown BRTs to perform much better than traditional models and comparably well to other machine-learning models [36,37,43,45,48,51], with some variability in comparative performance with other machine-learning methods depending on context [33,38,44,47].…”

No silver bullet for digital soil mapping: country-specific soil organic carbon estimates across Latin America

“…The combination of regression modeling approaches with geostatistics of independent model residuals (i.e., regression kriging) is a combined strategy that has been widely used to map SOC (Hengl et al, 2004;Mishra et al, 2009;Marchetti et al, 2012;Kumar et al, 2012;Peng et al, 2013;Adhikari et al, 2014;Yigini and Panagos, 2016;Nussbaum et al, 2014;Mondal et al, 2017). Machine learning algorithms such as random forests or support vector machines have also been used to increase statistical accuracy of soil carbon models (Martin et al, 2011;Hashimoto et al, 2017;Hengl et al, 2017) including applications for SOC mapping (Grimm et al, 2008;Sreenivas et al, 2016;Yang et al, 2016;Hengl et al, 2017;Delgado-Baquerizo et al, 2017;Ließ et al, 2016;Viscarra Rossel et al, 2014). Machine learning methods do not necessarily allow to extract information about the main effects of prediction factors in the response variable (e.g., SOC); consequently, a variable selection strategy is always useful to increase the interpretability of machine learning algorithms.…”