Geospatial mapping of soil organic carbon in Sirsi Forst division using remote sensing techniques

Introduction

Soil Organic Carbon (SOC) is a critical component of soil organic matter and a key indicator of soil health, integral to various ecological processes such as soil fertility, water retention and nutrient cycling (Lal, 2004). SOC plays a significant role in global carbon cycling and climate change mitigation, as it serves as a dynamic reservoir that can absorb or release carbon dioxide (CO₂) depending on management practices [1,2] . In forested ecosystems, SOC stocks are substantial due to continuous organic material input and minimal disturbance, making soils the largest and most stable carbon pool in forests [3] . Factors influencing SOC accumulation include climatic conditions, soil properties and vegetation type, with cooler and wetter regions generally exhibiting higher SOC levels due to slower decomposition rates [4,5] . Understanding SOC dynamics is essential for effective carbon sequestration and climate change mitigation [6,7] . Topography significantly influences Soil Organic Carbon (SOC) distribution by affecting water movement, erosion and deposition, with slope, aspect and elevation being key factors [8,9] . Accurate SOC assessment is vital for effective land resource management, environmental monitoring and climate change mitigation. Traditional methods such as the Walkley-Black wet oxidation and dry combustion are commonly used for SOC measurement. The Walkley-Black method is cost-effective but may underestimate SOC in high organic matter soils, whereas dry combustion offers greater accuracy but is labour-intensive and requires specialized equipment [10,11] . Recent advancements include digital soil mapping (DSM) techniques, which use remote sensing data, Geographic Information Systems (GIS) and machine learning algorithms to estimate SOC across large areas. DSM approaches utilize environmental covariates like topography, climate and vegetation indices to model SOC distribution [12] .

Geostatistical methods and machine learning algorithms, such as Random Forest (RF), Support Vector Machines (SVM) and Artificial Neural Networks (ANN), enhance SOC predictions by capturing complex, non-linear relationships [13] . Remote sensing technologies, including satellite imagery from Landsat, Sentinel-2 and MODIS, provide spatially continuous data on vegetation cover, soil moisture and land use changes, with vegetation indices like NDVI and NDMI reflecting SOC levels [14,15,16,17,18] . The integration of these techniques, particularly the RF model, has shown promise in improving SOC prediction accuracy and reliability compared to traditional methods [19] .

Methodology

To map Soil Organic Carbon (SOC) stock in various forest types within Uttara Kannada district, Sentinel-2 and Shuttle Radar Topography Mission (SRTM) data were employed, utilizing machine learning algorithms for predictive modeling. Sentinel-2, part of the European Space Agency's Copernicus program, provides high-resolution optical imagery, while SRTM offers elevation data essential for terrain analysis.

SRTM Data:

The SRTM digital elevation model (DEM) with 30 m spatial resolution was pre-processed using Google Earth Engine (GEE). Key terrain attributes, including elevation, slope and aspect, were derived from this DEM. These attributes are crucial for understanding terrain effects on SOC distribution [8, 9] .

Sentinel-2 Data:

Harmonized Sentinel-2 MSI Level-2A data was used, which includes 13 spectral bands spanning visible to shortwave infrared wavelengths. Bands 1, 9 and 10 were excluded due to their limited relevance for terrestrial studies. The Sentinel-2 bands utilized with a focus on those most applicable for SOC modeling.

Preprocessing:

The SRTM data was normalized and terrain attributes were calculated in GEE. For Sentinel-2 data, preprocessing involved normalization of pixel values to a 0-1 range using the "normalize" function. Cloud masking was applied with the "maskS2clouds" function and image collections were filtered by date (2024-01-01 to 2024-03-31) and cloud coverage (<50%). Median computation was performed to reduce outlier effects and the resultant image was clipped to the study area (Sirsi Forest Division).  LSWI (Land Surface Water Index): Calculated as (NIR -SWIR) / (NIR + SWIR), using bands 8A (NIR) and 11 (SWIR).  BSI (Bare Soil Index): Computed using ((SWIR + RED) -(NIR + BLUE)) / ((SWIR + RED) + (NIR + BLUE)), with bands 11 (SWIR), 4 (RED) and 2 (BLUE).

Predictor

Combined Data Extraction:

All image layers, including SRTM data, terrain attributes, Sentinel-2 bands and indices, were merged into a single image. This image was reprojected to EPSG:4326 and exported for further analysis in R.

Machine Learning for SOC Prediction

Raster values were extracted at field point locations using the "extrac t()" function, merging these with field data. The dataset was split into training (70%) and testing (30%) sets using 'createDataPartition()'. Predictors included Sentinel-2 bands and vegetation indices, with SOC as the response variable.

The Random Forest (RF) machine learning algorithm was used for SOC prediction. Random Forest was chosen for its ability to handle complex interactions between predictors and SOC. RF operates by building multiple decision trees from bootstrap samples of the data and averaging their predictions, thus managing complex relationships and improving prediction accuracy [20,21] . This approach ensures a comprehensive and robust analysis of SOC distribution across different forest types using advanced remote sensing and machine learning techniques.

Results and Discussion

Correlation of SOC Stock and predictive variables

Understanding the associations between SOC stock with each predictive covariate is crucial for assessing the contribution of each variable in the model. Figure 1. Represents the results of Pearson's correlation analysis, revealed that the relationships between SOC stock and the predictive variables. The results showed that the SOC stock is positively correlated with slope (0.12), B2 (0.04), B3 (0.07), B4 (0.14), B6 (0.03), B7 (0.04), B8 (0.35), B8A (0.003), NDVI (0.14), EVI (0.07), SAVI (0.05), NDMI (0.11), GNDVI (0.11) and LSWI (0.13), but negatively correlated with elevation (-0.13), aspect (-0.3), B5 (-0.02), B11 (-0.17), B12 (-0.16), and BSI (-0.13). These results indicate positive correlations among SOC stock and spectral bands, environmental covariates and spectral indices such as B2, B3, B6, B7, B8 and B8A. Similar results were reported by Duarte et al. 2022 [22] showed soil organic carbon stock was positively correlated with B blue , B green and B red. Badgley et al. 2017 [23] reported B8 band (NIR) was positively correlated because of its importance in vegetation modelling.

The geographic distribution of various predictive variables used in mapping SOC stock, including elevation, slope, aspect, NDVI, EVI, SAVI were represented in Figure 2, GNDVI, NDMI, LSWI and BSI were represented in Figure 3. The elevation in the study area ranges from 230 to 675 meters, while the slope varies from 0.00° to 44.59°. Aspect is represented on a scale of 1:1. NDVI values ranges from 0.4 to 88, reflecting significant variability in vegetation cover across the region. EVI values range between 0.12 and 1. Other predictive variables have the following ranges: SAVI (0.13 to 0.79), NDMI (-0.31 to 0.52), GNDVI (0.31 to 0.8), BSI (-0.46 to 0.29) and LSWI (-0.23 to 0.5) in the study area, with these results and several studies have been indicated NDVI [24, 25], EVI and SAVI are positively correlated [6,7] many studies have been depicted that soil organic carbon stock decreased with elevation [35] .

Increase in Node Purity (IncNodPurity)

The total decrease in node impurity from splitting on a variable, averaged over all trees were depicted in Figure 4. B11, NDVI and LSWI highest IncNodPurity, indicating they significantly improve the homogeneity of the nodes when used for splits. B3, B2 and B8A have the lowest IncNodePurity, indicate they are less critical for making splits. The ranking of variable importance for predicting soil organic carbon (SOC) stocks was depicted in Figure 4. Using the Random Forest (RF) model which scaled from 0 to 150. Among the predictors, the Normalized Difference Vegetation Index (NDVI) is identified as the most influential, showed an importance value of 100. This finding highlights that SOC stocks are strongly influenced by variations in vegetation cover.

The levels of SOC are closely associated with the density of natural vegetation and the remaining plant residues after harvesting, as supported by previous studies [17, 18, 27]. Other significant predictors with importance values above 75 include the Sentinel-2 Band 11, Land Surface Water Index (LSWI), Band 12 (B12), Bare Soil Index (BSI), Normalized Difference Moisture Index (NDMI), Band 7 and Enhanced vegetation index (EVI). Furthermore, the Soil-Aspect, Band 6, Band 4, Adjusted Vegetation Index (SAVI), Band 8, elevation, slope, Band 5, Gross Normalized difference vegetation index (GNDVI), Band 8A, Band2 and Band 3are also notable predictors, each with importance values exceeding 50. The application of the Normalized Difference Vegetation Index (NDVI) has been instrumental in highlighting the role of biomass and vegetation cover in predicting soil organic carbon (SOC) stocks [28] . Previous studies have demonstrated that SOC levels can be effectively estimated by considering vegetation presence alone [23,29] . Thus, NDVI serves as a reliable proxy for assessing SOC content. Furthermore, incorporating the Bare Soil Index (BSI) in the analysis underscores the relevance of examining bare soil conditions, which is particularly important in areas with highly fragmented secondary forests [22] . Additionally, topographic variables such as elevation and slope have been consistently found to be strongly associated with variations in soil organic carbon (SOC) in studies by Boettinger et al. (2010) and Hinge et al. (2018) [8,9] .

SOC stock in the western region as compared to the eastern region of the study area. SOC stock was highest in evergreen forests, followed by semi-evergreen and moist deciduous forest. The accurate estimation of soil organic carbon and its stock is crucial because it indicates the soil's ability to capture atmospheric carbon and function effectively as a carbon sink. This information is essential for various applications, including land resource monitoring, land evaluation and environmental modelling, as it provides insights into the carbon sequestration potential of soils. Utilizing digital soil mapping (DSM) techniques improves the accuracy and robustness of SOC estimates by integrating multiple data sources and environmental variables. These techniques generate high-quality spatial data that are invaluable for ongoing assessments of SOC dynamics and for advancing research in environmental and soil sciences [30] .

The study used machine learning algorithm, specifically Random Forest (RF) [31,32] to estimate soil organic carbon (SOC)

stocks in forested areas of the Sirsi forest division. Areas classified as non-forest were excluded from the SOC stock analysis and are showed as uncoloured regions on the maps. According to the RF algorithm, the SOC stock from 0-30 cm, 30-60 cm and 0-60 cm within the study area varied between 59.45 Mgha -1 to 76.59 Mgha -1 , 53. [6] carried out a study in the semi-arid zone of South India to examine the spatial distribution of total soil organic carbon stocks up to a certain depth of 30 cm. They found that the Western Ghats region in Karnataka, which is characterized by high rainfall, elevated terrain, dense vegetation and limited human activity, had substantially higher total carbon stocks, ranging between 8 and 14 kg m -2 . These findings are consistent with the results reported by Lo Seen et al. (2010) [33] which highlighted the impact of favourable climatic and ecological conditions on carbon sequestration in similar landscapes.

Conclusion

This study provides variation and amount of soil organic carbon stock among different depths in sirsi forest division. Through geospatial mapping of SOC stock contributes to the growing body of knowledge on digital soil mapping and remote sensing for SOC assessment, demonstrating its potential to aid in sustainable land management and climate change mitigation efforts. The methodologies and insights presented here can serve as a foundation for future research and policy-making focused on soil carbon management in forested and other terrestrial ecosystems.

Acknowledgements

We would like acknowledge Gowri B N without her presence, provided all the support and guidance.

Competing interests

Authors have declared that no competing interests exist

Authors' Contributions

This work was carried out in collaboration among all authors.

References

1. Batjes NH. Total carbon and nitrogen in the soils of the world. European Journal of Soil Science. 1996;47(2):151163.
2. Trumbore S. Radiocarbon and soil carbon dynamics. Annual Review of Earth and Planetary Sciences. 2009;37(1):47-66.
3. Sahoo UK, Singh SL, Gogoi A, Kenye, Sahoo SS. Active and passive soil organic carbon pools as affected by different land use types in Mizoram, Northeast India. PloS One. 2019;14(7):0219969.
4. Jobbágy EG, Jackson RB. The vertical distribution of soil organic carbon and its relation to climate and vegetation. Ecological Applications. 2000;10(2):423-436.
5. Conant RT, Ogle SM, Paul EA, Paustian K. Measuring and monitoring soil organic carbon stocks in agricultural lands for climate mitigation. Frontiers in Ecology and the Environment. 2011;9(3):169-173.
6. Mitran T, Mishra U, Lal R, Ravisankar T, Sreenivas K. Spatial distribution of soil carbon stocks in a semi-arid region of India. Geoderma Regional., 2018, 15.
7. Chatterjee A, Lal R. On-farm assessment of tillage impact on soil carbon and associated soil quality parameters. Soil and Tillage Research. 2009;104(2):270-277.
8. Hinge G, Surampalli RY, Goyal MK. Prediction of soil organic carbon stock using digital mapping approach in humid India. Environmental Earth Sciences. 2018;77:1-10.
9. Schillaci C, Lombardo L, Saia S, Fantappiè M, Märker M, Acutis M, et al. Modelling the topsoil carbon stock of agricultural lands with the Stochastic Gradient Treeboost in a semi-arid Mediterranean region. Geoderma. 2017;286:3545.
10. Fernandes MMH, Coelho AP, Fernandes C, da Silva MF, Marta CCD. Estimation of soil organic matter content by modeling with artificial neural networks. Geoderma. 2019;350:46-51.
11. Malone BP, Styc Q, Minasny B, McBratney AB. Digital soil mapping of soil carbon at the farm scale: A spatial downscaling approach in consideration of measured and uncertain data. Geoderma. 2017;290:91-99.
12. Wiesmeier M, Barthold F, Blank B, Kögel-Knabner I. Digital mapping of soil organic matter stocks using Random Forest modeling in a semi-arid steppe ecosystem. Plant and Soil. 2011;340:7-24.
13. Duarte E, Barrera JA, Dube F, Casco F, Hernández AJ, Zagal E, et al. Monitoring approach for tropical coniferous forest degradation using remote sensing and field data. Remote Sensing. 2020;12(16):2531.
14. Wulder MA, White JC, Loveland TR, Woodcock CE, Belward AS, Cohen WB, et al. The global Landsat archive: Status, consolidation, and direction. Remote Sensing of Environment. 2016;185:271-283.
15. Minasny B, McBratney AB, Malone BP, Wheeler I. Digital mapping of soil carbon. Advances in Agronomy. 2013;118:1-47.
16. Xiong X, Grunwald S, Myers DB, Kim J, Harris WG, Bliznyuk N, et al. Assessing uncertainty in soil organic carbon modeling across a highly heterogeneous landscape. Geoderma. 2015;251:105-116.
17. Behrens T, Zhu AX, Schmidt K, Scholten T. Multi-scale digital terrain analysis and feature selection for digital soil mapping. Geoderma. 2010;155(3-4):175-185.
18. Fu B, Wang Y, Campbell A, Li Y, Zhang B, Yin S, et al. Comparison of object-based and pixel-based Random Forest algorithm for wetland vegetation mapping using high spatial resolution GF-1 and SAR data. Ecological Indicators. 2017;73:105-117.
19. Duarte E, Zagal E, Barrera JA, Dube F, Casco F, Hernández AJ. Digital mapping of soil organic carbon stocks in the forest lands of Dominican Republic. European Journal of Remote Sensing. 2022;55(1):213-231.
20. Badgley G, Field CB, Berry JA. Canopy near-infrared reflectance and terrestrial photosynthesis. Science Advances. 2017;3(3):1602244.
21. John K, Isong IM, Kebonye NM, Ayito E, Agyeman P, Afu SM, et al. Using machine learning algorithms to estimate soil organic carbon variability with environmental variables and soil nutrient indicators in an alluvial soil. Land. 2020;9(12):487.
22. Wang B, Waters C, Orgill S, Gray J, Cowie A, Clark A, et al. High resolution mapping of soil organic carbon stocks using remote sensing variables in the semi-arid rangelands of eastern Australia. Science of the Total Environment. 2018;630:367-378.
23. Zhao YC, Shi XZ. Spatial prediction and uncertainty assessment of soil organic carbon in Hebei Province, China. In: Digital Soil Mapping: Bridging Research, Environmental Application and Operation. 2010;1:227-239.
24. Yang Y, Fang J, Tang Y, Ji C, Zheng C, He J, et al. Storage, patterns and controls of soil organic carbon in the Tibetan grasslands. Global Change Biology. 2008;14(7):1592-1599.
25. Jan B, Jeffrey U. Achieving the United Nations sustainable development goals: An enabling role for accounting research. Accounting, Auditing & Accountability Journal. 2018;31(1):2-24.
26. Grimm R, Behrens T, Märker M, Elsenbeer H. Soil organic carbon concentrations and stocks on Barro Colorado IslandDigital soil mapping using Random Forests analysis. Geoderma. 2008;146(1-2):102-113.
27. Padarian J, Minasny B, McBratney AB. Using deep learning for digital soil mapping. Soil. 2019;5(1):79-89.
28. Lo Seen D, Ramesh BR, Nair KM, Martin M, Arrouays D, Bourgeon G, et al. Soil carbon stocks, deforestation and land-cover changes in the Western Ghats biodiversity hotspot (India). Global Change Biology. 2010;16(6):17771792.
29. Dharumarajan S, Kalaiselvi B, Suputhra A, Lalitha M, Vasundhara R, Kumar KA, et al. Digital soil mapping of soil organic carbon stocks in Western Ghats, South India. Geoderma Regional. 2021;25:00387.
30. Wang B, Waters C, Orgill S, Gray J, Cowie A, Clark A, et al. High resolution mapping of soil organic carbon stocks using remote sensing variables in the semi-arid rangelands of eastern Australia. Science of the Total Environment. 2018;630:367-378.