This research was aimed at developing the mapping model of benthic habitat mapping using machine-learning classification algorithms and tested the applicability of the model in different areas. We integrated in situ benthic habitat data and image processing of WorldView-2 (WV2) image to parameterise the machine-learning algorithm, namely: Random Forest (RF), Classification Tree Analysis (CTA), and Support Vector Machine (SVM). The classification inputs are sunglint-free bands, water column corrected bands, Principle Component (PC) bands, bathymetry, and the slope of underwater topography. Kemujan Island was used in developing the model, while Karimunjawa, Menjangan Besar, and Menjangan Kecil Islands served as test areas. The results obtained indicated that RF was more accurate than any other classification algorithm based on the statistics and benthic habitats spatial distribution. The maximum accuracy of RF was 94.17% (4 classes) and 88.54% (14 classes). The accuracies from RF, CTA, and SVM were consistent across different input bands for each classification scheme. The application of RF model in the classification of benthic habitat in other areas revealed that it is recommended to make use of the more general classification scheme in order to avoid several issues regarding benthic habitat variations. The result also established the possibility of mapping a benthic habitat without the use of training areas.

There are not many discussion or previous works that specifically address the issue of integrating small field plot size (1 m²) and image at different spatial resolutions in the seagrass percent cover (PC) mapping using remote sensing. This is important to determine the spatial resolution of image that can still be effectively integrated with 1 × 1 m plot size field data. This research aimed at assessing the accuracy and spatial distribution of seagrass PC map modelled from image at different spatial resolutions, using seagrass field data measure at 1 m² plot size. Two multispectral satellite images namely WorldView-2 (2 m) and PlanetScope (3 m) were used for this research and simulated to 5 m, 10 m, 15 m, and 30 m. Kemujan and Lombok Island were selected as the study area, and seagrass beds in each island have different characteristics. Machine learning random forest regression was used to perform empirical modelling and the mapping accuracy was assessed using independent seagrass PC samples. The results indicated that 1 m² plot size is still effective to be integrated with image up to 30 m spatial resolution, where the RMSE and overall seagrass PC pattern is relatively similar but the level of information precision is reduced at lower spatial resolution. Furthermore, we found out that the main factor that strongly determines the success use of 1 m² plot size and the mapping accuracy is the configuration of the seagrass bed in the study area. Seagrass PC in the more continuous seagrass bed can be mapped with higher accuracy than in patchy seagrass bed.