Mapping Benthic Habitat Composition and Estimating Seagrass Percent Cover Using Stepwise and Machine Learning Regression Methods: A Case Study from Kwandang Bay, North Gorontalo, Indonesia Setiawan Djody Harahap (a*), Huwaida Nur Salsabila (b), Abhista Fawwaz Sahitya (c), Jennifer Wijaya (c), Safina Rajwaa Ananda (d), Pramaditya Wicaksono (e), Nurul Khakhim (e), Muhammad Kamal (e), Prima Widayani (e), Muhammad Banda Selamat (f)
a) Master in Remote Sensing, Faculty of Geography, Universitas Gadjah Mada, Sekip Utara, Sleman, Daerah Istimewa Yogyakarta 55281, Indonesia
*setiawandjody99[at]mail.ugm.ac.id
b) Department of Geography, Faculty of Mathematics and Natural Sciences, Universitas Negeri Makassar, Makassar, South Sulawesi 90224, Indonesia
c) Blue Carbon Research Group, Faculty of Geography, Universitas Gadjah Mada, Sekip Utara, Sleman, Daerah Istimewa Yogyakarta 55281, Indonesia
d) Cartography and Remote Sensing, Faculty of Geography, Universitas Gadjah Mada, Sekip Utara, Sleman, Daerah Istimewa Yogyakarta 55281, Indonesia
e) Faculty of Geography, Universitas Gadjah Mada, Sekip Utara, Sleman, Daerah Istimewa Yogyakarta 55281, Indonesia
f) Faculty of Marine Science and Fisheries, Hasanuddin University, Perintis Kemerdekaan km.10 Tamalanrea, Makassar, South Sulawesi 90245, Indonesia
Abstract
Understanding the spatial distribution of benthic habitats and seagrass ecosystems is essential for effective coastal ecosystem management. Furthermore, spatial information on seagrass percent cover (PCv) is an important parameter for monitoring the condition and health of seagrass meadows, as it reflects indicators of seagrass abundance that serve as measurable proxies for evaluating ecosystem resilience. This study presents a comprehensive approach to mapping benthic habitat composition and estimating seagrass PCv in Kwandang Bay, North Gorontalo, Indonesia. Using Sentinel-2 imagery, this study employed Random Forest classification to generate the benthic habitat composition map, while the seagrass PCv map was estimated using three regression techniques: Stepwise Regression (SWR), Random Forest Regression (RFR), and Support Vector Machine Regression (SVR). The field data used in this study were collected using a georeferenced photo-transect method. The performance of the benthic composition map was assessed using a confusion matrix, while the performance of each seagrass PCv model was evaluated using standard accuracy metrics, including the coefficient of determination (\(R^{2}\)), root mean squared error (RMSE), and 1:1 graphical plot to assess prediction reliability. This study also aims to compare the performance of classical and machine learning models in mapping seagrass PCv. This study underscores the potential of both traditional and modern analytical techniques for mapping seagrass PCv, where the resulting benthic composition map and seagrass PCv estimations provide a valuable spatial baseline for monitoring coastal ecosystems and seagrass dynamics, thereby supporting conservation strategies in Kwandang Bay
Keywords: Benthic habitat mapping, Seagrass percent cover mapping, Random Forest, Stepwise regression, Support Vector Machine regression, Sentinel-2
