Optimizing Pedotransfer Functions for Estimating Soil Bulk Density Using Boosted Regression Trees

Martín, Manuel; Seen, Danny Lo; Boulonne, Line; Jolivet, Claudy; Nair, K. M.; Bourgeon, Gérard; Arrouays, Dominique

doi:10.2136/sssaj2007.0241

Cited by 100 publications

(91 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Pedotransfer functions are often applied instead to predict soil BD on the basis of SOC or soil organic matter content and soil texture data (Arrouays et al, 2012). It has been shown that most pedotransfer functions are suitable only for the agro-pedo-climatic conditions prevailing at the sites used to fit these functions (Martin et al, 2009). Under different conditions, they lead to substantial systematic errors (De Vos et al, 2005;Nanko et al, 2014;Vasiliniuc and Patriche, 2015).…”

mentioning

confidence: 99%

Determining Soil Bulk Density for Carbon Stock Calculations: A Systematic Method Comparison

Walter¹,

Don²,

Tiemeyer³

et al. 2016

Soil Science Soc of Amer J

142

View full text Add to dashboard Cite

Accurate and effective determination of soil bulk density (BD) is needed to monitor soil organic C (SOC) stocks and SOC stock changes. However, BD measurements are often lacking in soil inventories and BD is estimated by pedotransfer functions with substantial uncertainty. In a systematic method comparison, we evaluated different methods for BD determination in the field by comparing the performance of MINI (5 cm 3 ) and BIG (250 cm 3 ) sample rings and of three driving hammer probes differing in diameter, material, and extraction method. Bulk density determined with 100-cm 3 sample rings was defined as the reference method (REF). All methods were tested at five depth increments in nine subplots at four sites with differing soil texture and SOC content. All methods determined BD in the depth increments with low systematic error (8% for probes and 2% for sample rings). The random error of the probe samples was, on average, 50% higher than that of the ring samples when the cores of the probes were adequately corrected for compaction or stretching. The BD was significantly overestimated (by 2%) when determined with MINI rings, and the variation in BD was not reduced with BIG sample rings rather than the smaller REF sample rings. The performance of the driving hammer technique varied widely among probe types and sites. The sheath probe had the smallest systematic error of all probes tested and is recommended for soil inventories. All methods for estimating BD had smaller errors than pedotransfer functions.Abbreviations: BD, bulk density; CV, coefficient of variation; MPE, mean prediction error; SDPE, standard deviation of the prediction error; SOC, soil organic carbon. C arbon storage in soils exceeds that in vegetation and the atmosphere (Ciais et al., 2013). Thus, small changes in soil organic C (SOC) stocks could have severe impacts on the global C cycle. Reliable measurements of C concentration are an important prerequisite for detecting such small changes in SOC stocks (Goidts et al., 2009). Information on soil bulk density (BD) is essential in converting weight-based concentration data to volume-or area-based stock data. However, BD is a parameter that is only partly or never sampled in many soil inventories (Gruneberg et al., 2014;Reynolds et al., 2013;Saby et al., 2008). Pedotransfer functions are often applied instead to predict soil BD on the basis of SOC or soil organic matter content and soil texture data (Arrouays et al., 2012). It has been shown that most pedotransfer functions are suitable only for the agro-pedo-climatic conditions prevailing at the sites used to fit these functions (Martin et al., 2009). Under different conditions, they lead to substantial systematic errors (De Vos et al., 2005;Nanko et al., 2014;Vasiliniuc and Patriche, 2015). For a soil with a BD of 1.4 g cm −3 , a systematic measurement error of −0.01 to −0.51 g cm −3 (De Vos et al., 2005) would result in SOC stocks being underestimated by 1 to 36%. Katja Core Ideas• Little is known about the methodological errors of bulk densit...

show abstract

mentioning

confidence: 99%

Determining Soil Bulk Density for Carbon Stock Calculations: A Systematic Method Comparison

Walter¹,

Don²,

Tiemeyer³

et al. 2016

Soil Science Soc of Amer J

142

View full text Add to dashboard Cite

show abstract

“…Recently, these principals have been applied to the prediction of both D b (Jalabert et al, 2010;Martin et al, 2009) and organic carbon stock (Wiesmeier et al, 2011;Grimm et al, 2008) at the point scale with considerable success. Methods commonly used to explicitly include landscape attributes in the modelling process are artificial neural networks (ANNs) (Keshavarzi et al, 2010) and random forests (RFs) (Prasad et al, 2006).…”

Section: K P Taalab Et Al: Modelling Soil Bulk Density At the Landmentioning

confidence: 99%

“…Many studies differentiate between topsoil and subsoil by depth (De Vos et al, 2005;Katterer et al, 2006). However, the lower depth of the topsoil layer can vary from 15 cm (Bellamy et al, 2005) to 30 cm (Martin et al, 2009). The data used in this study were sampled by horizon, meaning that there was not a uniform sampling depth between points and the number of samples taken at a given location was dependent on soil morphology.…”

Section: Data Preprocessingmentioning

confidence: 99%

“…Indeed, tree-based models have been used to successfully predict D b using a mix of landscape data and soil data (Martin et al 2009). In terms of model performance, RF achieved similar results to a number of other studies, all of which have used textural properties as predictors (Tranter et al, 2007;De Vos et al, 2005;Heuscher et al, 2005).…”

Section: Model Performancementioning

confidence: 99%

“…The importance of putting D b in a landscape context is supported by the successful stratification of previous regression models by land use (Steller et al, 2008;Moreira et al, 2009) and parent material (Hallett et al, 1998;Calhoun et al, 2001). However, these factors have been explicitly included in the modelling process only relatively recently (Martin et al, 2009;Jalabert et al, 2010). Of the landscape variables included, land use, parent material and soil classification are deemed to be consistently important predictors.…”

Section: Variable Importancementioning

confidence: 99%

See 2 more Smart Citations

Modelling soil bulk density at the landscape scale and its contributions to C stock uncertainty

et al. 2013

View full text Add to dashboard Cite

Soil bulk density (D_b) is a major contributor to uncertainties in landscape-scale carbon and nutrient stock estimation. However, it is time consuming to measure and is, therefore, frequently predicted using surrogate variables, such as soil texture. Using this approach is of limited value for estimating landscape-scale inventories, as its accuracy beyond the sampling point at which texture is measured becomes highly uncertain. In this paper, we explore the ability of soil landscape models to predict soil D_b using a suite of landscape attributes and derivatives for both topsoil and subsoil. The models were constructed using random forests and artificial neural networks.

Using these statistical methods, we have produced a spatially distributed prediction of D_b on a 100 m × 100 m grid, which was shown to significantly improve topsoil carbon stock estimation. In comparison to using mean values from point measurements, stratified by soil class, we found that the gridded method predicted D_b more accurately, especially for higher and lower values within the range. Within our study area of the Midlands, UK, we found that the gridded prediction of D_b produced a stock inventory of over 1 million tonnes of carbon greater than the stratified mean method. Furthermore, the 95% confidence interval associated with total C stock prediction was almost halved by using the gridded method. The gridded approach was particularly useful in improving organic carbon (OC) stock estimation for fine-scale landscape units at which many landscape–atmosphere interaction models operate

show abstract

Large Differences in Global and Regional Total Soil Carbon Stock Estimates Based on SoilGrids, HWSD, and NCSCD: Intercomparison and Evaluation Based on Field Data From USA, England, Wales, and France

Tifafi

Guenet

Hatté

2018

Global Biogeochemical Cycles

147

View full text Add to dashboard Cite

Soils are the major component of the terrestrial ecosystem and the largest organic carbon reservoir on Earth. However, they are a nonrenewable natural resource and especially reactive to human disturbance and climate change. Despite its importance, soil carbon dynamics is an important source of uncertainty for future climate predictions and there is a growing need for more precise information to better understand the mechanisms controlling soil carbon dynamics and better constrain Earth system models. The aim of our work is to compare soil organic carbon stocks given by different global and regional databases that already exist. We calculated global and regional soil carbon stocks at 1 m depth given by three existing databases (SoilGrids, the Harmonized World Soil Database, and the Northern Circumpolar Soil Carbon Database). We observed that total stocks predicted by each product differ greatly: it is estimated to be around 3,400 Pg by SoilGrids and is about 2,500 Pg according to Harmonized World Soil Database. This difference is marked in particular for boreal regions where differences can be related to high disparities in soil organic carbon concentration. Differences in other regions are more limited and may be related to differences in bulk density estimates. Finally, evaluation of the three data sets versus ground truth data shows that (i) there is a significant difference in spatial patterns between ground truth data and compared data sets and that (ii) data sets underestimate by more than 40% the soil organic carbon stock compared to field data.

show abstract

Optimizing Pedotransfer Functions for Estimating Soil Bulk Density Using Boosted Regression Trees

Cited by 100 publications

References 34 publications

Determining Soil Bulk Density for Carbon Stock Calculations: A Systematic Method Comparison

Determining Soil Bulk Density for Carbon Stock Calculations: A Systematic Method Comparison

Modelling soil bulk density at the landscape scale and its contributions to C stock uncertainty

Large Differences in Global and Regional Total Soil Carbon Stock Estimates Based on SoilGrids, HWSD, and NCSCD: Intercomparison and Evaluation Based on Field Data From USA, England, Wales, and France

Contact Info

Product

Resources

About