This article originally appeared in issue 98 of Position magazine, written by Cuan Petheram, principal research scientist at CSIRO and was joint project leader on the Northern Australia Water Resource Assessment.
Developing pure, perennial data resources for water in the Top End with digital soil mapping.
Northern Australia has long been big on ideas but short on data. The national science agency, CSIRO, has now delivered the most extensive, integrated study of water and agriculture resources for the north – covering 200,000 square kilometres across multiple river catchments.
What makes the Northern Australia Water Resource Assessment stand out is not just the vast scope but the range of
disciplines, and assessment techniques, applied together to deliver the fine-scale but region-wide data. The assessment was commissioned by the Australian government to deliver knowledge of the scale, nature, location and distribution of the likely environmental, social and economic opportunities — and risks — of irrigated development. A key input to the assessment of water resource development is an understanding of the soil and landscape resources available, their spatial distribution, and their limitations to use.
Specifically, an understanding of the potential suitability of soils for a range of crops, planting seasons and irrigation management is vital. But how do you map soil parameters and the suitability of soil and land for over 120 land uses across an area of 200,000 square kilometres of diverse soil
types in that time frame?
Traditional soil mapping techniques were impractical. Instead an alternative mapping approach, digital soil mapping (DSM) was adopted. But skills sets and project management have also been critical to delivery.
Applying digital soil mapping in the real-world production environment
With traditional soil surveying approaches, a scientist relies largely on their interpretation of spatial patterns from satellite images or aerial photographs and existing field observations to develop a conceptual model for detailed field investigation. This is time intensive and subjective.
DSM has evolved with gains in computing power, adoption of increasingly sophisticated statistical methods and increased access to high resolution spatial datasets, including climate, remote sensing and digital elevation models.
CSIRO senior experimental scientist Ross Searle, project leader of Data Sources for Agriculture Decision Making, said the assessment wouldn’t have been possible without DSM.
“We couldn’t get close to doing the scale of work using traditional methods, given the funding and time available,” said Mr. Searle. “We would have taken many years using traditional soil mapping methods.”
“Australia is possibly unique in that we do have a perceived large potential for development of those landscapes that we don’t know much about. This was a way to evaluate much larger areas of land.”
CSIRO has been at the forefront of the application of digital soil mapping and modern digital approaches in land suitability analysis across very large and remote areas, he said.
“Until about five years ago, the use of DSM was typically confined to small areas and mostly utilised by researchers. “We’ve now applied DSM in a real world production environment.”
Taking human bias – and limitations – out of field work
DSM involves assembling a series of spatial datasets, each one of which relates to important soil formation processes in the area of interest.
A stratified random sampling approach, in this case conditioned Latin hypercube sampling (cLHS), was used to establish a field sample design. This sampling takes into consideration existing soil measurements in the area of interest, but chooses sites without any human bias and captures the full range of variability within and across each of the spatial datasets.
Pre-existing (legacy) data is inevitably sparse and geographically biased, so it was still essential to collect new soil and landscape data in the field.
The number of field sampling points was determined by the budget and logistical considerations. But one of the challenges in using DSM techniques in remote areas is accessing the statistically determined sample location with field soil sampling equipment such as drilling rigs.
“As a traditional soil surveyor, you go out and use your understanding of the landscape and build a model based on your sampling. DSM sends you to parts of the landscape you wouldn’t ever go to, such as outcrops, because you need to cover the whole range of potential values,” said Mr. Searle.
Working in remote locations with heavy and valuable drilling equipment in tow, some innovation was required to force the statistical model to sample in readily accessible locations, without the loss of too much statistical power.
Field work was still a significant undertaking — 10 teams on 12-day field trips. Data from nearly 60,000 existing soil survey sites, north of latitude 20º S, were combined with data collected from a total of 736 new sites at multiple depths.
CSIRO research technician Seonaid Philip, who coordinated the field work, sample testing and mapping, said DSM made it possible to deliver in the time frame and provide reliable data. DSM also provided a consistent approach working across three jurisdictions, Western Australia, the Northern Territory and Queensland.
“The science in DSM has matured and now there’s a core knowledge in the field with experts using DSM, it was very satisfying. We have tested the reality, applying it and understanding it and having confidence in the work we’re doing,” Ms Philip said.
Samples were sent back to the laboratory for wet and dry chemistry and other analysis such as mid infrared spectroscopy. These measurements were then used to train machine learning models, in this case a random forest statistical model. The random forest models were then combined with the spatial covariate to extrapolate the models across the entire study area to produce maps of soil attributes.
The products are spatially explicit data and maps at a spatial resolution of approximately 90 metres on the ground for 16 soil parameters determined by the project, including: soil depth; soil EC; plant available water capacity; soil pH; drainage; permeability; rockiness; erodibility and exchangeable sodium percentage.
Quantifying the uncertainty: repeatability provides confidence in predictions
DSM offers a number of benefits over traditional soil mapping techniques – increased operational efficiencies, speed, utility of digital outputs and objective output datasets. But perhaps the greatest advantage of DSM, said Mr. Searle, is that because the gridded outputs are generated using statistical methods it is possible to quantify the uncertainty of each pixel for each soil parameter output.
“There’s no such thing as certainty in these highly variable environments. We can now collect more data and establish how sure we are about the value we’re predicting,” said Mr. Searle.
“Say we have 1000 data sets, we might use 900 and we hold back 100. We produce a map and we intersect that with the 100 data sets remaining and see what the difference is. We do that 10 times and create 10 different models using subsets of the sampling. By having that many we can see the range of values.”
The richness and quantity of data allows that repeatability and comparison.
“There’s now broader availability of covariates: from satellite imagery to radiometrics, to build our models with,” said Mr. Searle.
“The software and algorithms we use are all public domain through the R statistical language. CSIRO’s access to a supercomputer resources has been one of the key enablers. We couldn’t do it on this scale without that resource.”
Building land suitability models using DSM
The gridded soil parameter outputs from digital soil mapping have a variety of applications, ranging from mapping of areas suitable for building farm dams to use in hydrology and agriculture production models. To date, their main application has been in the generation of land suitability maps.
Land suitability analysis determines the potential of land for specific uses, based on a rule based approach to the local environmental attributes and qualities.
Land suitability maps use that information to help indicate those parts of the landscape that are more and less suitable for different land uses. Here, ‘land use’ indicates a unique combination of crop type, irrigation method and growing season.
The assessment matched those land use requirements to soil, land and climate limitations.
A set of land suitability data and maps was produced for 126 crop by season by irrigation type-combinations for all three areas at a spatial resolution of approximately 90 metres on the ground. The soil attribute data were also used to evaluate suitability of offstream storages and as input to hydrological models.
Ms. Philip said what’s been clear is that the DSM and land suitability products delivered as part of the assessment hold up statistically.
“As far as quantitative and qualitative (expert knowledge) reliability are concerned, these products have performed really well. Traditional soil surveys have little information on reliability. This process allows us to produce different products that can be applied in so many ways.”
This information will be a valuable resource for those seeking to make decisions about development in the three study areas, and will help direct future surveys and sampling effort.
A team approach
Mr. Searle said DSM required a whole new skill set for the traditional soil scientist.
“You still need to have the soils knowledge to do it well but you need the statistical understanding as well,” he said. “The DSM community gets around this by trying to work as teams, it’s much more effective when you can access the range of skills.”
More than 140 experts worked on the assessment, an example of the reach of newer science as well as the value of managing complex logistics, workflows and project and personnel management.
The digital soil mapping and subsequent land suitability analysis were a collaborative approach between CSIRO and Western Australia, the Northern Territory and Queensland. All three jurisdictions were heavily involved in field data collection, improvement of the data layers, development of the land suitability framework and the technical assessment of the modelling products.
The Northern Australia Water Resource Assessment also investigated: climate; surface water hydrology; groundwater hydrology; agriculture and aquaculture viability; water storage; socio-economics; Indigenous water values, rights and development aspirations; and aquatic and marine ecology.
All data collected and analysed as part of the assessment is publicly available via reports and discoverable by location through web applications including NAWRA-explorer.
Stay up to date by getting stories like this delivered to your mailbox.
Sign up to receive our free weekly Spatial Source newsletter.