ASSESSMENT OF RAINFALL INFLUENCE ON SENTINEL-1 TIME SERIES ON AMAZONIAN TROPICAL FORESTS AIMING DEFORESTATION DETECTION IMPROVEMENT
Keywords: Sentinel-1, Time series, Change Detection, Rainfall Influence, Forests, Amazon
Abstract. This work aims to determinate the relationship between C-band SAR backscattering measurements over Amazonian tropical forests and hourly precipitation rates, and to study the feasibility of a SAR-anomaly masking method based on orbital rain measurements. To do so, a comprehensive dataset of ESA’s Sentinel-1 backscattering data and the concomitant GPM-IMERG precipitation data was collected and analysed. Backscattering anomalies were characterized in a statistically meaningful way. GAM models were then adjusted to the backscatter-rain data pairs. The computed models show a positive correlation between non-anomalous backscattering values and accumulated rain, of approximately 0,2 dB/mm·h−1 and 0,4 dB/mm·h−1 for VV and VH polarizations. Negative anomalies, which can easily mislead deforestation algorithms, have a strong negative correlation with rain rate observed at the time of the SAR acquisition. This is especially true for VV measurements. The subsequent anomaly masking procedure, based on computed accumulated and hourly rain thresholding, yielded unsatisfactory results. These poor results are probably due to the coarse resolution of the 0.1° GPM-IMERG data, which is insufficient to track anomaly-generating atmospheric events such as storm rain cells. Rain-related changes in SAR backscattering can compromise deforestation detection algorithms, and further research and sensor developing is needed to increase spatial resolution of precipitation measures, to reach an optimal backscattering anomaly screening.