Title

Soil Grids of Victorian - Soil Clay % 

Alternate title

SOIL_CLAY_PC 

Credit

The project team acknowledges the support provided by members of the National Digital Soil Mapping Working group and the project team that delivered the Soil and Landscape Grid of Australia through the Terrestrial Ecosystems Research Network (TERN). 

Supplemental Information

Related Documents: None 

Status

Under development  

Code

4283 

Maintenance and update frequency

Not planned  

Title

DIGITAL geotiff 3 

Topic category

• Geoscientific information

Use limitation

Creative Commons by Attribution Whilst the Soil Grids show an immense amount of fine scale detail, and are our best spatially continuous and exhaustive estimates of soil attributes across all of Victoria, they are most appropriately used for assessments of regional to state-wide trends of soil properties and their relationship with their environment and pedogenesis. Care should be taken when using the grids for local assessments and it is recommended that the confidence intervals are included at this scale. Site based assessments should be made when assessing and making decisions at local scales. 

Classification

Unclassified  

Statement

Dataset Source: The soil grids have been produced by Agriculture Victoria Research, a division of the Department of Economic Development, Jobs, Transport and Resources. Development of the grids has involved methodologies generally referred to as Digital Soil Mapping (DSM). The International Union of Soil Sciences Digital Soil Mapping Working Group defines DSM as creation and the population of a geographically referenced soil database, generated at a given resolution by using field and laboratory observation methods coupled with environmental data through quantitative relationships. In essence, it uses more easily obtained, spatially exhaustive, `predictor¿ datasets to infer likely soil property values at a given location and resolution. The Soil Grids of Victoria are a contribution to other State, National and International digital soil mapping efforts. The grids are Victoria¿s first state-wide attempt at this efficient and cost-effective approach to mapping soils at fine-scale resolution in a consistent and easy to use format. The methodology used to develop the Soil Grids of Victoria has been based on that refined by the Australian Soil and Landscape Grid. Data and knowledge embedded into existing soil related datasets, e.g. soil profile and land mapping collections, have been key inputs. Although, the maps have been produced at a fine spatial scale, 90 m grid resolution, the predicted value within each grid pixel has been derived from models calibrated from available data across Victoria. This means the effects of localised conditions may be somewhat muted. Furthermore, each property has been modelled independently. Co-variation between properties such as particle size distribution, i.e. proportion of sand, silt and clay, and field capacity and wilting point have not been considered. Input soil data is of varying quality being influenced by measurement technique, spatial geo-referencing and age. In all instances, data from a variety of measurement techniques have been combined and this data has been modified as it has been harmonised to target depths. Pragmatic decisions have been made when sourcing and treating data to maximise the amount and distribution of data across Victoria to develop the predictive models. Except for pH and Bulk Density, the model calibration and validation data has included estimated values derived from MIR spectroscopy. These data have a larger uncertainty than direct laboratory measurements and therefore impact the accuracy of resulting map predictions. Input site data has been sourced from soil surveys, many dating back to the 1950s. For dynamic soil properties, such as soil pH, it should be noted that no date filter has been applied. Dataset Originality: Derived 

Description

Collection Method: Survey and modelling 

Description

Key steps in the production of the Victorian soil grids are described below: 1. Collation and preparation of site based observations and measurements of soil. 2. Creation and preparation of gridded environmental predictor datasets. 3. Creation and application of Cubist models to generate soil property surfaces. 4. Expert assessment of modelled soil property grids and refinement. 5. Post processing of modelled soil property datasets. Most likely soil property value ranges were determined in consultation with pedologists and by checking the measured input data. A smoothing filter (each cell being allocated an average value from its eight surrounding cells) was applied to all the grids to remove only the spurious predictions. Where necessary, additional selective filters were applied to pixels of the mean prediction grids with data outside the acceptable range and average values from neighbouring pixels were allocated. 

Classification

Unclassified  

Attribute Quality

Comments

Model fit summary statistics (Clay %) Depth 5cm 15cm 30cm 60cm 100cm 200cm R2 0.5 0.53 0.45 0.38 0.38 0.32 LCCC 0.67 0.69 0.6 0.54 0.54 0.49 RMSE 10.196 10.289 13.344 14.791 14.356 14.451 ME -0.346 0.002 -1.453 0.19 0.314 0.83 Av Obs. 22.6 24.5 31 39.9 45.7 43.24 Av Mod. 22.3 24.5 29.5 40.1 46 44.1 Lin¿s concordance correlation co-efficient (LCCC) assesses covariation and correspondence between the predictions and the original data. Values > 0.9 denote near perfect agreement, values between 0.75 and 0.9 show substantial agreement, between 0.6 and 0.75 show moderate agreement and those < 0.6 indicate poor agreement (Lin 1989). The Root Mean Square Error quantifies the inaccuracy of the predictions and the Mean Error the prediction bias. 

Positional Accuracy

Comments

Not Known 

Conceptual Consistency

Comments

Not Known 

Missing Data

Comments

Victoria 

Excess Data