Spatializing and Digitizing Available Soil Information in Ontario, Canada

Thumbnail Image
Journal Title
Journal ISSN
Volume Title
University of Guelph

This doctoral research aimed to predict soil information in Ontario, specifically soil organic carbon (SOC) content and soil series, using conventional soil maps (CSMs) and survey reports in conjunction with various digital soil mapping tools and techniques. The study examined the feasibility of disaggregation techniques to spatially characterize CSMs and addressed the associated challenges. It compared two disaggregation techniques, DSMART and Pure Polygon Disaggregation (PPD), with modified sampling and classifier approaches, for spatializing soil series maps. The study derived and spatialized SOC information from the disaggregated soil series maps and predicted both soil series maps and SOC content at the provincial scale. The research compared the conditioned Latin hypercube sampling (cLHS) and simple random sampling (SRS) approaches using three classifier algorithms within DSMART to disaggregate the soil taxonomic units of a sub-watershed in Southern Ontario, Canada. The results indicate that both SRS and cLHS performed better with C5.0 and random forest (RF) algorithms. Furthermore, the study compared the DSMART and PPD techniques for a regional municipality in Ontario, Canada, and found that they yielded similar accuracy assessment results. Notably, PPD proved significantly less computationally intensive and time-consuming than DSMART. Two methods were proposed in the study for generating predictive SOC% values using soil profile information from existing CSMs: the reassigned method and the weighted-probable approach. The C5.0-probable approach, utilizing disaggregated maps generated through both DSMART and PPD, yielded the most promising results. Based on these findings, the study generated disaggregated soil series maps for Ontario and created a mean SOC% map for the province using the weighted-probable approach. To address the class imbalance, the study oversampled synthetic data from underrepresented soil series when building the training dataset. In conclusion, this study underscores the capacity of digital soil mapping techniques to offer spatialized and digitized soil information through the disaggregation of soil series maps and the generation of soil property estimations utilizing CSMs. These techniques have the potential to be a complementary tool for enhancing conventional soil data collection and knowledge generation methods, thereby, providing valuable insights for informing policy and the implementation of sustainable and climate-smart agricultural practices.

digital soil mapping, machine learning algorithms, conventional soil map disaggregation, soil organic carbon prediction, sampling techniques
Easher, T.H., Saurette, D., Chappell, E., Lopez, F. de J. M., Gasser, M.-O., Gillespie, A., Heck, R. J., Heung, B., & Biswas, A. (2023). Sampling and classifier modification to DSMART for disaggregating soil polygon maps. Geoderma, 431, 116360�??.
Chappell, E., Easher, T.H., Saurette, D., Biswas, A. (2021). Soil Organic Carbon: Past, Present, and Future Research. In: Rakshit, A., Singh, S., Abhilash, P., Biswas, A. (eds) Soil Science: Fundamentals to Recent Advances. Springer, Singapore.