To measure the reliability of each zone's electricity grid, we also plan to use the open VIIRS nighttime lights data. An unreliable grid is associated with electric power outages -- by detecting evenings that light output is significantly below baseline, we can estimate the number of times the grid is down; we then correlate these outages with electricity grid reliability using previous research (Mann et al., 2016). Unlike for the electricity consumption indicator, the monthly VIIRS data is not adequate for reliability estimation as it is averaged over too long of a time period. We must instead use the daily imagery.

Open VIIRS daily DNB was fetched through NOAA/NGDC. Unfortunately, the raw daily DNB imagery does not come with correlated geolocations, and it is not corrected for stray light, lighting, lunar illumination, or cloud cover. Therefore, we must apply DNB imagery formatting, imagery calibration, sort DNB by image captured time, remove cloud cover and other artifacts, before finally extracting image statistics. The temporal frequency of data points is also challenge; with only a limited number of about 15 night scenes per month sampled at a non-regular rate, more evaluation is needed to obtain a sufficient sampling of data to calculate an outage rate. Looking forward, electricity reliability will be one of the most challenging indicators to complete, and will not scale well until we have automated methods of dealing with the aforementioned challenges.

Conclusions

We found that we were able to see large fluctuations in radiance that seemed to indicate electricity reliability. However, the daily VIIRS data was markedly more challenging to process than the monthly version due to the lack of preprocessing. Furthermore, we found that the data is not as well organized, so it takes time to manually query data sources and select appropriate images. We also don’t have a high-quality model to link changes in radiance with reliability as this is still an active area of research. VIIRS can only provide single snapshots, so we must somehow correlate a few discrete time points with the electricity grid reliability over a long period of time. These issue should be addressed before moving forward with this indicator.