ASRIS Continental-scale soil property predictions 2001

Abstract

This data and its metadata statement were supplied to the Bioregional Assessment Programme by a third party and are presented here as originally supplied.

The Australian Soil Resources Information System (ASRIS) has compiled a national database of existing soil property information. The following details the point-modelling of soil properties from the ASRIS database. Continental soil property models are constructed from this database using decision trees. These relate the soil property to the environment through a suite of environmental predictor variables at the locations where measurements are observed. These decision tree models are then used to extend predictions, from known points to the entire continental extent, by applying the rules derived to the exhaustively available environmental variables.

The models developed have good to fair predictive ability. While they explain the broad changes in property across the ASRIS extent, their overall reliability varies spatially and depends on the property. In general, topsoil models are stronger than subsoil models. There is however a large amount of unexplained variation in all models. This is to a large degree expected given the heterogeneous nature of the database.

Certainty surfaces are provided for all predictions. They represent a combination of local point model performance, point density and environmental representativeness. The last two components emphasise the degree of extrapolation in extending predictions from the point data to the entire extent.

Dataset History

The Australian Soil Resources Information System (ASRIS) has compiled a national database of existing soil property information. Soil Properties are modelled at a resolution of 250m for those river basins incorporating the intensively used agricultural areas. It is broken down into 18 component regions with abbreviated region names as assigned. These jointly serve as separate modelling and assessment regions. At the 250m resolution there are over 43 million pixels at which predictions are required. Predictions however are aggregated to approximately a 1.1km (0.01 degrees) resolution for presentation. Where the soil data do not support modelling for the entire extent a reduced extent may be used. After the initial quality control stage 135490 points were available in the ASRIS database for soil property modelling, albeit of varying completeness for individual properties. A small number of these observations fall outside the ASRIS extent. Further data preparation involved conversion to common units/scales, investigating method differences and calibrating comparable methods where possible, interrogating unusual observations and deleting obvious outliers. This was done for each soil property separately.

The date at which the samples were taken or analysed was not considered due to the relative incompleteness of those fields in the database.

Multiple readings at the same location were allowed. Given we could not make the distinction between time series at the same point and observations that were taken at different locations but were actually tagged with the same point, it was decided to allow both.

Certainty surfaces are provided for all predictions. They represent a combination of local point model performance, point density and environmental representativeness. The last two components emphasise the degree of extrapolation in extending predictions from the point data to the entire extent.

Dataset Citation

CSIRO (2001) ASRIS Continental-scale soil property predictions 2001. Bioregional Assessment Source Dataset. Viewed 13 March 2019, http://data.bioregionalassessments.gov.au/dataset/632c7249-ec00-4ccb-8758-f9026aeec53e.