Temporal Influences on Thresholding for Satellite Retrieval of Water Surface Areas: A Case in the Poyang Lake, China

Ping Song,Yuanbo Liu,Jian Peng,Chun Ye,Dongbo Zhao

DOI: https://doi.org/10.1109/WICOM.2010.5601045

2010-01-01

Abstract:Satellite provides valuable data for monitoring the Earth's surface processes. Water surface can be delineated using Normalized Difference Vegetation Index (NDVI) or Normalized Difference Water Index (NDWI), by which an optimal threshold value is often determined empirically to separate water from non-water areas. The empirical determination prevents water delineation from automatic processing of multi-temporal images. It is yet a puzzle that what and how the optimal value is dependent on. To explore this issue, we used the multi-temporal Moderate Resolution Imaging Spectroradiometer (MODIS) data in the Poyang Lake of China for year 2008. Through thresholding, we extracted water surface areas using both products with and without correction for temporal influences. For evaluation, we selected aerosol optical thickness (AOT), solar zenith angle and sensor zenith angle as proxy indicators of the temporal factors. We further analyzed the relationships between the optimal thresholds and the three indicators. Our results showed that for the NDVI method, both AOT (R2=0.26) and sensor zenith angle (R2=0.52) had significant correlations with the optimal values determined from the data without radiometric correction. After radiometric correction, the correlations weakened with R2=027 and 0.49 for AOT and sensor zenith angle, respectively. Alternatively, for the NDWI method, AOT had the highest correlations (R2=0.18) with the optimal threshold values determined from the data without radiometric correction. It indicated that the NDWI method was less sensitive to temporal influences. With radiometric correction, the correlation became weak (R2=0.03). The weakened correlations were evidence that temporal influences on the thresholding for retrieval of water surfaces from multi-temporal satellite data. Overall, the NDWI method was a better choice to derive water surface areas using multi-temporal data than the NDVI method did- - . To get reliable retrievals of water surface, it should be careful to apply the retrieval methods using satellite data without radiometric correction.