Monitoring of Spatio-Temporal Carbon Stock Variation in Dudhwa Tiger Reserve, Uttar Pradesh, India using Remote Sensing and Machine Learning Based Approach

Ghosh Sayanta , Soni Aniruddh , Sharma Jitendra Vir
International Workshop on AI-driven Spatio-temporal Data Analysis for Wildlife Conservation, November 2023, Pages 25–26,

Temporal variation in forest cover, the largest terrestrial ecosystem on Earth, influences the climate at both local, regional, and global scales through physical, chemical, and biological processes. At the same time, forests sequester and store more carbon dioxide than any other terrestrial ecosystem and act as a "natural brake" in climate variation. Here, we have made an attempt to assess the spaio-temporal variation in forest biomass combining field-based and remote sensing and machine learning approaches. For this purpose, Fractional Vegetation Cover (FVC) layers based on Linear Spectral Unmixing (LSU) Algorithm have been developed using cloud-free multi-temporal LANDSAT data for Dudhwa Tiger Reserve, Uttra Pradesh, India from the year 2001 to 2022. Linear Regression Model (LRM) have been developed between Field based forest biomass and FVC on the basis of field data collected from 60 sampling plots of 0.1 ha across three different forest strata, namely, Very Dense Forest (VDF), Moderate Dense Forest (MDF) and Open Forest (OF). LRM indicates strong positive correlation having R2 values 0.718 for VDF, 0.73 for MDF and 0.76 for OF forest strata. Also, the predicted biomass thus obtained shows strong positive correlation with observed biomass.

Results highlights that in VDF, carbon stock shows a decreasing trend till 2018 (332 t/ ha) since the year 2001 (347 t/ha) before further increase during present year (339 t/ha). Simultaneously, temporal variation in FVC also suggests the same trend for the forest cover under VDF strata which is playing pivotal role in increasing trend of forest carbon stock from the 2018 onwards. Also, we have compared between best possible FVC model based on three vegetation index (NDVI, MSAVI and EVI) which highlights the FVC model based on NDVI shows highly significant correlation (R2=0.73, p<0.005) with the field-based forest biomass. Degradation matrix also developed using the temporal FVC layers for the delineation of degradation patches and trend analysis of forest degradation. Outcome of the paper will be helpful for the policy makers in visualizing proper development plan to regulate the Land-use and forest cover dynamics for achieving the higher carbon sequestration rate which would in turn helps to maintain the balance in the global climate scenerio.

FVC, Forest Biomass, Linear Spectral Unmixing, Carbon Stock, NDVI, Linear Regression Model