Gerard B. M. Heuvelink

SoilGrids1km — Global Soil Information Based on Automated Mapping

Background Soils are widely recognized as a non-renewable natural resource and as biophysical carbon sinks. As such, there is a growing requirement for global soil information. Although several global soil information systems already exist, these tend to suffer from inconsistencies and limited spatial detail. Methodology/Principal Findings We present SoilGrids1km — a global 3D soil information system at 1 km resolution — containing spatial predictions for a selection of soil properties (at six standard depths): soil organic carbon (g kg−1), soil pH, sand, silt and clay fractions (%), bulk density (kg m−3), cation-exchange capacity (cmol+/kg), coarse fragments (%), soil organic carbon stock (t ha−1), depth to bedrock (cm), World Reference Base soil groups, and USDA Soil Taxonomy suborders. Our predictions are based on global spatial prediction models which we fitted, per soil variable, using a compilation of major international soil profile databases (ca. 110,000 soil profiles), and a selection of ca. 75 global environmental covariates representing soil forming factors. Results of regression modeling indicate that the most useful covariates for modeling soils at the global scale are climatic and biomass indices (based on MODIS images), lithology, and taxonomic mapping units derived from conventional soil survey (Harmonized World Soil Database). Prediction accuracies assessed using 5–fold cross-validation were between 23–51%. Conclusions/Significance SoilGrids1km provide an initial set of examples of soil spatial data for input into global models at a resolution and consistency not previously available. Some of the main limitations of the current version of SoilGrids1km are: (1) weak relationships between soil properties/classes and explanatory variables due to scale mismatches, (2) difficulty to obtain covariates that capture soil forming factors, (3) low sampling density and spatial clustering of soil profile locations. However, as the SoilGrids system is highly automated and flexible, increasingly accurate predictions can be generated as new input data become available. SoilGrids1km are available for download via http://soilgrids.org under a Creative Commons Non Commercial license.

Modelling soil variation: past, present, and future

Abstract The soil mantles the land, except where there is bare rock or ice, and it varies more or less continuously. Many of its properties change continuously in time, too. We can measure the soil at only a finite number of places and times on small supports, and any statement concerning the soil at other places or times involves prediction. Variation in soil is also complex, so complex that no description of it can be complete, and so prediction is inevitably uncertain. Soil scientists should be able to quantify this uncertainty, and manage it. This means representing the variation by models that may be in part deterministic, but cannot be wholly so; they must have some random element to represent the unpredictable variation. Here we review three families of statistically based models of soil variation that are currently in use and trace their development since the mid-1960s. In particular, we consider classification and geostatistics for modelling the spatial variation, time series analysis and physically based approaches for modelling temporal variation, and space–time Kalman filtering for predicting soil conditions in space and time simultaneously. Each of these attaches to its predictions quantitative estimates of the prediction errors. Past, present and future research has been, is, and will be directed to the development of models that diminish these errors. A challenge for the future is to investigate approaches that merge process knowledge with measurements. For soil survey, this would be achieved by integration of pedogenetic knowledge and field observations through the use of data assimilation techniques, such as the space–time Kalman filter.

Spatial aggregation and soil process modelling

Abstract Nonlinear soil process models that are defined and calibrated at the point support cannot at the same time be valid at the block support. This means that in the situation where model input is available at point support and where model output is required at block support, spatial aggregation should take place after the model is run. Although block kriging does both in one pass, it is sensible to separate spatial aggregation from spatial interpolation. Contrary to aggregation, interpolation should better take place before the model is run because this enables a more efficient use of the spatial distribution characteristics of individual inputs. When a model is run with interpolated inputs, it is important not to ignore the interpolation error. Substituting conditional expectations in place of probability distributions into a nonlinear model leads to bias, essentially for the same reason that aggregating inputs prior to running a model is not the same as aggregating the output after the model is run. Running a model with inputs that are probability distributions will usually call for a Monte Carlo simulation approach. This causes a substantial increase in the numerical load, but apart from eliminating bias, an important advantage is that it shows how uncertainties in model inputs propagate to the model output. Many models used in soil science suffer not only from input error but also from model error, which is support- and case-dependent. Case dependency implies that model error can only be assessed realistically through validation. A major problem in validation is that the validation data are often collected at a much smaller support than the aggregated model predictions.

Using simulated annealing for resource allocation

Many resource allocation issues, such as land use- or irrigation planning, require input from extensive spatial databases and involve complex decisionmaking problems. Spatial decision support systems (SDSS) are designed to make these issues more transparent and to support the design and evaluation of resource allocation alternatives. Recent developments in this field focus on the design of allocation plans that utilise mathematical optimisation techniques. These techniques, often referred to as multi-criteria decision-making (MCDM) techniques, run into numerical problems when faced with the high dimensionality encountered in spatial applications. In this paper we demonstrate how simulated annealing, a heuristic algorithm, can be used to solve high-dimensional non-linear optimisation problems for multi-site land use allocation (MLUA) problems. The optimisation model both minimises development costs and maximises spatial compactness of the land use. Compactness is achieved by adding a non-linear neighbourhood objective to the objective function. The method is successfully applied to a case study in Galicia, Spain, using an SDSS for supporting the restoration of a former mining area with new land use.

Real-time automatic interpolation of ambient gamma dose rates from the Dutch radioactivity monitoring network

Detection of radiological accidents and monitoring the spread of the contamination is of great importance. Following the Chernobyl accident many European countries have installed monitoring networks to perform this task. Real-time availability of automatically interpolated maps showing the spread of radioactivity during and after an accident would improve the capability of decision makers to accurately respond to a radiological accident. The objective of this paper is to present a real-time automatic interpolation system suited for natural background radioactivity. Interpolating natural background radiation allows us to better understand the natural variability, thus improving our ability to detect accidents. A real-time automatic interpolation system suited for natural background radioactivity presents a first step towards a system that can deal with radiological accidents. The interpolated maps are produced using a combination of universal kriging and an automatic variogram fitting procedure. The system provides a map of (1) the kriging prediction, (2) the kriging standard error and (3) the position of approximate prediction intervals relative to a threshold. The maps are presented through a Web Map Service (WMS) to ensure interoperability with existing Geographic Information Systems (GIS).

SoilGrids250m: Global gridded soil information based on machine learning

This paper describes the technical development and accuracy assessment of the most recent and improved version of the SoilGrids system at 250m resolution (June 2016 update). SoilGrids provides global predictions for standard numeric soil properties (organic carbon, bulk density, Cation Exchange Capacity (CEC), pH, soil texture fractions and coarse fragments) at seven standard depths (0, 5, 15, 30, 60, 100 and 200 cm), in addition to predictions of depth to bedrock and distribution of soil classes based on the World Reference Base (WRB) and USDA classification systems (ca. 280 raster layers in total). Predictions were based on ca. 150,000 soil profiles used for training and a stack of 158 remote sensing-based soil covariates (primarily derived from MODIS land products, SRTM DEM derivatives, climatic images and global landform and lithology maps), which were used to fit an ensemble of machine learning methods—random forest and gradient boosting and/or multinomial logistic regression—as implemented in the R packages ranger, xgboost, nnet and caret. The results of 10–fold cross-validation show that the ensemble models explain between 56% (coarse fragments) and 83% (pH) of variation with an overall average of 61%. Improvements in the relative accuracy considering the amount of variation explained, in comparison to the previous version of SoilGrids at 1 km spatial resolution, range from 60 to 230%. Improvements can be attributed to: (1) the use of machine learning instead of linear regression, (2) to considerable investments in preparing finer resolution covariate layers and (3) to insertion of additional soil profiles. Further development of SoilGrids could include refinement of methods to incorporate input uncertainties and derivation of posterior probability distributions (per pixel), and further automation of spatial modeling so that soil maps can be generated for potentially hundreds of soil variables. Another area of future research is the development of methods for multiscale merging of SoilGrids predictions with local and/or national gridded soil products (e.g. up to 50 m spatial resolution) so that increasingly more accurate, complete and consistent global soil information can be produced. SoilGrids are available under the Open Data Base License.

Mapping Soil Properties of Africa at 250 m Resolution: Random Forests Significantly Improve Current Predictions

80% of arable land in Africa has low soil fertility and suffers from physical soil problems. Additionally, significant amounts of nutrients are lost every year due to unsustainable soil management practices. This is partially the result of insufficient use of soil management knowledge. To help bridge the soil information gap in Africa, the Africa Soil Information Service (AfSIS) project was established in 2008. Over the period 2008–2014, the AfSIS project compiled two point data sets: the Africa Soil Profiles (legacy) database and the AfSIS Sentinel Site database. These data sets contain over 28 thousand sampling locations and represent the most comprehensive soil sample data sets of the African continent to date. Utilizing these point data sets in combination with a large number of covariates, we have generated a series of spatial predictions of soil properties relevant to the agricultural management—organic carbon, pH, sand, silt and clay fractions, bulk density, cation-exchange capacity, total nitrogen, exchangeable acidity, Al content and exchangeable bases (Ca, K, Mg, Na). We specifically investigate differences between two predictive approaches: random forests and linear regression. Results of 5-fold cross-validation demonstrate that the random forests algorithm consistently outperforms the linear regression algorithm, with average decreases of 15–75% in Root Mean Squared Error (RMSE) across soil properties and depths. Fitting and running random forests models takes an order of magnitude more time and the modelling success is sensitive to artifacts in the input data, but as long as quality-controlled point data are provided, an increase in soil mapping accuracy can be expected. Results also indicate that globally predicted soil classes (USDA Soil Taxonomy, especially Alfisols and Mollisols) help improve continental scale soil property mapping, and are among the most important predictors. This indicates a promising potential for transferring pedological knowledge from data rich countries to countries with limited soil data.

Latin hypercube sampling of Gaussian random fields

Following the method of Stein, this article shows how a Latin hypercube sample can be drawn from a Gaussian random field. In a case study the efficiency of Latin hypercube sampling is compared experimentally to that of simple random sampling. The model outputs studied are the mean and the 5- and 95-percentile of the areal fraction where point concentration of zinc in the topsoil exceeds a given threshold. The Latin hypercube sampling procedure slightly distorts the short-distance correlation, and in an artificial example, it is shown that this distortion is modest for small samples and vanishes for large samples.

Uncertainty analysis in environmental modelling under a change of spatial scale

Although environmental processes at large scales are to a great degree the resultant of processes at smaller scales, models representing these processes can vary considerably from scale to scale. There are three main reasons for this. Firstly, different processes dominate at different scales, and so different processes are ignored in the simplification step of the model development. Secondly, input data are often absent or of a much lower quality at larger scales, which results in a tendency to use simpler, empirical models at the larger scale. Third, the support of the inputs and outputs of a model changes with change of scale, and this affects the relationships between them. Given these reasons for using different models at different scales, application of a model developed at a specific scale to a larger scale should be treated with care. Instead, models should be modified to suit the larger scale, and for this purpose uncertainty analyses can be extremely helpful. If upscaling disturbed the balance between the contributions of input and model error to the output error, then an uncertainty analysis will show this. Uncertainty analysis will also show how to restore the balance. In practice, application of uncertainty analysis is severely hampered by difficulties in the assessment of input and model error. Knowledge of the short distance spatial variability is of paramount importance to input error assessment with a change of support, but current geographical databases rarely convey this type of information. Model error can only be estimated reliably by validation, but this is not easy because the support of model predictions and validation measurements is usually not the same.

Soil water content interpolation using spatio-temporal kriging with external drift

Abstract In this study, two techniques for spatio-temporal (ST) kriging of soil water content are compared. The first technique, spatio-temporal ordinary kriging, is the simplest of the two, and uses only information about soil water content. The second technique, spatio-temporal kriging with external drift, uses also the relationship between soil water content and net-precipitation to aid the interpolation. It is shown that the behaviour of the soil water content predictions is physically more realistic when using spatio-temporal kriging with external drift. Also, the prediction uncertainties are slightly smaller. The data used in this study consist of Time Domain Reflectometry (TDR) measurements from a 30-day irrigation experiment on a 60×60-m grassland in the Netherlands.