Meyer Bohn and Bradley Miller – Iowa Water Center Conference April 6-8, 2021
Abstract
Abstract
The objectives of this study were to assess spatial predictions of topsoil thickness from models produced from ensemble machine learning algorithms along with assessing the uncertainty estimations associated with those models. Boosting is one example of an ensemble method, which attempts to improve accuracy by combining multiple weaker models into a unified, stronger model. The inherent problem with ensemble learning is that model accuracy is evaluated with subsets of the sample population, which can result in model overfitting. Cross-validation (CV) partitions the sample population into multiple random training and validation subsets. The training data calibrate a model and test prediction accuracy on the validation subset. To determine if a model is truly robust, accuracy should be assessed by testing model predictions on an independent validation (IV) subset. To evaluate model accuracy in this study, over 900 samples from central Iowa were used for model training and over 600 digital terrain derivatives were generated as predictor variables to develop mathematical models with the machine-learning algorithm, Cubist. Fifty-four observations were selected from the three townships and reserved for independent validation. Three ensemble methods, 1) bootstrapped-bagged 2) bootstrapped-boosted, and 3) 10 fold cross-validation (CV)-boosted were applied to determine which method was most robust. A 90% confidence interval was calculated from ten prediction realizations of the bootstrapped-bagged method to determine how many IV observations were contained within the estimated range of uncertainty.
The CV-boosted method was most robust in prediction with an IV relative error of 46% (Mean Absolute Error (MAE) = 26.5 cm). The bootstrapped-bagged and bootstrapped-boosted models had IV relative errors of 83% and 51%, respectively. For validation of the prediction interval, 53% of the IV observations were contained within the range of uncertainty and the average interval width was 51 cm. Maps of statistical uncertainty revealed that the model confidently predicts topsoil thickness in upland soil-landscapes and performs poorly in fluvial and pluvial areas.
References
Why care about topsoil thickness in Iowa agricultural soils
- Cruse, R.M. 2016. Economic impacts of soil erosion in Iowa Economic impacts of soil erosion in Iowa: What is the impact of existing soil erosion rates on crop yield and subsequent income alterations?
- Montgomery, D.R. Matson, P.A. 2007. Soil erosion and agricultural sustainability. Proc. Natl. Acad. Sci. U. S. A. 104, 13268–13272.
Soil databases for training data
- National Cooperative Soil Survey. National Cooperative Soil Characterization Database. Available online. Accessed 1 Jan 2018.
- National Cooperative Soil Survey. National Soil Information System Pedons. Accessed 1 Jan 2018.
Accessing and analyzing soils training data
- R Core Team. 2020. R: A language and environment for statistical computing. R version 3.6.1. Foundation for Statistical Computing, Vienna, Austria. https://www.R-project.org/. (accessed 6 Jan. 2020)
- Beaudette, D.E., P. Roudier, and A.T. O’Geen. 2013. Algorithms for quantitative pedology: A toolkit for soil scientists, Computers & Geosciences. 52: 258–268.
Machine learning concepts
- Rocca, J. and B. Rocca. 2019. Ensemble methods: bagging, boosting, and stacking. Understanding the key concepts of ensemble learning. Accessed 1 Mar 2021. Available at: https://towardsdatascience.com/ensemble-methods-bagging-boosting-and-stacking-c9214a10a205
Digital soil mapping methods
- Odgers, N. 2017. Digital soil mapping with covariates. Accessed 1 Jan 2021. Available at: http://pierreroudier.github.io/teaching/20171014-DSM-Masterclass-Hamilton/2017-10-09-dsm-with-covariates.html
- Miller, B.A., 2017. Digital Soil Mapping and Pedometrics, in: International Encyclopedia of Geography: People, the Earth, Environment and Technology. John Wiley & Sons, Ltd, Oxford, UK, pp. 1–8. https://doi.org/10.1002/9781118786352.wbieg0318
- Miller, B.A., Koszinski, S., Hierold, W., Rogasik, H., Schröder, B., Van Oost, K., Wehrhan, M., Sommer, M. 2016. Towards mapping soil carbon landscapes: Issues of sampling scale and transferability. Soil Tillage Res. 156, 194–208. https://doi.org/10.1016/j.still.2015.07.004
Digital terrain analysis methods
- Conrad, O., B. Bechtel, M. Bock, H. Dietrich,E. Fischer,L. Gerlitz,J. Wehberg,V. Wichmann,and J. Böhner. 2015. System for Automated Geoscientific Analyses (SAGA) v. 2.1.4, Geosci. Model Dev., 8, 1991-2007, doi:10.5194/gmd-8-1991-2015.
- GRASS Development Team. 2018. Geographic Resources Analysis Support System (GRASS) Software, Version 7.6. Open Source Geospatial Foundation. https://grass.osgeo.org.
- Miller, B.A., Schaetzl, R.J., 2015. Digital classification of hillslope position. Soil Sci. Soc. Am. J. 79, 132–145. https://doi.org/10.2136/sssaj2014.07.0287
Machine Learning methods with the Cubist algorithm
- Quinlan, J.R. 1992. Learning with continuous classes. Proceedings of the 5th Australian Joint Conference on Artificial Intelligence p. 343–348.
- Quinlan, J.R. 1993. Combining instance-based and model-based learning. In: Kaufmann, M. (ed.), Proceedings of the Tenth International Conference on Machine Learning p. 236–243.
- Quinlan, J.R. 1994. C4.5: programs for machine learning. Mach. Learn. 16: 235–240.
- Kuhn, M. 2008. Caret package. Journal of Statistical Software. 28(5).