Mangrove species mapping through phenological analysis using random forest algorithm on Google Earth Engine

Providing rapid, reliable, and accurate mapping methods is essential to support mangrove species inventory and conservation. The vegetative phenological analysis approach can be incorporated to differentiate mangrove species based on time-series image data. This study aims to (1) characterize the phenological signatures of mangrove species in the observed area, and (2) use the vegetative phenological patterns for mangrove species mapping using the random forest (RF) algorithm on the Google Earth Engine (GEE) platform. We characterized the vegetative phenological signatures of mangrove species using 82 filtered Sentinel-2 multi-temporal images from 2018 to 2020. Three vegetation indices, namely Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), and Soil-Adjusted Vegetation Index (SAVI), were compared to observe the mangrove vegetative phenophases trends using the GEE platform. The Mean Decrease of Gini (MDG) and the Mean Decrease of Accuracy (MDA) were calculated to determine months with critical vegetative phenophases of each mangrove species. There were six mangrove species classes and one intertidal class included in the classification at the Clungup Mangrove Conservation (CMC) area in Malang, Indonesia. Each mangrove species phenophases showed that the curve value increased in the rainy season and decreased in the dry season based on the vegetation index value fluctuations. We found that using RF-based classification, we can use these phenophases patterns to differentiate between one mangrove species to another. Our finding shows that the wet months in the middle to the end of the rainy season play an important role in distinguishing between mangrove species. The vegetative phenological analysis resulted in high accuracy in mapping mangrove species in the study area, with an overall accuracy of 72.01% for EVI, 81.82% for SAVI, and 82.51% for NDVI. The results demonstrated that mangrove species could be differentiated and mapped using vegetative phenological analysis based on multitemporal images.