Updating urban extents with nighttime light imagery by using an object-based thresholding method

Abstract

Abstract To address the problem of assessing large-scale urban dynamics from incompatible time series DMSP/OLS night-time light (NTL) data, this study proposed an Object-based Urban Thresholding method for NTL image data (i.e., NTL-OUT method) to estimate the optimal thresholds of urban objects in different NTL images. The optimal threshold for an urban object was determined by comparing the reference threshold with that of a target year through object-based normalization. Threshold normalization assumed the stability or limited change of NTL intensity over time for pseudo invariant pixels. An experiment was conducted in China to detect its urban areas in 2005 and 2010 from DMSP/OLS F152005 and F182010 NTL images by using the estimated object-based thresholds predicted from DMSP/OLS F152000 NTL data and reference urban data. Results showed that the proposed method can successfully relate thresholds among years. The estimated thresholds were highly correlated with the reference, with R 2 of 0.96 and 0.92 and RMSE of 2.9 and 4.1 pixels for 2005 and 2010, respectively. At the object level, R 2 and RMSE of the estimated vs. the reference urban areas yielded 0.94 and 12.5 pixels and 0.88 and 21.2 pixels for 2005 and 2010, respectively. At the provincial level, R 2 reached 0.94 and 0.91 with RMSE of 230.7 and 336.7 pixels for 2005 and 2010, respectively. The assessment of derived urban maps for 141 sample cities yielded an overall accuracy of 95.19% and 94.09% and Kappa of 0.56 and 0.53 in 2005 and 2010, respectively. The proposed method shows great potentials to update urban extents in a timely and cost-effective manner and to detect global urban dynamics.