Fig. 16. The NTL change maps, the time series plot of a
selected pixel, and high-resolution images for Puerto Rico. (a) The
accumulated annual NTL change maps from 2013 to 2021 with the latest
detected change year presented. (b) The day-of-day NTL change map in
2017 over the region enlarged from the white rectangle from Fig. 16a.
(c) The time series plot of a selected pixel (red cross in Fig. 16b) and
the corresponding VZA-COLD detection results, in which the red, green,
blue, and grey colors indicate the different VZA intervals, lines
represent the estimated models, small dots are the DNB observations, and
the large magenta dots are the detected changes. (d) The high-resolution
Google Earth image in July 2017. The red rectangle represents the
location and size of the selected pixel. (e) The high-resolution Google
Earth image in October 2017. The red rectangle represents the location
and size of the selected pixel. The dark background in Fig. 16a and Fig.
16b is the Esri ArcMap Dark Gray Canvas base map.
4.2. Quantitative evaluation
Based on 1,088 stratified random samples and following the “good
practice” (Olofsson et al., 2014), the overall accuracy of the map is
99.54%\(\pm\)0.19% (95% confidence interval), with a user’s accuracy
of 68.25%\(\pm\)3.40% (95% confidence interval) and producer’s
accuracy of 66.89%\(\pm\)16.19% (95% confidence interval) for the NTL
change category.