Web-based mangrove distribution and carbon stock monitoring system in Papua using CNN on satellite imagery.

Despite being one of the most critical ecosystems in regard to biodiversity as well as being an effective storage of carbon, mangrove forests are disappearing globally through deforestation. In this paper, we propose a web-based system to monitor mangrove spatial distribution and calculate the carbon storage valuation using Convolutional Neural Network classification model with of U-Net architecture. This research focuses on the development of localized mangrove classification algorithm that is applicable in Papua, Indonesia using cloud-free Sentinel-2 satellite imagery from the year of 2016. 4 mangrove classes of Delta, Estuary, Lagoon, and Open Coast mangroves along with other classes such as Water, Non Mangroves Forest, Open Ground, and Urban are represented with balanced data points and augmented with Random Forest classifier on the entire Region of Interest as an input to the model. The model were performed sufficiently in distinguishing mangroves and non-mangroves land covers with the best F1-score achieving of 0.96 and a categorical accuracy of 0.96 on the 10th epochs. [ABSTRACT FROM AUTHOR]