Spatial prediction of soil organic carbon stocks across contrasting Andean basins, Peru

Soil organic carbon stocks (SOCS) are critical components of the global carbon cycling and play a central role in climate change mitigation. However, their dynamics in high‐altitude Andean ecosystems remain poorly understood despite their importance for carbon sequestration. The significant spatial...

Full description

Bibliographic Details
Main Authors: Carbajal Llosa, Carlos Miguel, Tumbalobos Dextre, Merely, Condori Ataupillco, Levi Tatiana, Cuellar Condori, Nestor Edwin, Gavilan, Carla
Format: Preprint
Language:Inglés
Published: Elsevier B.V. 2025
Subjects:
Online Access:http://hdl.handle.net/20.500.12955/2952
https://doi.org/10.1016/j.geodrs.2025.e01026
_version_ 1855490558650220544
author Carbajal Llosa, Carlos Miguel
Tumbalobos Dextre, Merely
Condori Ataupillco, Levi Tatiana
Cuellar Condori, Nestor Edwin
Gavilan, Carla
author_browse Carbajal Llosa, Carlos Miguel
Condori Ataupillco, Levi Tatiana
Cuellar Condori, Nestor Edwin
Gavilan, Carla
Tumbalobos Dextre, Merely
author_facet Carbajal Llosa, Carlos Miguel
Tumbalobos Dextre, Merely
Condori Ataupillco, Levi Tatiana
Cuellar Condori, Nestor Edwin
Gavilan, Carla
author_sort Carbajal Llosa, Carlos Miguel
collection Repositorio INIA
description Soil organic carbon stocks (SOCS) are critical components of the global carbon cycling and play a central role in climate change mitigation. However, their dynamics in high‐altitude Andean ecosystems remain poorly understood despite their importance for carbon sequestration. The significant spatial heterogeneity of SOCS in mountainous terrain makes accurate quantification and mapping challenging. This study evaluated the performance of geospatial regression and machine learning (ML) approaches for predicting SOCS in two Peruvian Andean basins: Torobamba and Coata. We compared Geographically Weighted Regression (GWR), GWR with collinearity analysis (GWRC), their kriging‐adjusted variants, and ML models (Random Forest, Gradient Boosting). Models were built using key SOCS covariates for each basin and validated through 5‐fold cross‐validation with Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and coefficient of determination (R²). In Torobamba, GWRC markedly improved performance, reducing the RMSE by 79–90% and achieving R² up to 0.99. In contrast, Coata, showed only modest improvements (RMSE reductions of 7.8–9.8%, R² = 0.30–0.39). ML models performed poorly (negative R²), likely due to feature selection, parameter tuning, or limited sample size. Overall, locally weighted regression approaches (GWRK/GWRCK) outperformed conventional ML methods for SOCS prediction in complex mountain environments, particularly with small to medium sample sizes. These results highlight the importance of accounting for spatial non‐stationarity in SOCS and provide methodological guidance for SOCS mapping in Andean ecosystems.
format Preprint
id INIA2952
institution Institucional Nacional de Innovación Agraria
language Inglés
publishDate 2025
publishDateRange 2025
publishDateSort 2025
publisher Elsevier B.V.
publisherStr Elsevier B.V.
record_format dspace
spelling INIA29522025-12-03T16:57:54Z Spatial prediction of soil organic carbon stocks across contrasting Andean basins, Peru Carbajal Llosa, Carlos Miguel Tumbalobos Dextre, Merely Condori Ataupillco, Levi Tatiana Cuellar Condori, Nestor Edwin Gavilan, Carla Digital soil mapping Soil organic carbon stock Geographically weighted regression Machine learning regression algorithms Andes Cartografía digital de suelos Reservas de carbono orgánico del suelo Regresión ponderada geográficamente Algoritmos de regresión de aprendizaje automático https://purl.org/pe-repo/ocde/ford#4.01.04 Regresión de paso cauteloso; stepwise regression; Cuenca hidrográfica; Watersheds; Perú; Peru Soil organic carbon stocks (SOCS) are critical components of the global carbon cycling and play a central role in climate change mitigation. However, their dynamics in high‐altitude Andean ecosystems remain poorly understood despite their importance for carbon sequestration. The significant spatial heterogeneity of SOCS in mountainous terrain makes accurate quantification and mapping challenging. This study evaluated the performance of geospatial regression and machine learning (ML) approaches for predicting SOCS in two Peruvian Andean basins: Torobamba and Coata. We compared Geographically Weighted Regression (GWR), GWR with collinearity analysis (GWRC), their kriging‐adjusted variants, and ML models (Random Forest, Gradient Boosting). Models were built using key SOCS covariates for each basin and validated through 5‐fold cross‐validation with Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and coefficient of determination (R²). In Torobamba, GWRC markedly improved performance, reducing the RMSE by 79–90% and achieving R² up to 0.99. In contrast, Coata, showed only modest improvements (RMSE reductions of 7.8–9.8%, R² = 0.30–0.39). ML models performed poorly (negative R²), likely due to feature selection, parameter tuning, or limited sample size. Overall, locally weighted regression approaches (GWRK/GWRCK) outperformed conventional ML methods for SOCS prediction in complex mountain environments, particularly with small to medium sample sizes. These results highlight the importance of accounting for spatial non‐stationarity in SOCS and provide methodological guidance for SOCS mapping in Andean ecosystems. To the Soil, Water, and Foliar Laboratory (LABSAF) network technicians, especially of La Molina, Canaan, ´ and Illpa Experimental Agrarian Stations headquarters. Special thanks go to Marilia Coila Mamani and Fredy Flores Galindo for their help collecting soil samples 2025-12-03T15:01:44Z 2025-12-03T15:01:44Z 2025-11-06 info:eu-repo/semantics/preprint Carbajal, C., Tumbalobos-Dextre, M., Condori-Ataupillco, T., Cuellar-Condori, N., & Gavilan, C. (2025). Spatial prediction of soil organic carbon stocks across contrasting Andean basins, Peru. Geoderma Regional, e01026. https://doi.org/10.1016/j.geodrs.2025.e01026 2352-0094 http://hdl.handle.net/20.500.12955/2952 https://doi.org/10.1016/j.geodrs.2025.e01026 eng urn:issn:2352-0094 Geoderma Regional info:eu-repo/semantics/openAccess https://creativecommons.org/licenses/by/4.0/ application/pdf application/pdf Elsevier B.V. NL Instituto Nacional de Innovación Agraria Repositorio Institucional - INIA
spellingShingle Digital soil mapping
Soil organic carbon stock
Geographically weighted regression
Machine learning regression algorithms
Andes
Cartografía digital de suelos
Reservas de carbono orgánico del suelo
Regresión ponderada geográficamente
Algoritmos de regresión de aprendizaje automático
https://purl.org/pe-repo/ocde/ford#4.01.04
Regresión de paso cauteloso; stepwise regression; Cuenca hidrográfica; Watersheds; Perú; Peru
Carbajal Llosa, Carlos Miguel
Tumbalobos Dextre, Merely
Condori Ataupillco, Levi Tatiana
Cuellar Condori, Nestor Edwin
Gavilan, Carla
Spatial prediction of soil organic carbon stocks across contrasting Andean basins, Peru
title Spatial prediction of soil organic carbon stocks across contrasting Andean basins, Peru
title_full Spatial prediction of soil organic carbon stocks across contrasting Andean basins, Peru
title_fullStr Spatial prediction of soil organic carbon stocks across contrasting Andean basins, Peru
title_full_unstemmed Spatial prediction of soil organic carbon stocks across contrasting Andean basins, Peru
title_short Spatial prediction of soil organic carbon stocks across contrasting Andean basins, Peru
title_sort spatial prediction of soil organic carbon stocks across contrasting andean basins peru
topic Digital soil mapping
Soil organic carbon stock
Geographically weighted regression
Machine learning regression algorithms
Andes
Cartografía digital de suelos
Reservas de carbono orgánico del suelo
Regresión ponderada geográficamente
Algoritmos de regresión de aprendizaje automático
https://purl.org/pe-repo/ocde/ford#4.01.04
Regresión de paso cauteloso; stepwise regression; Cuenca hidrográfica; Watersheds; Perú; Peru
url http://hdl.handle.net/20.500.12955/2952
https://doi.org/10.1016/j.geodrs.2025.e01026
work_keys_str_mv AT carbajalllosacarlosmiguel spatialpredictionofsoilorganiccarbonstocksacrosscontrastingandeanbasinsperu
AT tumbalobosdextremerely spatialpredictionofsoilorganiccarbonstocksacrosscontrastingandeanbasinsperu
AT condoriataupillcolevitatiana spatialpredictionofsoilorganiccarbonstocksacrosscontrastingandeanbasinsperu
AT cuellarcondorinestoredwin spatialpredictionofsoilorganiccarbonstocksacrosscontrastingandeanbasinsperu
AT gavilancarla spatialpredictionofsoilorganiccarbonstocksacrosscontrastingandeanbasinsperu