Mapping mangrove alliances using historical data in Fiji

The mapping of mangrove alliance distributions is limited because of lack of training data and inaccessibility of sites. Mapping mangrove alliances is important for monitoring carbon storage as well as socio-economic services to local communities. This research uses alliance field data from 1978 along with current mangrove distribution for Rewa River Delta and Suva-Navua coast in Fiji to train a Random Forest model. Classification of mangrove alliances used Sentinel-1 and − 2 images along with elevation data which resulted in a 94% classification accuracy for Rewa River Delta and 74.5% Suva-Navua. Alliances within the Rewa River Delta, including Mixed, Dogo, Boreti, Landward and Tiri, were classified with greater than 85% accuracy. In comparison, most alliances in the Suva-Navua had less than 67% accuracy; the exception was Coastal Fringing alliance which represents 56% of the area and had a 92% classification accuracy. White and red mangroves were better classified when they had larger area coverage. The Random Forest model identified SWIR, NIR and elevation data as the most important variables for discriminating mangrove alliances. Compared to limited other studies that mapped mangrove alliances using optical data alone, this analysis resulted in equal or better classification results. These results show the potential of using historic data for mapping contemporary mangrove alliances in regions that often have limited validation data or accessibility. The next step is mapping recovery of mangrove alliances post-cyclone events to help conservation groups identify focused mitigation efforts.