Incorporation of historical disturbance identified with landtrendr algorithm for land cover mapping In Malaysia

In Malaysia, land area under oil palm plantation has been increasing. Meanwhile voluntary measures to improve sustainability of palm oil production have been introduced including regulation of land conversion to oil palm plantations. The objective of this project is to assess the utility of Google Earth Engine with the LandTrendr algorithm for classifying land cover, as a first step towards developing a tool for land cover change detection in Peninsula Malaysia to support Roundtable on Sustainable Oil Palm (RSPO) certification. Ground validation data on land cover and disturbance events from satellite imagery were used to calibrate LandTrendr to detect and map change from forest to oilpalm, other vegetation or urban; other vegetation to oilpalm; and oilpalm to oilpalm (replanting). The resulting disturbance rasters were used with a 2019 multispectral Landsat mosaic in a Random Forests supervised classification. The classified maps of 2019 land cover showed an improvement in accuracy with the addition of LandTrendr rasters over using only Landsat imagery. Our results suggest that disturbance history provides useful ancillary information to support remote sensing mapping and LandTrendr could potentially become a useful tool for detecting land cover change in the tropics. The addition of LandTrendr rasters resulted in a 0.453 percentage point increase in overall accuracy from 59.992% to 60.445%. Overall accuracy improved for the target land covers - oil palm and rubber, as well as forest and urban land covers, while decreasing for other land cover classes. Highest accuracy was obtained for forest, oil palm and rice. The main source of error was from other land covers being incorrectly classified as oil palm. Confusion between the ‘other vegetation’ class and the ‘other agriculture’ class, and between urban areas and bare ground were also major sources of error