Interpretable regularized modelling of domain balanced soil fertility for spatial nutrient prioritization in data limited paddy systems
Abstract
Spatially explicit soil fertility diagnosis is still difficult in data sparse paddy cultivated areas because single soil test indicators cannot separate soil chemical condition, primary nutrient supply, micronutrient balance and yield relevance. This study developed a soil-test driven, domain-balanced Soil Fertility Index (SFI) framework using 64 georeferenced long-term paddy-field samples from Hanumakonda district, Telangana, India. Ten laboratory indicators (pH, electrical conductivity, organic carbon, available nitrogen, phosphorus, potassium, zinc, iron, copper, and manganese) were transformed into agronomic scores and aggregated into Chemical Condition, Macronutrient, Micronutrient, and Overall Soil Fertility indices. Regularized regression models were used to approximate the derived fertility-index targets from the original soil-test variables and to identify the conditional contribution of individual indicators within each index formulation. Ridge, Lasso and ElasticNet regression models were evaluated using five-fold cross-validation within the training dataset ( n = 51) and independent test-set evaluation ( n = 13). Among the evaluated models, Lasso provided the best prediction of Overall SFI (R ² = 0.594, RMSE = 0.033, RPD = 1.63), whereas ElasticNet performed best for the Micronutrient Index (R ² = 0.683, RPD = 1.85) and showed stable prediction of the Macronutrient Index (R ² = 0.577, RPD = 1.60) Relative zone mapping showed that low to moderate priority classes dominated the predicted overall SFI surface, while the Micronutrient Index produced the sharpest local management contrast. The predicted fertility surfaces were mapped across approximately 140,387 ha within the empirical point-covered sampling domain using a 30 m spatial analysis grid. Independent yield validation supported the agronomic relevance of the framework. Available nitrogen and organic carbon were strongly correlated with yield. Yield validation showed that predicted MNI relative zones significantly separated productivity classes, with mean yield increasing from 22.52 to 26.81 q acre ⁻¹ (Kruskal-Wallis H = 17.52, p < 0.001), whereas Overall SFI, MiNI and SFI 5core zones did not show significant yield separation. The proposed framework provides an interpretable and reproducible decision-support approach for prioritizing soil testing and site-specific nutrient management in data-sparse paddy landscapes.
Related
- Collection-informed environmental prioritization of unused household pharmaceuticals using FASS 2025 classification, risk quotients, and TOPSIS-based decision analysis
- Integrating land-use dynamics, water quality, and multi-criteria ecological suitability for fish sanctuary prioritization in three major river basins of Bangladesh
- Well posedness of a regularized-Hibler model of sea-ice dynamics
- Final step of the 32-year Lake Nyos degassing adventure: Natural CO2 recharge is to be balanced by discharge through the degassing pipes
- Characteristics and consequences of H2O undersaturation and low melt fertility in mountain belt interiors
BibTeX
@article{venkateswarlu2026interpretable,
title = {{Interpretable regularized modelling of domain balanced soil fertility for spatial nutrient prioritization in data limited paddy systems}},
author = {Meeniga Venkateswarlu and Chitteti Ravali and Vaibhav Bhagwan Pandit and Sumanta Chatterjee and Aminullah Noorzai},
journal = {Scientific Reports},
year = {2026},
month = oct,
doi = {10.1038/s41598-026-75319-0},
url = {https://www.nature.com/articles/s41598-026-75319-0},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Interpretable regularized modelling of domain balanced soil fertility for spatial nutrient prioritization in data limited paddy systems
AU - Meeniga Venkateswarlu
AU - Chitteti Ravali
AU - Vaibhav Bhagwan Pandit
AU - Sumanta Chatterjee
AU - Aminullah Noorzai
PY - 2026
DA - 2026/10/09
JO - Scientific Reports
DO - 10.1038/s41598-026-75319-0
UR - https://www.nature.com/articles/s41598-026-75319-0
AB - Spatially explicit soil fertility diagnosis is still difficult in data sparse paddy cultivated areas because single soil test indicators cannot separate soil chemical condition, primary nutrient supply, micronutrient balance and yield relevance. This study developed a soil-test driven, domain-balanced Soil Fertility Index (SFI) framework using 64 georeferenced long-term paddy-field samples from Hanumakonda district, Telangana, India. Ten laboratory indicators (pH, electrical conductivity, organic carbon, available nitrogen, phosphorus, potassium, zinc, iron, copper, and manganese) were transformed into agronomic scores and aggregated into Chemical Condition, Macronutrient, Micronutrient, and Overall Soil Fertility indices. Regularized regression models were used to approximate the derived fertility-index targets from the original soil-test variables and to identify the conditional contribution of individual indicators within each index formulation. Ridge, Lasso and ElasticNet regression models were evaluated using five-fold cross-validation within the training dataset ( n = 51) and independent test-set evaluation ( n = 13). Among the evaluated models, Lasso provided the best prediction of Overall SFI (R ² = 0.594, RMSE = 0.033, RPD = 1.63), whereas ElasticNet performed best for the Micronutrient Index (R ² = 0.683, RPD = 1.85) and showed stable prediction of the Macronutrient Index (R ² = 0.577, RPD = 1.60) Relative zone mapping showed that low to moderate priority classes dominated the predicted overall SFI surface, while the Micronutrient Index produced the sharpest local management contrast. The predicted fertility surfaces were mapped across approximately 140,387 ha within the empirical point-covered sampling domain using a 30 m spatial analysis grid. Independent yield validation supported the agronomic relevance of the framework. Available nitrogen and organic carbon were strongly correlated with yield. Yield validation showed that predicted MNI relative zones significantly separated productivity classes, with mean yield increasing from 22.52 to 26.81 q acre ⁻¹ (Kruskal-Wallis H = 17.52, p < 0.001), whereas Overall SFI, MiNI and SFI 5core zones did not show significant yield separation. The proposed framework provides an interpretable and reproducible decision-support approach for prioritizing soil testing and site-specific nutrient management in data-sparse paddy landscapes.
SN - 2045-2322
ER -