Comparing Machine Learning Models and Hybrid Geostatistical Methods Using Environmental and Soil Covariates for Soil pH Prediction

Tziachris, Panagiotis; Aschonitis, Vassilis; Chatzistathis, Theocharis; Papadopoulou, Maria; Doukas, Ioannis

doi:10.3390/ijgi9040276

Cited by 21 publications

(11 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Random forests, on the other hand, not only incorporate randomization and bootstrapping into the construction of individual Categorization and Regression trees, but also aggregate the output of many individual trees. These characteristics are thought to counter overfitting and generally improve accuracy [22,24,29,[63][64][65][66][67]. The current study disagrees with these findings in the context of soil pH prediction.…”

Section: Cart Vs Random Forest Modelscontrasting

confidence: 94%

Assessing the geographic specificity of pH prediction by classification and regression trees

et al. 2021

View full text Add to dashboard Cite

Soil pH effects a wide range of critical biogeochemical processes that dictate plant growth and diversity. Previous literature has established the capacity of classification and regression trees (CARTs) to predict soil pH, but limitations of CARTs in this context have not been fully explored. The current study collected soil pH, climatic, and topographic data from 100 locations across New York’s Temperate Deciduous Forests (in the United States of America) to investigate the extrapolative capacity of a previously developed CART model as compared to novel CART and random forest (RF) models. Results showed that the previously developed CART underperformed in terms of predictive accuracy (RRMSE = 14.52%) when compared to a novel tree (RRMSE = 9.33%), and that a novel random forest outperformed both models (RRMSE = 8.88%), though its predictions did not differ significantly from the novel tree (p = 0.26). The most important predictors for model construction were climatic factors. These findings confirm existing reports that CART models are constrained by the spatial autocorrelation of geographic data and encourage the restricted application of relevant machine learning models to regions from which training data was collected. They also contradict previous literature implying that random forests should meaningfully boost the predictive accuracy of CARTs in the context of soil pH.

show abstract

Section: Cart Vs Random Forest Modelscontrasting

confidence: 94%

Assessing the geographic specificity of pH prediction by classification and regression trees

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Comparing RF and CB with RF-OK and CB-OK predictions suggests that OK, combined with the RF and CB method, increases R 2 value by 11 to 23% in the SOCS prediction process, respectively. The hybrid methods are combined to improve the accuracy of spatial modeling predictions by other researchers (Keskin and Grunwald., 2018;Hengl et al, 2018;Tziachris et al, 2020). Other researchers (Kumar et al, 2012;Mirzaee et al, 2016;Song et al, 2017) also believed when machine learning models are combined with geostatistical methods; prediction accuracy is increased.…”

Section: Evaluation Of Machine Learning Modelsmentioning

confidence: 99%

Spatial prediction of soil organic carbon stocks in an arid rangeland using machine learning algorithms

Rostaminia

Rahmani

Mousavi

et al. 2021

Environ Monit Assess

View full text Add to dashboard Cite

method. Robust and popular random Forest (RF), cubist (CB) along with random forest-ordinary kriging (RF-OK), and cubist-ordinary kriging (CB-OK) hybrid ML models were applied to the prediction of SOCS. Ten-fold CV was implemented for modeling performance and uncertainty map. According to data analysis, the maximum, minimum, and average values of SOCS are 44.50, 10.50, and 20.50 (ton. ha −1 ) at the surface depth (0-30 cm), respectively. In general, normalized and standardized height covariates had a higher effect related to other predictors. On the other hand, two remote sensing (RS) indices, including salinity ratio (salinity) and GNDVI index, had a better impact on SOCS variability. The external validation of model performance indicated that RF-OK with (R 2 = 0.75, RMSE = 6.33 ton. ha −1 ) with the high and low uncertainty range (3.33-9.50 ton. ha −1 ) was the outperformed ML model in compare with other models as RF (R 2 = 0.65, RMSE = 7.38 ton. ha −1 ), CB-OK (R 2 = 0.56, RMSE = 9.22 ton. ha −1 ), and CB (R 2 = 0.33, RMSE = 10.42 ton. ha −1 ). In general, the hybrid models improved the accuracy of RF and CB with increased 0.11 until 0.23 of R 2 , and 1.05 to 1.2 (ton. ha −1 ) decreased RMSE of model's prediction. Hence, we conclude that the topographic attributes (especially normalized and standardized height) were the most critical factors in controlling surface SOCS in arid rangelands when combining with robust RF ML model, and optimized soil sampling methods like RF-cLHS can prepare acceptable soil properties maps.

show abstract

“…However, this finding was not supported by Hengl et al (2017) who reported that estimated R 2 using XGBoost was lower for coarse fragments and soil pH compared with RF. Tziachris et al (2020) have also found that both the XGBoost and RF models presented the best results in the prediction of soil pH. Contrary to that highlighted by Chen et al (2022), the model performances did not decrease at deeper depth intervals in terms of SOC and SOCD.…”

Section: Resultsmentioning

confidence: 90%

Digital assessments of soil organic carbon storage using digital maps provided by static and dynamic environmental covariates

Shirazi

Shahbazi

Rezaei

et al. 2023

Soil Use and Management

View full text Add to dashboard Cite

Understanding the vertical and lateral distribution of soil organic carbon (SOC) and soil organic carbon density (SOCD) is indispensable for soil use and environmental management because of their vital role in soil quality assessments. Primarily, they are needed in calculating soil organic carbon storage (SOCS). The objective of this research was to provide digital maps of SOC and SOCD variation as well as their uncertainties at multiple standardized depths (H1: 0–5, H2: 5–15, H3: 15–30, H4: 30–60 and H5: 60–100 cm) using a parsimonious model with optimized terrain‐related attributes and satellite‐derived data. SOCS were evaluated at soil subgroup levels. An area of about 808 km2 with varying elevation, plant cover and lithology from the Miandoab region, West Azerbaijan Province, Iran was selected as a case study area. A total of 386 soil samples were collected from 104 profiles comprising various soil genetic horizons. A continuous spline function was then fitted to the target properties in advance of creating a dataset at five standard depth intervals (following the GlobalSoilMap project). These were then grouped into three classes including top (H1), middle (H2, H3 and H4) and bottom (H5) depths to ease interpretation. Static and dynamic covariates (30‐m resolution) were derived from a digital elevation model (DEM) and a suite of Landsat‐8 spectral imageries, respectively. Four candidate models including stepwise multiple linear regression (SMLR), random forest (RF), cubist (CU) and extreme gradient boosting (XGBoost) Tree were tested in this study. Finally, the digital maps at 30‐m resolution of SOC and SOCD and their uncertainties were prepared using the best‐fit model and the bootstrapping method, respectively. Four soil subgroups (Gypsic Haploxerepts, Typic Calcixerepts, Typic Haploxerepts and Xeric Haplocalcids) were identified across the study area. The covariates had variable contributions on the evaluated models. The XGBoost Tree model generally outperformed other models for prediction of SOC and SOCD (R2 = 0.60, on average). Regardless of soil subgroups, the uncertainty analysis showed that the SOCD map had a low prediction interval range value indicating high accuracy. Additionally, the highest SOCS and SOCD was observed at the top followed by middle and bottom depths in the study area. All subgroups exhibited a decreasing trend of SOCD with increasing depth. A similar trend was also observed for SOCS. The highest SOCD (on average) was observed in Gypsic Haploxerepts (4.71 kg C/m2) followed by Typic Calcixerepts (4.46 kg C/m2), Typic Haploxerepts (4.45 kg C/m2) and Xeric Haplocalcids (4.40 kg C/m2). Overall, the SOCS normalized by area within soil order boundaries was greater in Inceptisols than Aridisols across the study area. The findings of this study provide critical information for sustainable management of soil resources in the area for agricultural production and environmental health in the Miandoab region of Iran.

show abstract

Comparing Machine Learning Models and Hybrid Geostatistical Methods Using Environmental and Soil Covariates for Soil pH Prediction

Cited by 21 publications

References 36 publications

Assessing the geographic specificity of pH prediction by classification and regression trees

Assessing the geographic specificity of pH prediction by classification and regression trees

Spatial prediction of soil organic carbon stocks in an arid rangeland using machine learning algorithms

Digital assessments of soil organic carbon storage using digital maps provided by static and dynamic environmental covariates

Contact Info

Product

Resources

About