Khaledian, Y. and B.A. Miller. Selecting appropriate machine learning methods for digital soil mapping. Applied Mathematical Modelling 81: 401-418. doi: 10.1016/j.apm.2019.12.016.
Tag: Digital Soil Mapping
This study examines the spatial patterns and accuracies of predictions made by different spatial modelling methods on sample sets taken at two different scales. These spatial models are then tested on independent validation sets taken at three different scales. Each spatial modelling method produced similar, but unique, maps of soil organic carbon content (SOC%). Kriging approaches excelled at internal spatial prediction with more densely spaced sample points.
Soil mapping, classification, and pedologic modelling have been important drivers in the advancement of our understanding of soil. Advancement in one of these highly interrelated areas tend to lead to corresponding advances in the others. Traditionally, soil maps have been desirable for purposes of land valuation, agronomic planning, and even in military operations. The expansion of the use of soil knowledge to address issues beyond agronomic production, such as land use planning, environmental concerns, energy security, water security, and human health, to name a few, requires new ways to communicate what we know about the soils we map as well as bringing forth research questions that were not widely considered in earlier soils studies.
The objective of this study was to evaluate the ability of high-resolution, minimally invasive sensor data to predict spatial variation of soil organic carbon stocks within highly degraded peatland soils in northeast Germany. Soil organic carbon density was related to elevation, electrical conductivity, and peat thickness. Modeling peat thickness based on sensor data needs additional research, but seems to be a valuable set of covariates in digital soil mapping.
Proponents of digital soil mapping sometimes criticize traditional soil mapping for using a discrete data model to describe a continuous surface (field) and lacking a quantified estimation of error. Although most of my research is on digital soil mapping, I like to give due credit to the accomplishments of traditional soil mapping. I understand the need for … Continue reading The Real Benefits of Digital Soil Mapping
A comparison of direct and indirect approaches for mapping SOC stocks from rule-based, multiple linear regression models applied at the landscape scale via spatial association. The final products for both strategies are high-resolution maps of SOC stocks (kg m‾²), covering an area of 122 km², with accompanying maps of estimated error. Although the indirect approach fit the spatial variation better and had a lower mean estimated error for the topsoil stock, the mean estimated error for the total SOC stock (topsoil + subsoil) was lower for the direct approach. The optimal approach would depend upon the intended use of the map.
Classification of elevation rasters with this digital model of hillslope position represent base maps that can be used to (1) improve research on toposequences by providing explicit definitions of each hillslope element’s location, (2) facilitate the disaggregation of soil map unit complexes, and (3) identify map unit inclusions that occur due to subtle topographic variation.
Results suggest that models with limited predictor pools can substitute other predictors to compensate for unavailable variables. However, a better performing model was always found by considering predictor variables at multiple scales. Although the scale effect of the modifiable area unit problem is generally well known, this study suggests digital soil mapping efforts would be enhanced by the greater consideration of predictor variables at multiple analysis scales.