Insights from big spatial data through machine learning techniques for prudent management of natural resources Ramachandra T V1,2,3,*, Negi Paras1,4, Setturu Bharath1 1Energy and Wetlands Research Group, Environmental Information System (EIACP), Center for Ecological Sciences, New Bioscience Building, Third Floor, E-Wing [Near D-Gate], Indian Institute of Science, Bangalore, 560012, Karnataka, India 2Centre for Sustainable Technologies (ASTRA), Indian Institute of Science, Bangalore, 560 012, Karnataka, India 3Centre For Infrastructure, Sustainable Transportation and Urban Planning, Indian Institute of Science, Bangalore, 560 012, Karnataka, India 4Department of Remote Sensing and GIS, Soban Singh Jeena University, Almora, Uttarakhand, India *Email: tvr@iisc.ac.in
Online published on 20 June, 2023. Abstract Evaluation of Land Use Land Cover (LULC) changes play a vital role in understanding the landscape dynamics that have been influencing climate, biodiversity, hydrology, and ecology of a region. The information of temporal LULC aids decision-makers in framing sustainable land use policies for nature conservation. Anthropogenic pressure, especially unplanned developmental activities, has contributed towards fragmenting contiguous forests, thus affecting their structure and loss of habitat for endemic taxa. LULC changes in the Bellary district, Karnataka have been assessed through temporal remote sensing data. Classification of remote sensing data for estimating the spatial extent of land uses has been done through supervised machine learning algorithms namely random forest (RF), support vector machine (SVM), and parametric maximum likelihood classifier (MLC). The performance of these algorithms was evaluated through accuracy assessments. Results reveal that RF has the highest overall accuracy (88.94%) and Kappa value (0.76) compared to overall Kappa of MLC (85.51%, 0.74) and SVM (85.47%, 0.63). Based on this, RF was considered for temporal data analyses, which highlighted the decline of forest cover from 2.61% (1973) to 0.74% (2022). The built-up has increased from 0.27% (1973) to 2.43% (2022), and agriculture from 68.21% (1973) to 84.95% (2022). Fragmentation of contiguous forests is evident from the decline in the interior or intact forests from 6.73% (1973) to 2.41% (2022) and the increase in the non-forest areas such as built-up, agriculture, etc., amounting now to 89.81%. Results highlight the need for immediate policy interventions for the conservation and protection of the remnant forest patches. Top Keywords LULC, Forest fragmentation, Supervised learning techniques, Machine learning. Top |