RESEARCH COMMUNICATIONS CURRENT SCIENCE, VOL. 118, NO. 8, 25 APRIL 2020 1275 *For correspondence. (e-mail: bsharad0404@gmail.com) Detection and mapping of seagrass meadows at Ritchie’s archipelago using Sentinel 2A satellite imagery Sharad Bayyana 1, *, Satish Pawar 1 , Swapnali Gole 2 , Sohini Dudhat 2 , Anant Pande 2 , Debashis Mitra 3 , Jeyaraj Antony Johnson 4 and Kuppusamy Sivakumar 2 1 Indian Institute of Remote Sensing, Dehradun 248 001, India 2 Department of Endangered Species Management, Wildlife Institute of India, Dehradun 248 002, India 3 Marine and Atmospheric Science Department, Indian Institute of Remote Sensing, Dehradun 248 001, India 4 Department of Habitat Ecology, Wildlife Institute of India, Dehradun 248 002, India This study presents an attempt to utilize seagrass data acquired from field surveys to compare classification models for mapping seagrasses using Sentinel-2A satellite data. Out of three models tested, viz. Random Forest, Support Vector Machine and K-Nearest Neighbor; Random Forest classification model proved most effective in the given scenario with 0.99 model accuracy. Seagrasses present as deep as 21 m were de- tected post water column correction, presenting the capability of Sentinel-2A satellite in detecting sub- merged benthic habitat. Keywords: Depth Invariant Index, Ritchie’s archipelago, seagrass, Sentinel-2A. SEAGRASS meadows, one of the most productive ecosys- tems on the planet, are estimated to lose 7% of their global area annually 1 . Spatial data analysis for seagrass studies towards their sustainable management and con- servation has been an emerging field. Globally, satellite remote sensing tools have proven to be cost effective in comparison to conventional field surveys 2–8 and tradi- tional geospatial methods such as aerial photography 9 . Since, satellite sensors are repeatable in their path and are geometrically accurate, change detection in seagrass distribution over temporal scale is possible 10–12 . Landsat imagery has been efficiently used in seagrass and benthic substrate mapping, despite its spectral and spatial limita- tions 12–15 . Multispectral imagery from compact airborne spectrographic imager (CASI) with satellite imagery of Landsat and Spot, has been shown to exhibit more accurate results from airborne high-resolution sensor compared to aerial photography in classification of sub- merged benthic features including seagrasses 2 . Sensing of submerged benthic vegetation in the coastal waters is achieved with multispectral observations (400– 650 nm) of reflected radiance in the visible range which is enhanced with finer spatial resolution 16 . Certain regres- sion models developed for mapping benthic features have opened up the doors to overcome the limitations of atten- uation of radiance within the water column 17–19 . Assum- ing that variance in reflectance from same benthic substrate is primarily due to its presence at various depths and the diffused attenuation coefficient ( K d ) is same for all the bands 17,18 , regression from logarithmic values of individual bands provides proxy attenuation coefficients which are independent of depth 20 . Assessment of sub- merged sea grasses is reliable with remote sensing when appropriate correction (such as water depth correction) is applied to satellite images 21 . Medium resolution multi- spectral satellite images from Landsat OLI were effective in mapping of submerged benthic features with applica- tion of depth invariant index (DII), which is independent of depth effect 22 . High resolution multispectral imagery such as Sentinel-2A with 10 m spatial resolution has also proved effective to detect and estimate the cover of seagrass beds along the coast of Lombok in Indonesia 23 . The quality of results post DII when utilized for VHR Worldview-2 imageries was significantly high (up to 83% at Kotok Island in Indonesia) 24 . In India, seagrass are distributed along the coastline of nine states and two union territories with major patches found along Tamil Nadu (Palk Bay and Gulf of Mannar), Odisha, Gujarat, Lakshadweep Islands and Andaman and Nicobar Islands 25 . Remote sensing for the seagrass detection was first initiated at Lakshadweep islands to study the coral reefs and seagrass beds using black and white aerial photographs 26 . Later, loss of seagrass habi- tats in Gulf of Mannar group of islands due to anthropo- genic activities was detected using LISS III satellite imagery 27 . Seagrass area was estimated to be around 85.5 sq. km around the islands of Gulf of Mannar based on IRS-1D LISS III satellite data from 1998 (ref. 28). Earlier, a few studies have utilized conventional field survey methods to map seagrass ecosystems in the Andaman and Nicobar group of islands 29–31 . One study used satellite geospatial data (LISS III and LISS IV) for the mapping across the entire Andaman islands 32 . Seagrass meadows in Andaman and Nicobar Islands serve as foraging grounds for globally threatened species such as dugongs, green sea turtles 33 , and act as nurseries for several species of fish and invertebrates and thus support fisheries in the islands. In the light of proposed infra- structure developments in the islands 34 , understanding the extent of seagrass distribution in the islands will be use- ful in identifying critical areas to aid their conservation and management. In this study, we mapped the seagrass meadows at Ritchie’s archipelago (henceforth RA; 11 46N–1219N and 9254E–9308E) within the Andaman and Nicobar group of islands using multi-spectral imager (MSI) Senti- nel-2A satellite imagery ( Supplementary Table 1). Ritchie’s archipelago is a group of 13 islands, east of the main group of Andaman islands, consisting of two