The Capability of Machine Learning Combined with PlanetScope for Detecting Seaweed Using the OBIA Method
Arpin Hardiana (a*), Nurjannah Nurdin (b, c)

a) Doctor Program in Environmental Science, Faculty of graduate school, Hasanuddin University, Jl. Perintis Kemerdekaan Km. 10, Makassar 90245, South Sulawesi, Indonesia
b) Marine Science Department, Faculty of Marine Science and Fisheries, Hasanuddin University, Jl. Perintis Kemerdekaan Km. 10, Makassar 90245, South Sulawesi, Indonesia
c) Research Center and Development for Marine, Coastal, and Small Island, Hasanuddin University, Jl. Perintis Kemerdekaan Km. 10, Makassar 90245, South Sulawesi, Indonesia

arpinhardi[at]gmail.com

Abstract

Seaweed has high economic value and plays a crucial role in supporting the well-being of coastal communities and maintaining the balance of marine ecosystems. Accurate detection and mapping of seaweed distribution are crucial for planning sustainable cultivation. This study aims to evaluate the ability of a machine learning algorithm combined with high-resolution PlanetScope satellite imagery to map the dynamics of seaweed cultivation areas over 36 months (2022-2024) using PlanetScope satellite imagery with the Object-Based Image Analysis (OBIA) approach. The OBIA method enables more contextual spatial analysis by segmenting images into homogeneous objects, which are then classified using the Nearest Neighbour machine learning algorithm. The analysis process includes segmentation, extraction of spectral, textural, and spatial features, and object classification based on the training model. The results demonstrate that integrating PlanetScope and OBIA yields high classification accuracy in distinguishing seaweed objects from other types of vegetation and aquatic substrates. The machine learning algorithm has been proven capable of processing complex multivariate data to improve detection accuracy. The study found that peak seaweed cultivation occurs between March and September in most coastal areas. The highest cultivation area varies annually, with peaks in July 2022 (198.51 Ha), April 2023 (264.56 Ha), and March 2024 (205.62 Ha), while the final quarter of the year showed a significant decline, particularly in November 2023 (7.58 Ha) and 2024 (27.98 Ha), with an accuracy of 70.91%. These findings significantly contribute to the use of machine learning-based remote sensing technology to support efficient, accurate, and sustainable coastal resource management. Therefore, this approach has great potential for routine monitoring of seaweed cultivation areas in various coastal areas of Indonesia.

Keywords: Seaweed, Machine learning, Planetscope, OBIA

Topic: Topic B: Applications of Remote Sensing