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.
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.
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%.
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.
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.
The 1:50,000 national soil survey of the Netherlands, completed in the early 1990s after more than three decades of mapping, is gradually becoming outdated. Large-scale changes in land and water ...management that took place after the field surveys have had a great impact on the soil. Especially oxidation of peat soils has resulted in a substantial decline of these soils. The aim of this research was to update the national soil map for the province of Drenthe (2680 km
2) without additional fieldwork through digital soil mapping using legacy soil data. Multinomial logistic regression was used to quantify the relationship between ancillary variables and soil group. Special attention was given to model-building as this is perhaps the most crucial step in digital soil mapping. A framework for building a logistic regression model was taken from the literature and adapted for the purpose of soil mapping. The model-building process was guided by pedological expert knowledge to ensure that the final regression model is not only statistically sound but also pedologically plausible. We built separate models for the ten major map units, representing the main soil groups, of the national soil map for the province of Drenthe. The calibrated models were used to estimate the probability of occurrence of soil groups on a 25 m grid. Shannon entropy was used to quantify the uncertainty of the updated soil map, and the updated soil map was validated by an independent probability sample. The theoretical purity of the updated map was 67%. The estimated actual purity of the updated map, as assessed by the validation sample, was 58%, which is 6% larger than the actual purity of the national soil map. The discrepancy between theoretical and actual purity might be explained by the spatial clustering of the soil profile observations used to calibrate the multinomial logistic regression models and by the age difference between calibration and validation observations.
This study compared the efficiency of geostatistical digital soil mapping (DSM) with conventional soil mapping (CSM) for updating soil class and property maps of a cultivated peatland in the ...Netherlands. For digital soil class mapping, the generalized linear geostatistical model was used. Digital mapping of the soil organic matter (SOM) content and peat thickness was done by universal kriging. The conventional soil class map was created by free survey, while the property maps were created with the representative profile description (RPD) and map unit means (MUM) methods. For each method, we computed the effort invested in the mapping in terms of the sampling and cost densities. The accuracies of the created soil maps were estimated from independent probability sample data. The results showed that for DSM, the cost density could be reduced by a factor of three compared with CSM without compromising accuracy. The map purity of both maps was around 55%. For conventional soil property mapping, the MUM maps were more accurate than the RPD maps. For SOM, CSM-MUM (RMSE 7.5%) performed better than DSM (RMSE 12.1%), although accuracy differences were not significant. For peat thickness, DSM (RMSE 23.3 cm) performed slightly better than CSM-MUM (RMSE 24.9 cm). Despite the differences in accuracy being small, the digital soil property maps were produced more efficiently. The cost density was a factor of 3.5 smaller. We conclude that for updating conventional soil maps in the Dutch peatlands, geostatistical DSM can be more efficient, although not necessarily more accurate, than CSM.
This study compares the performance of five popular equal-area projections supported by Free and Open Source Software for Geo-spatial (FOSS4G)—Sinusoidal, Mollweide, Hammer, Eckert IV and Homolosine. ...A set of 21,872 discrete distortion vindicatrices were positioned on the ellipsoid surface, centred on the cells of a Snyder icosahedral equal-area grid. These indicatrices were projected on the plane and the resulting angular and distance distortions computed, all using FOSS4G. The Homolosine is the only projection that manages to minimise angular and distance distortions simultaneously. It yields the lowest distortions among this set of projections and clearly outclasses when only land masses are considered. These results also indicate the Sinusoidal and Hammer projections to be largely outdated, imposing too large distortions to be useful. In contrast, the Mollweide and Eckert IV projections present trade-offs between visual expression and accuracy that are worth considering. However, for the purposes of storing and analysing big spatial data with FOSS4G the superior performance of the Homolosine projection makes its choice difficult to avoid.
Global mapping of soil salinity change Ivushkin, Konstantin; Bartholomeus, Harm; Bregt, Arnold K. ...
Remote sensing of environment,
09/2019, Letnik:
231
Journal Article
Recenzirano
Odprti dostop
Soil salinity increase is a serious and global threat to agricultural production. The only database that currently provides soil salinity data with global coverage is the Harmonized World Soil ...Database, but it has several limitations when it comes to soil salinity assessment. Therefore, a new assessment is required. We hypothesized that combining soil properties maps with thermal infrared imagery and a large set of field observations within a machine learning framework will yield a global soil salinity map. The thermal infrared imagery acts as a dynamic variable and allows us to characterize the soil salinity change. For this purpose we used Google Earth Engine computational environment. The random forest classifier was trained using 7 soil properties maps, thermal infrared imagery and the ECe point data from the WoSIS database. In total, six maps were produced for 1986, 2000, 2002, 2005, 2009, 2016. The validation accuracy of the resulting maps was in the range of 67–70%. The total area of salt affected lands by our assessment is around 1 billion hectares, with a clear increasing trend. Comparison with 3 studies investigating local trends of soil salinity change showed that our assessment was in correspondence with 2 of these studies. The global map of soil salinity change between 1986 and 2016 was produced to characterize the spatial distribution of the change. We conclude that combining soil properties maps and thermal infrared imagery allows mapping of soil salinity development in space and time on a global scale.
•Soil properties maps combined with thermal imagery can be used to map soil salinity.•Thermal imagery can act as a dynamic variable.•Soil salinisation is increasing on global scale.
•Maps of P-Olsen and Exch-K are less accurate than pH and organic carbon.•Smoothing led to an overestimation of QUEFTS spatial yield predictions.•Calculate-then-interpolate provided most accurate ...QUEFTS yield predictions.
Using fertilisers is indispensable for closing yield gaps in Sub Saharan Africa. Current fertiliser recommendations, however, are often blanket recommendations which do not take spatial variation in soil conditions within a region or country into account. Soil maps can potentially support fertiliser recommendations at a higher spatial resolution. The QUantitative Evaluation of the Fertility of Tropical Soils (QUEFTS) model is a decision support tool that predicts crop yields as an indicator of soil fertility and can be used to evaluate yield responses to fertilisers. It was designed for field level output and runs on field-specific soil information. The aim of this study was to compare two methods for developing maps of QUEFTS output, i.e. maize yield and the yield-limiting nutrient, with Rwanda as a case study. We used a database containing soil analysis results of 999 samples collected across Rwanda. Transfer functions were applied to predict the required P-Olsen and Exchangeable K input for QUEFTS based on the soil data. For the “Calculate-then-Interpolate” (CI) method, transfer functions and QUEFTS were applied to point data, and the final output was then interpolated using random forest modelling. For the “Interpolate-then-Calculate” (IC) method, maps of the soil parameters were developed first, before applying calculations. Implications of the chosen method (i.e. CI or IC) on QUEFTS predictions on a national scale were evaluated using set-aside locations. Results showed low precision and accuracy of QUEFTS maize yield predictions across Rwanda. The CI method performed better in predicting QUEFTS yield and yield-limiting nutrient than the IC method. Correlations between mapped yield predictions and predictions on set-aside evaluation locations were similar for the CI (r = 0.444) and IC (r = 0.439) methods. The poorer performance of the IC method was mostly due to overestimation of yields, which was most likely caused by the effect of smoothing on the soil maps used as input for QUEFTS. We conclude that the CI method is the preferred method for spatial application of QUEFTS.
Peatlands offer a series of ecosystem services including carbon storage, biomass production, and climate regulation. Climate change and rapid land use change are degrading peatlands, liberating their ...stored carbon (C) into the atmosphere. To conserve peatlands and help in realising the Paris Agreement, we need to understand their extent, status, and C stocks. However, current peatland knowledge is vague—estimates of global peatland extent ranges from 1 to 4.6 million km2, and C stock estimates vary between 113 and 612 Pg (or billion tonne C). This uncertainty mostly stems from the coarse spatial scale of global soil maps. In addition, most global peatland estimates are based on rough country inventories and reports that use outdated data. This review shows that digital mapping using field observations combined with remotely-sensed images and statistical models is an avenue to more accurately map peatlands and decrease this knowledge gap. We describe peat mapping experiences from 12 countries or regions and review 90 recent studies on peatland mapping. We found that interest in mapping peat information derived from satellite imageries and other digital mapping technologies is growing. Many studies have delineated peat extent using land cover from remote sensing, ecology, and environmental field studies, but rarely perform validation, and calculating the uncertainty of prediction is rare. This paper then reviews various proximal and remote sensing techniques that can be used to map peatlands. These include geophysical measurements (electromagnetic induction, resistivity measurement, and gamma radiometrics), radar sensing (SRTM, SAR), and optical images (Visible and Infrared). Peatland is better mapped when using more than one covariate, such as optical and radar products using nonlinear machine learning algorithms. The proliferation of satellite data available in an open-access format, availability of machine learning algorithms in an open-source computing environment, and high-performance computing facilities could enhance the way peatlands are mapped. Digital soil mapping allows us to map peat in a cost-effective, objective, and accurate manner. Securing peatlands for the future, and abating their contribution to atmospheric C levels, means digitally mapping them now.