Machine learning for predicting soil classes in three semi-arid landscapes

Brungard, Colby; Boettinger, Janis L.; Duniway, Michael C.; Wills, Skye; Edwards, Thomas C.

doi:10.1016/j.geoderma.2014.09.019

Cited by 298 publications

(196 citation statements)

References 51 publications

Supporting

Mentioning

182

Contrasting

Unclassified

Order By: Relevance

“…This increase was clearly within normal fluctuations of RMSE by bagging and random covariate selection (Spiess, 2016). Brungard et al (2015) even improved the prediction accuracy of RF by recursive covariate elimination.…”

Section: Model Buildingmentioning

confidence: 98%

“…RF (Breiman, 2001), another method of balancing the instability of CART, averages a committee of fully grown trees. Two mechanisms are used to de-correlate trees and consequently reduce the variance in the predictions: (1) bootstrap sampling (bagging) creates a different response vector for each tree, and (2) at each node only m try < p randomly selected covariates are tested as candidates for binary splitting.…”

Section: Random Forest (Rf)mentioning

confidence: 99%

“…Geodata availability and deemed importance often determine what covariates are used for DSM. However, Brungard et al (2015) showed that a priori preselection of covariates using pedological expertise might result in a decreased accuracy of soil class predictions. Using comprehensive environmental geodata for DSM improves prediction accuracy because soil-forming factors are likely better represented by a larger number of covariates.…”

Section: Introductionmentioning

confidence: 99%

“…Unless combined with either lasso or boosting, large sets of covariates are difficult to process through LM, EDK or GAM. But these methods allow for simple model interpretation with partial effects plots (Faraway, 2005, p. 73) Generally, more complex approaches seem to yield more accurate predictions than simpler DSM methods (Liess et al, 2012;Brungard et al, 2015). The effects of interactions between covariates on responses can be modelled by tree-based methods, but single trees (CART; e.g.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

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

Nussbaum¹,

Spiess²,

Baltensweiler³

et al. 2018

SOIL

215

View full text Add to dashboard Cite

Abstract. The spatial assessment of soil functions requires maps of basic soil properties. Unfortunately, these are either missing for many regions or are not available at the desired spatial resolution or down to the required soil depth. The field-based generation of large soil datasets and conventional soil maps remains costly. Meanwhile, legacy soil data and comprehensive sets of spatial environmental data are available for many regions.Digital soil mapping (DSM) approaches relating soil data (responses) to environmental data (covariates) face the challenge of building statistical models from large sets of covariates originating, for example, from airborne imaging spectroscopy or multi-scale terrain analysis. We evaluated six approaches for DSM in three study regions in Switzerland (Berne, Greifensee, ZH forest) by mapping the effective soil depth available to plants (SD), pH, soil organic matter (SOM), effective cation exchange capacity (ECEC), clay, silt, gravel content and fine fraction bulk density for four soil depths (totalling 48 responses). Models were built from 300-500 environmental covariates by selecting linear models through (1) grouped lasso and (2) an ad hoc stepwise procedure for robust external-drift kriging (georob). For (3) geoadditive models we selected penalized smoothing spline terms by component-wise gradient boosting (geoGAM). We further used two tree-based methods: (4) boosted regression trees (BRTs) and (5) random forest (RF). Lastly, we computed (6) weighted model averages (MAs) from the predictions obtained from methods 1-5.Lasso, georob and geoGAM successfully selected strongly reduced sets of covariates (subsets of 3-6 % of all covariates). Differences in predictive performance, tested on independent validation data, were mostly small and did not reveal a single best method for 48 responses. Nevertheless, RF was often the best among methods 1-5 (28 of 48 responses), but was outcompeted by MA for 14 of these 28 responses. RF tended to over-fit the data. The performance of BRT was slightly worse than RF. GeoGAM performed poorly on some responses and was the best only for 7 of 48 responses. The prediction accuracy of lasso was intermediate. All models generally had small bias. Only the computationally very efficient lasso had slightly larger bias because it tended to under-fit the data. Summarizing, although differences were small, the frequencies of the best and worst performance clearly favoured RF if a single method is applied and MA if multiple prediction models can be developed.Published by Copernicus Publications on behalf of the European Geosciences Union.

show abstract

Section: Model Buildingmentioning

confidence: 98%

Section: Random Forest (Rf)mentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

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

Nussbaum¹,

Spiess²,

Baltensweiler³

et al. 2018

SOIL

215

View full text Add to dashboard Cite

show abstract

“…Random forests models have been applied to modeling the presence and absence of invasive plants and rare lichens and outperform other models (classification trees, logistic regression, linear discriminant analysis) in several measures of accuracy (Cutler et al, 2007). In addition, random forests models have been applied to predict soil classes in semiarid environments and perform as well as or better than other model types in all cases (Brungard et al, 2015). Random forests, developed for ecological applications by Breiman and Culter (Breiman, 2001;Cutler et al, 2007;Breiman and Cutler, 2009), is a machine learning algorithm using decision trees to identify or predict classes of data.…”

mentioning

confidence: 99%

Modeling Rare Endemic Shrub Habitat in the Uinta Basin Using Soil, Spectral, and Topographic Data

Baker¹,

Fonnesbeck

Boettinger

2016

Soil Science Society of America Journal

Self Cite

View full text Add to dashboard Cite

Conserving rare plants is important to biodiversity. Shrubby reed-mustard [Schoenocrambe suffrutescens (Rollins) S.L. Welsh and Chatterly] (SRM), a shrub endemic to the Uinta Basin, faces habitat loss due to energy development of fossil fuels. Little is known about the soil and site properties required for successful establishment, growth, and survival of SRM. Our goal was to create random forests models to identify potential SRM habitat and to verify habitat suitability in the field with easily measured properties. The soil properties (SP) model indicated that CaCO 3 equivalent, silt, and dry color value predicted the presence or absence of SRM with 10% out-of-bag (OOB) error. Soil properties were correlated with Landsat 5 Thematic Mapper spectral data and a digital elevation model in a Soil-Spectral Correlation (SSC) model. Yellowness, 3/2 normalized difference ratio (NDR), and 3/1 NDR were most strongly correlated with CaCO 3 equivalent and silt, and they predicted SRM habitat with 28% OOB error. guided by SSC model output, 250 additional field presence or absence points were used to train the Spectral-Topographic (ST) model, which was validated by an independent set of SRM plant locations. The ST model using area solar radiation, 7/2 NDR, 5/2 NDR, and the normalized difference vegetation index gave an OOB error of 23%. The ST model can be used to identify potential habitat across a large area. Once identified, easily measured soil and site data from the SP model can verify SRM habitat suitability. These models can help land managers locate and check key soil properties at potential SRM habitat sites.Abbreviations: BLM, Bureau of Land Management; DEM, digital elevation model; NDR, normalized difference ratio; NDVI, normalized difference vegetation index; OOB, outof-bag; SP, soil properties; SRM, shrubby reed-mustard; SSC, soil-spectral correlation; ST, spectral-topographic. P reserving populations of rare plants is important to maintain biodiversity, which increases genetic diversity, provides possible sources of food and medicinal plants, and provides ecosystem services such as protection of soil and water resources, nutrient cycling, and pollutant adsorption or uptake (Australian Department of the Environment, 1993). A species may be rare for a variety of reasons, including low tolerance to environmental change, preference for a very specific habitat, limited dispersal capabilities, etc. Endemic plants are often characterized by few populations concentrated in a narrow geographic distribution. Soil and geologic properties have been successfully used to explain the distribution of some rare endemic plants. For example, endemic species grow in soils formed on serpentinite with low Ca/Mg ratios and high levels of heavy metals (Cr, Co, Ni) inherited from the ultramafic parent material (Alexander et al., 2007;Lazarus et al., 2011;Proctor, 1971;van der Ent and Reeves, 2015 Core Ideas• Soil and site properties predict rare plant distribution.• Rare plant habitat can be predicted by random forests modeling of data...

show abstract

Soil characteristics are associated with gradients of big sagebrush canopy structure after disturbance

et al. 2019

Self Cite

View full text Add to dashboard Cite

Reestablishing shrub canopy cover after disturbance in semi‐arid ecosystems, such as sagebrush steppe, is essential to provide wildlife habitat and restore ecosystem functioning. While several studies have explored the effects of landscape and climate factors on the success or failure of sagebrush seeding, the influence of soil properties on gradients of shrub canopy structure in successfully seeded areas remains largely unexplored. In this study, we evaluated associations between soil properties and gradients in sagebrush canopy structure in stands that had successfully reestablished after fire and subsequent seeding treatments. Using a dataset collected across the Great Basin, USA, of sagebrush stands that had burned and reestablished between 1986 and 2013, we tested soil depth and texture, soil surface classification, biological soil crusts plus mean historical precipitation, solar heatload, and fire history as modeling variables to explore gradients in sagebrush canopy structure growth in terms of cover, height, and density. Deeper soils were associated with greater sagebrush canopy structure development in terms of plant density and percent cover, coarser textured soils were associated with greater sagebrush cover and density, and more clayey soils were typically associated with greater height. Biological crust presence was also positively associated with enhanced sagebrush canopy growth, but adding more demographically or morphologically explicit descriptions of biocrust communities did not improve explanatory power. Increasing heatload had a negative effect on sagebrush canopy structure growth, and increased mean annual precipitation was only associated with greater sagebrush height. Given that conservation and restoration of the sagebrush steppe ecosystems has become a priority for land managers, the associations we identify between gradients in post‐fire sagebrush canopy structure growth and field‐identifiable soil characteristics may improve planning of land treatments for sagebrush restoration and the understanding of semi‐arid ecosystem functioning and post‐disturbance dynamics.

show abstract

Machine learning for predicting soil classes in three semi-arid landscapes

Cited by 298 publications

References 51 publications

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

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

Modeling Rare Endemic Shrub Habitat in the Uinta Basin Using Soil, Spectral, and Topographic Data

Soil characteristics are associated with gradients of big sagebrush canopy structure after disturbance

Contact Info

Product

Resources

About