Predicting reference soil groups using legacy data: A data pruning and Random Forest approach for tropical environment (Dano catchment, Burkina Faso)

Abstract

Predicting taxonomic classes can be challenging with dataset subject to substantial irregularities due to the involvement of many surveyors. A data pruning approach was used in the present study to reduce such source errors by exploring whether different data pruning methods, which result in different subsets of a major reference soil groups (RSG) – the Plinthosols – would lead to an increase in prediction accuracy of the minor soil groups by using Random Forest (RF). This method was compared to the random oversampling approach. Four datasets were used, including the entire dataset and the pruned dataset, which consisted of 80% and 90% respectively, and standard deviation core range of the Plinthosols data while cutting off all data points belonging to the outer range. The best prediction was achieved when RF was used with recursive feature elimination along with the non-oversampled 90% core range dataset. This model provided a substantial agreement to observation, with a kappa value of 0.57 along with 7% to 35% increase in prediction accuracy for smaller RSG. The reference soil groups in the Dano catchment appeared to be mainly influenced by the wetness index, a proxy for soil moisture distribution.

Synthetic resampling strategies and machine learning for digital soil mapping in Iran

“…Therefore, the trained ML algorithms on the ROS resampled data could not precisely be generalized to the unseen data. This is one reason why we obtained a poor performance of ML algorithms using ROS resampled data, which is in line with the findings of Hounkpatin et al (), who pointed out the poor power of generalization of RF models trained by ROS resampled data. Contrary to our findings, Sharififar, Sarmadian, Malone, and Minasny () indicated a significant improvement in ML learning when they made balanced soil data using ROS.…”

“…Generally speaking, the performances of ML algorithms trained on the original imbalanced dataset indicated that the two ensemble‐based models (RF and XGBoost) show high and fairly similar accuracy (Table ). Our findings are in line with results of several DSM literature reviews, which all confirmed the power of ensemble‐based models compared to the other common ML algorithms (Brungard et al, ; Hounkpatin et al, ). However, a closer inspection of the calculated recall values of individual classes revealed that a considerable number of minority classes will be misclassified as majority classes, such as Chernozems, Phaeozems and Solonetz (Table ).…”

“…In this respect, datasets that do not follow this criterion are called imbalanced data. Most soil class datasets are therefore imbalanced data (Hounkpatin et al, ). This stems from the spatial distribution pattern of soils, which is usually not equal.…”

“…Digital soil mapping (DSM) has been successfully applied for mapping of soil classes at a large range of scales, including regional and local scales (Brungard, Boettinger, Duniway, Wills, & Edwards, ; Hounkpatin et al, ; Schmidt, Behrens, & Scholten, ), the national scale (Adhikari, Minasny, Greve, & Greve, ; Ramcharan et al, ) and the continental scale (Hengl et al, ; Teng, Rossel, Shi, & Behrens, ). Digital soil mapping can integrate soil point observations (e.g., classified soils) with various sources of grid‐based geospatial covariates (e.g., satellite imagery, digital elevation models and climate data).…”

Temporal Variations of Soil Organic Carbon and pH at Landscape Scale and the Implications for Cropping Intensity in Rice-Based Cropping Systems

“…Firstly, soil sample locations were identified according to the number of samples in the respective soil polygon map of each upazila. Then, the samples were matched individually within the grids following objective selection [35] based on physiography, land types and soil series.…”