Assessment of global, national and regional‐level digital soil mapping products at different spatial supports

Abstract

Digital soil mapping provides estimates of soil properties and, in turn, an understanding of the spatial variation in soil across landscapes. In recent years, several digital soil mapping products have been released across the world. An end user in New South Wales (NSW), Australia, could choose between a global product (SoilGrids), a national product (Soil and Landscape Grid of Australia) and a state product (NSW digital soil maps). All predict at grid resolutions of 250 m or less, which approaches the required detail for within-field sitespecific management such as variable-rate application of agricultural inputs.In this study, the quality of clay and soil organic carbon (SOC) products (0-30 cm) from these three publicly available digital soil maps (DSM) were validated with an external dataset of 394 soil observations, collected from 14 farms across eastern Australia. Each farm was sampled using stratified random sampling, allowing for unbiased estimates of prediction quality at three spatial supports: point (soil core), strata (management zone) and farm. For the clay attribute, this study observed Lin's concordance correlation coefficient (LCCC) values of 0.22-0.37 at the point spatial support, far below reported values. This improved at the strata spatial support with LCCC values of 0.20-0.60 and the farm spatial support with values of 0.07-0.54. For SOC, LCCC values ranged from 0.25 to 0.45 for points, 0.31-0.55 for strata and 0.31-0.60 for farms. The grid resolution of DSMs depends largely on the resolution of the covariates used, and current efforts look to incorporate more recent, finer resolution covariates to create higher-resolution soil maps. However, the results of this study suggest that both the grid resolution and spatial support used in DSMs need further consideration. Rather than letting the resolution of covariates determine the resolution of DSMs, this should also be determined by the number of soil observations available for modelling, their distribution in geographic and attribute space and the prediction quality.

Proximal and remote sensing – what makes the best farm digital soil maps?

“…While many DSM studies promote the importance of the products for farmers, many farmers do not know they exist; or if they do, they may not trust the quality of the predictions. A study by Han et al (2022) demonstrated how poor global, national, and state DSM products were at representing withinfarm variability across a collection of diverse cropping fields and farms in Australia.…”

Importance of Terrain and Climate for Predicting Soil Organic Carbon Is Highly Variable across Local to Continental Scales

“…Spatial prediction machine learning models that incorporate terrain information may provide a powerful method to enhance process-based SOC models (see “Future Research” Section in the Results and Discussion Section). Recent studies have also pointed out that digital soil mapping (DSM) products generated over large spatial extents were unable to capture SOC variability at the local scale, , calling for DSM products to be created at different spatial scales. Accounting for the relationships between terrain and SOC at different scales may, therefore, be a key component for improving the monitoring, mapping, and predicting of SOC concentrations, thus supporting more accurate carbon accounting at various scales, as required for project-level interventions to national- and global-scale reporting …”