Analysis of oil palm plantations using multi-sensor and multi-temporal remotely sensed data in Indonesia and Malaysia.

Southeast Asia, in particular Indonesia and Malaysia are the biggest producers of palm oil in the world and their tropical forest and peatlands are threatened by high deforestation rates and plantation expansion. This results in loss of biodiversity and loss of a significant amount of carbon pools. In order to reduce biodiversity losses and carbon emissions the major aim is to develop an accurate method to map the extent of oil palm plantations and generate insight about regions of low and high yields to improve plantation monitoring. The objectives of the research project are; (1) to identify suitable vegetation indices and sensors for the mapping of oil palm plantations, (2) to identify their different growth stages, (3) to estimate the potential yields and to (4) to (semi-) automate the above mentioned processes. In order to achieve these objectives a broad range of available satellite data will be used. Multiple data types of satellite imagery will be fused to obtain highest accuracies of land cover classification maps. The satellite data sets of interest include optical and radar images as well as a digital elevation model. Satellite images in the optical spectrum are commonly used to identify surface features due to their reflectance characteristics but they can also be identified by their shape and texture. Synthetic Aperture Radar (SAR) imagery will be used in addition which has the advantage that it is independent from cloud and light conditions. The main focus will lay on actual and archived Landsat and MODIS imagery and ALOS PALSAR-1 and -2, EnviSAT ASAR and Sentinel-1. In order to distinguish between oil palm plantations and other land cover classes, the random forest classifier will be used to perform the land cover classification. It is an algorithm which creates an ensemble of decision trees to assign pixels or objects to the respective class. The oil palm maps will be used as a land use maps to exclude all other land use and land cover classes from further analysis. Remotely sensed biophysical parameters such as fraction of absorbed photosynthetic active radiation (fAPAR) and leaf area index (LAI) will be assimilated in oil palm productivity models and spatially as well as temporally improve. Statistical yield values of previous years will be used to calibrate and validate the model performance. The main expected outcomes will be (1) a map of the actual extent of oil palm plantations, (2) their actual growth stages and (3) a model to estimate oil palm yield using satellite derived state variables. The results achieved at this point show the actual distribution of oil palm plantations in Malaysian Borneo including a list of most important sensors, spectral bands, SAR polarizations and vegetation indices to map oil palm plantations.