An Improved Method for Estimating the Percentage Impervious Surface Area from MODIS and DMSP-OLS Night time Light Data

Abstract

The percentage of impervious surface area (ISA%) in a watershed has been suggested to be an important indicator for evaluating watershed health since the 1990s. Thus, an accurate and frequently updated ISA% map is needed by decision-makers and environmental managers. Recently, Pok developed an easily implemented method (hereafter referred to as the Pok17 method) for estimating ISA% from a moderate resolution imaging spectroradiometer (MODIS) time-series and the Defense Meteorological Satellite Program’s Operational Line-scan System (DMSP-OLS) nighttime light (NTL) data. However, it was found that the Pok17 method systematically overestimated ISA % values in rural areas (i.e., pixels with lower ISA% values). In this study, we improved the original Pok17 method to mitigate these overestimations. Firstly, we analyzed the cause of the overestimation in the Pok17 method and found that it was because a large uncertainty existed in the enhanced vegetation index-adjusted NTL index (EANTLI) values of rural areas, which are used as inputs for ISA% estimations in the Pok17 method. In this study, we proposed to use original NTL data instead of EANTLI values in rural areas. For urban and suburban areas, the EANTLI data were still used to correct the saturation problem and blooming effects in the original NTL data. The results showed that the improved Pok17 method outperformed the original one with a root mean square error (RMSE) of 10.3%, a system error (SE) of 4.3%, and a determination coefficient of 0.88. A remarkable improvement was found in pixels with ISA% values less than 20%, with the RMSE reduced from 9.7% to 8.1% and the SE reduced from 6.1% to 3.3%. This improvement is significant for evaluating watershe health because the ISA% value of the watershed is obtained by accumulating the ISA % value of each pixel, and pixels with lower ISA% values usually account for most of the area in the watershed.