Fig. 3. Analysis of optimal buffer size for cloud/snow removal. (a) Calibration accuracies for the cloud/snow buffer with different moving window sizes. (b) The cloud/snow mask with 5x5 pixel buffer for a NASA Lunar-BRDF-corrected Black Marble product output (VNP46A2) from tile h10v04 on day-of-year 45 in 2015.
3.2. Viewing zenith angle stratification
The DNB observations were grouped into four VZA intervals, 0-20°, 20°-40°, 40°-60°, and 0-60°, to mitigate the angular effect. Some pixels have uniform DNB values among different VZA ranges, while others have uneven DNB magnitudes and large variations across the VZA (e.g., Fig 4). This VZA-related DNB radiance disparity is mainly caused by the complex and case-by-case joint angular effects of their viewing angle and local geometry (Li et al., 2019a; Tan et al., 2022; Wang et al., 2021). For low building and open area pixels with streetlights fully shielded, such as rural settlements without adjacent occlusions (Fig. 4a), almost no differences in the DNB value were observed among different VZA. While significant VZA interval related disparities were observed in areas with multiple story buildings and rural settlements surrounded by dense tree canopy (Fig. 4b-c). In downtown areas with skyscrapers, near-nadir observations have the largest overall radiance and variations (Fig. 4b). In the dense residential areas, however, the off-nadir DNB radiances are dramatically larger than the ones with lower VZA values (Fig 4c). This inconsistency of the DNB radiances consequently led to inequalities in the sensitivity of monitoring NTL changes, therefore disparities in the occurrences, magnitude, and timing of the changes across the VZA. Fig. 5 shows a sample with disparate magnitudes and occurrences of DNB radiance change among the different VZA ranges. The near-nadir time series has a larger magnitude of change in early 2016 than the ones with 20°-40° VZA, while this 2016 change is hardly noticeable from the off-nadir time series with 40°-60° VZA (Fig 5). Thus, to enhance the temporal consistency for angular affected NTL emission areas, we divided the VZA range into three equal ranges, 0-20°, 20°-40°, and 40°-60°, following the near-nadir (0-20°) and off-nadir (40°-60°) divisions of the NASA’s Black Marble product. A fourth VZA interval of 0-60° is also included to balance the trade-off between DNB time series temporal consistency and temporal density, as time series observations stratified by VZA will substantially reduce the data temporal density and therefore reduce their capability in monitoring NTL changes that can be only observed for a very short time (e.g., loss of electric grid power supply or large-scale gathering events ).