Using time series PALSAR gamma nought mosaics for automatic detection of tropical deforestation: A test study in Riau, Indonesia

- HV γ0 systematically decreases after deforestation.
- Thresholding of HV γ0 decreases provided 72-96% accuracy.
- Time variation in γ0 related to precipitation caused lower accuracy.
- Yearly integration of time-series maps can improve accuracy to 82-96%.
- HH γ0 did not systematically change after deforestation.

This study intended to demonstrate the effectiveness of phased array type L-band synthetic aperture radar (PALSAR) time series mosaics for automatic detection of tropical deforestation in Riau province, Indonesia. Using six time series PALSAR mosaics, the characteristics of HH and HV gamma nought (γ0) in natural forests and deforested areas from 2007 to 2010 and the accuracy of deforestation detection by using a threshold were investigated. We obtained the following results: (1) Applying a simple thresholding method to time series differences of γ0HV was effective for fully automatic detection of deforestation areas. When we used a fixed threshold, the accuracy ranged from 72% to 96% (average of 87%). (2) γ0HH did not always show systematic changes after deforestation. (3) The temporal variation of γ0 for deforested areas was larger than that for natural forests. These variations in γ0 were correlated with 10-day accumulated precipitation. High accumulated precipitation decreased the γ0 difference between deforested areas and natural forests, causing decreased accuracy of deforestation detection. (4) Integration of the results from two different dates in a given year can reduce the detection error due to time variations and provide highly accurate results (average accuracy of 91%, minimum of 82%). The accuracy values quoted except hilly areas, shadow, and lay-over. The proposed method is effective for detection of deforestation larger than 1 ha. The deforestation mapping method proposed in the study can be utilized for assessment of yearly changes in forested areas and is useful for tracking changes in forest cover on large scales.