ERS-1 observations and potential for use in tropical forest monitoring

Forest/non-forest discrimination is possible using two ERS-1 SAR images acquired under different soil moisture conditions, as long as the soil surface is not too rough; rough soil appears to give similar response to primary forest under all reasonable soil moisture conditions. Even a small amount of regrowth gives similar backscatter to primary forest at the angle of incidence of ERS-1, although there is a weak trend of backscatter increasing with age in both observations and model results. Non-forest areas can be automatically detected by thresholding difference images. Pre-processing using simulated annealing or segmentation to reduce speckle greatly increases the accuracy of this forest/non-forest discrimination but both detect small areas of change in primary forest areas. These methods are fairly machine intensive but the run times are still low enough for them to be viable for operational data analysis. Averaging blocks of pixels to reduce speckle produces results comparable to the more sophisticated methods as the window size increases, but cannot be used for small area detection. An automatic monitoring system to detect changes in forest boundaries would therefore need to acquire images soon after deforestation and before significant regrowth could occur. Since images under wet and dry conditions are also required it seems likely that images need to be acquired on a monthly basis