Using Nighttime Light Data And MODIS Data For Large-scale Impervious Surface Mapping Research

Impervious surface is defined as water-proof synthetic surfaces. They are primarily associated with human activities and habitation through the construction of transportation infrastructure and buildings. Impervious surface is one of the factors to quantitative evaluation of urban development, urban ecology and the urban environment, has been used widely used in urban land use classification, assessment of the resident population and the urban environment, urban planning etc. But large-scale impervious surface as a driver factor of global urbanization, mapping it in a regional and global scale still to be a challenge.Coarse spatial resolution satellite remote sensing data can get quick, wide range of surface data, they are effective tools of mapping large-scale impervious surface. The coarse spatial resolution data sources generally contain MODIS, AVHRR, DMSP-OLS, VIIRS-DNB etc. But the coarsespatial resolution data such as MODIS, AVHRR still exist problems such as mixed pixel phenomenon which exhibit the mapping accurate. DMSP-OLS,?RS-DNB as nighttime light data which focus the city area as region of interest are very potential for large-scale impervious surface mapping, but at the same time, the nighttime light itself also exist some problems, such as the data saturation, low spatial resolution and other issues.Data from the U.S. Defense Meteorological Satellite Program's Operational Line-scan System (DMSP-OLS) as low resolution nighttime light remote sensing data which contains variables related to city information, also closely related to urban impervious surface, is a useful tool for large scale IS mapping. In this research we present a new integration variable-normalized impervious surface index (NISI) which combined with DMSP-OLS and MODIS NDVI data to overcome the traditional threshold technique which omits small proportional IS and less spatial information.This research indicate that a combination of DMSP-OLS and MODIS NDVI variables provided a better estimation performance than single DMSP-OLS or MODIS NDVI variable, and the integrated approach forsettlement mapping at the regional scale was promising. The combination variable with the linear method greatly improved the spatial patterns of IS distribution and the mapping accuracy. This paper provided a rapid and accurate approach to estimate fractional IS from DMSP-OLS and MODIS imagery with linear regression method at the regional scale accurately which save much time and human labor.DMSP-OLS are often used to map impervious surface area (ISA) distribution at regional and global scales, but its coarse spatial resolution and data saturation produce high inaccuracy in ISA estimation. Suomi National Polar-orbiting Partnership (SNPP) Visible Infrared Imaging Radiometer Suite's Day/Night Band (VIIRS-DNB) with its high spatial resolution and dynamic data range may provide new insights but has not been fully examined in mapping ISA distribution. In this paper, a new variable—Large-scale Impervious Surface Index (LISI)—is proposed to integrate VIIRS-DNB and Moderate Resolution Imaging Spectroradiometer (MODIS) normalized difference vegetation index (NDVI) data for mapping ISA distribution. A regression model was established, in which LISI was used as an independent variable and the reference ISA from Landsat images was a dependent variable. The results indicated a better estimation performance using LISI than using a single VIIRS-DNB or MODIS NDVI variable. The LISI-based approach provides accurate spatial patterns from high values in core urban areas to low values in rural areas, with an overall root mean squared error of 0.11. The LISI-based approach is recommended for fractional ISA estimation in a large area.Impervious surface (IS) is very important for urbanization and environment issues. The Defense Meteorological Satellite Program's Operational Line-scan System (DMSP-OLS), Terra Moderate Resolution Imaging Spectroradiometer's Normalized Difference Vegetation Index (MODIS NDVI) and the new variable Suomi National Polar-orbiting Partnership (NPP) Visible Infrared Imaging Radiometer Suite's Day/Night Band (VIIRS-DNB) provide novel ways to map the large-scale IS. But the accuracy of the urban IS mapping from these coarse remote sensing datasets remains low, and these variables generally lack of versatility for large scales. In this paper, we analyzed the DMSP-OLS low resolution nighttime light data, the MODIS NDVI medium resolution data and the VIIRS-DNB new nighttime light data and compared these three datasets performance in the study of large-scale IS mapping. In addition, with the individual variables we generated new combination variables, and applied these new combination variables for different study areas. Analysis the results suggests, data combination can provide more information than individual variables, especially the new variable Large-scale Impervious Surface Index (LISI) we proposed, which combined with MODIS NDVI and VIIRS-DNB variables performance more versatility and accuracy than the Vegetation Adjusted Normalized Urban Index (VANUI). LISI is more suitable for large-scale impervious surface mapping.Machine learning language as a non-parameter of algorithm both can used to classification and regression analysis, much research has been used the classification machine learning language with DMSP-OLS data and vegetation data for impervious surface mapping, and got good results. But the regression machine learning language rarely used to DMSP-OLS data or vegetation datain the large-scale impervious surface mapping research until now. Meanwhile, many researches show that nonparametric machine learning languages perform better than linear regression method. Therefore, this study used BP-ANN and SVM regression method and NISI variables to mapping large-scale fractional impervious surface. In addition, the results which based on individual variable of DMSP-OLS and MODIS NDVI were also used for comparison.