Fish Habitat Mapping Using Local Knowledge, UAV, and PlanetScope Imagery: A Case Study on Humpback Mahseer (Tor remadevii) in the Cauvery River

By Francesca Dragut

Abstract
The critically endangered Humpback Mahseer (Tor remadevii) fish species, endemic to the Cauvery River in South India, requires habitat maps to support species reintroduction, an effort coordinated by the Wildlife Association of South India (WASI). To localize reintroduction areas, an automated and scalable method is needed to derive habitat maps. WASI acquired a UAV image of a 3km river stretch, on which 8 habitats were identified: pool, shallow, rapid, pebble beach, rock, sand, riparian vegetation, and scrub jungle. Although UAV images are effective in classifying river habitat, they are costly and inefficient to acquire in multiple areas. This study explored the use of PlanetScope (PS) imagery and textures (homogeneity, entropy, variance) for habitat classification using local knowledge to derive training samples from UAV. Three Random Forest models were trained on one river stretch and tested on two others, using: (1) all PS bands and textures, (2) only textures on all PS bands, (3) red and near-infrared (NIR) bands and textures. Results show that an iterative process including local knowledge is efficient in deriving training samples from UAV imagery. On average, the model trained only on textures performed best in all areas: 0.56 recall and precision, and 0.89 accuracy. Although the model trained on textures and bands performed better for some habitats of the test areas (e.g. pebble beach, rock), it performed poorly in others (e.g. rapids, with some metrics reaching 0).This study shows that combining UAV, local knowledge, and high-resolution satellite imagery is effective in habitat mapping. While previous studies used textures in combination with other predictors, this study showed that using textures alone is a feasible and more generalizable alternative. This approach will enable WASI experts to up-scale habitat mapping across the Cauvery River, supporting data-driven conservation and species reintroduction efforts.