Type Raster Dataset
Soil Landscapes of the United States (SOLUS) is a national map product developed by the National Cooperative Soil Survey and focused on providing a consistent set of spatially continuous soil property maps to support large scope soil investigations and land use decisions. SOLUS maps use continuous property mapping, which predicts soil physical or chemical properties in horizontal and vertical dimensions. The soil properties are represented across a continuous range of values. Raster datasets of select soil properties can be predicted at specified depths or depth intervals.
SOLUS maps use a digital soil mapping framework that combines multiple sources of soil survey data with environmental covariate data and machine learning. The SOLUS100 maps are a set of 100m spatial resolution maps that predict 20 soil properties (listed below with units) at seven depths with uncertainty estimates for the contiguous United States.
Very Fine Sand (%)
Fine Sand (%)
Medium Sand (%)
Coarse Sand (%)
Very Coarse Sand (%)
Total Sand (%)
Silt (%)
Clay (%)
pH
Soil Organic Carbon (%)
Calcium Carbonate Equivalent (%)
Gypsum Content (% by weight)
Electrical Conductivity (mmhos/cm)
Sodium Adsorption Ratio
Cation Exchange Capacity (meq/100g)
Effective Cation Exchange Capacity (meq/100g)
Oven Dry Bulk Density (g/cm^3)
Depth to Bedrock (cm)
Depth to Restriction (cm)
Rock Fragment Volume (%)
Each property listed above is predicted at 0, 5, 15, 30, 60, 100, and 150cm depths. Each property-depth prediction has an estimate of uncertainty expressed as the relative prediction interval (RPI) that ranges from 0 to 1. RPI is a relative measure of uncertainty with high values being more uncertain. It is computed as the ratio of the 95% PI width to the training set 95% quantile width (97.5% quantile value - 2.5% quantile value). Values closer to 0 indicate lower uncertainty and values closer to 1 indicate higher uncertainty. Values greater than 1 indicate that the prediction at that location is outside the range of the training data used for that property at that depth. Prediction interval high and low are also provided. Property prediction and uncertainty layers follow the naming convention:
propertyname_depth_cm_p (predicted property values)
propertyname_depth_cm_rpi (relative prediction interval)
propertyname_depth_cm_l (prediction interval low)
propertyname_depth_cm_h (prediction interval high)
A table with layer filename, property, depth, filetype, scalar, description, and units is included with the maps: Final_Layer_Table_20231215.csv
There are no credits for this item.
There are no access and use limitations for this item.
| West | -130.573795 | East | -62.664008 |
| North | 52.394589 | South | 21.111398 |
| Maximum (zoomed in) | 1:5,000 |
| Minimum (zoomed out) | 1:150,000,000 |
Nauman, T.W., S. Kienast-Brown, S.M. Roecker, C. Brungard, D. White, J. Philippe, J.A. Thompson. (2024). Soil Landscapes of the United States (SOLUS): developing predictive soil property maps of the conterminous US using hybrid training sets. Manuscript in preparation.
The period in which SOLUS100 maps were created.
SOLUS100 maps were reviewed internally by National Cooperative Soil Survey scientists prior to release
SOLUS100 maps were reviewed internally by National Cooperative Soil Survey scientists prior to release
SOLUS100 property maps follow a digital soil mapping workflow that includes: 1. preparing and testing various point-location datasets 2. filling gaps in field training data with previous soil survey property estimates 3. building empirical models for each property at seven standard depths 4. testing model performance with conventional and spatial cross validation 5. rendering predictions and uncertainty metrics
The model development for SOLUS100 property maps included five general steps: (a) selecting the optimal set of existing property measurements, (b) combining field samples with an appropriate sample of gNATSGO random training points, (c) fitting a final model and estimating global uncertainty metrics from a large random sample of the mapping domain, (d) comparing global uncertainty to various scales of cross validation uncertainty metrics to determine an appropriate spatial cross validation scale to represent a lower bound of performance, and (e) use of the final model to render predictions. Quantile Random Forests from the R software Ranger package was used for predictive models.
Training data selection and amount varied by property and depth, but most often included National Cooperative Soil Survey Soil Characterization Database (SCD) data, National Soil Information System (NASIS) field sites directly linked to component or linked to a SSURGO spatial home map unit component by a name match, and most models only used observations sampled since the year 2000 (a combination selected in 41 out of 121 models).
Accuracy of predictive models was evaluated using a random 10-fold cross validation (CV) to estimate a high bound of performance and a spatial 10-fold CV to estimate a lower bound of performance for each property at each depth. 10-fold CV R2 values ranged from 0.45 to 0.87 with a mean of 0.65 while spatial CV R2 values ranged from 0.38 to 0.85 with a mean of 0.58.
Final predictions were masked for areas of open water using the 2019 U.S. National Land Cover Dataset to remove all water bodies from final predictions.
Final predictions were masked using the SOLUS100 modeled depth to bedrock layer so that property predictions below the predicted depth to bedrock are not included in the maps.
Final predictions were corrected for extreme values outside of the range of training set possibilities for each soil property by adjusting all predictions above the maximum of the training set to that maximum value, and similarly adjusted all predictions below the minimum training set value to that minimum value.
Soil texture fraction predictions were normalized to 100% in the final maps by adjusting values using a sum normalization (e.g. adjusted sand = sand / [sand + silt + clay]) to ensure logical consistency.
Final predictions were masked for areas labeled NOTPUB in the SSURGO database.
Covariate data was a set of 145 environmental data layers at 100m spatial resolution representing topography, climate, vegetation, and parent material borrowed from the work of Ramcharan et al., 2018. Sources include U.S. National Elevation Dataset, PRISM, MODIS, Landsat, long-term surface water and bare ground occurrence, aero radiometric data, land cover classes, potential natural vegetation, and surficial geology.
Training data sources included National Cooperative Soil Survey Soil Characterization Database (SCD) with laboratory measurements, National Soil Information System (NASIS) pedon observations with taxonomic names but no laboratory measurements, and Gridded National Soil Survey Geographic Database (gNATSGO) weighted average property estimates for model training.
naming convention for soil property and uncertainty layers
National Soil Survey Handbook
depth to bedrock (cm), right censored at 201cm
National Soil Survey Handbook
calcium carbonate equivalent (%)
National Soil Survey Handbook
cation exchange capacity (pH 7) (meq/100g)
National Soil Survey Handbook
clay content (%)
National Soil Survey Handbook
oven dry bulk density (g/cm^3)
National Soil Survey Handbook
electrical conductivity (mmhos/cm)
National Soil Survey Handbook
effective cation exchange capacity (meq/100g)
National Soil Survey Handbook
rock fragment volume (%)
National Soil Survey Handbook
gypsum content (% by weight)
National Soil Survey Handbook
pH measured using 1:1 method (-log10([H+]))
National Soil Survey Handbook
depth to any restrictive layer (cm), right censored at 201cm
National Soil Survey Handbook
coarse sand content (%)
National Soil Survey Handbook
fine sand content (%)
National Soil Survey Handbook
medium sand content (%)
National Soil Survey Handbook
total sand content (%)
National Soil Survey Handbook
very coarse sand content (%)
National Soil Survey Handbook
sodium absorption ratio
National Soil Survey Handbook
silt content (%)
National Soil Survey Handbook
soil organic carbon (%)
National Soil Survey Handbook
very fine sand content
National Soil Survey Handbook