Digital mapping of soil organic and inorganic carbon status in India

• Published in: Geoderma Vol. 269; pp. 160 - 173
• Main Authors: Sreenivas, Kandrika, Dadhwal, V.K., Kumar, Suresh, Harsha, G. Sri, Mitran, Tarik, Sujatha, G., Suresh, G. Janaki Rama, Fyzee, M.A., Ravisankar, T.
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.05.2016
• Subjects: Carbon; Carbon density; Carbon stock; Data mining; Density; Estimates; India; Mathematical models; Pools; Random forests; Raw materials; Soil (material); Soil carbon; Spatial prediction; India; Carbon stock; Spatial prediction; Data mining; Carbon density; Random forests; Soil carbon
• ISSN: 0016-7061; 1872-6259
• DOI: 10.1016/j.geoderma.2016.02.002

Reliable mapping of soil organic carbon (SOC) and Soil Inorganic Carbon (SIC) densities and estimates of their pool size are important from global warming perspective to understand the sequestration potential and losses. In this study, first spatially explicit mapping of SOC and SIC at 250m resolution and an estimate of their pool size in India was undertaken using a large number of remote sensing derived data layers and data mining approach. The SOC and SIC densities up to 100cm depth or paralithic contact (whichever is shallower) were estimated for 1198 soil samples located across India using a stratified random sampling that integrated land use, soil, topography and agro-ecological regions. Using Random forests (RF) based spatial prediction procedure with climatic, land cover, rock type, soil type, multi-year NDVI, irrigation status as independent input variables, models for predicting carbon density at 250m spatial resolution were developed. For modelling with RF algorithm, about 898 soil profile observations (75% observations) were used, while the rest of 300 (25% of total observations) were used for validation. It was observed that the data distribution of sample points don't have significant influence on RF model predictions. The relationship between observed and predicted values was characterized by Mean Squared Deviation (MSD) and Root Mean Squared Error (RMSE) parameters. The SOC, SIC and total soil carbon pool size of India has been estimated at 22.72±0.93 Pg,12.83±1.35 Pg and 35.55±1.87 Pg, respectively, which are comparable to previous studies while providing first spatially explicit 250m map of their distribution. The spatial distribution indicates that majority of the carbon stock resides in the northern part of India. The soil carbon stock of eastern India has contribution from organic carbon, while the western portion has contribution mainly from inorganic carbon. •Modelling carbon densities using Random forests based data mining•Influence of sample data distribution on model performance•Important variables: land use for organic and temperature for inorganic carbon•Digital mapping of soil organic and inorganic carbon densities•Estimation of soil organic and inorganic carbon stocks for India