Power outage and environmental justice in Winter Storm Uri: an analytical workflow based on nighttime light remote sensing

Figures & data

Figure 1. Workflow of outage detection using NASA’s Black Marble product suites.

Figure 2. (a) National Weather Services (NWS) winter weather forecast near Texas, (b) counties that declared Winter Storm Uri as a major disaster.

Table 1. Descriptions of layers extracted from NASA’s Black Marble product.

Layer	Value Range	Product
Sensor Viewing Zenith	Ranging from −90° to 90°	VNP46A1
Sensor Viewing Azimuth	Ranging from −180° to 180°, 0° represents the north direction, negative values represent the east, and positive values represent the west	VNP46A1
UTC Time	Ranging from −12 to 12, represent time in the UTC	VNP46A1
BRDF-corrected Radiance	Ranging from 0 to 6553.4	VNP46A2
Mandatory Quality Flag (QF)	00 (high-quality & persistent), 01 (high-quality & ephemeral), 02 (poor-quality), 255 (no retrieval)	VNP46A2

Table 2. The list of tables and socio-economic variables extracted from the 2019 American Community Survey.

Variable Name	Description
Ratio of White	Ratio of White American population (one race) to total population
Ratio of African American	Ratio of African American population (one race) to total population
Ratio of American Indian and Alaska Native	Ratio of American Indian and Alaska Native population (one race) to total population
Ratio of Asian	Ratio of Asian American (one race) to total population
Ratio of Latino/Hispanic	Ratio of Latino/Hispanic population to total population
Ratio of 25 years old + and hold a degree less than college degree	Ratio of population over 25 years old and with a degree lower than a college degree to total population
Ratio of commute time less than 30 min	Ratio of population living in areas with less than 30 min commuting time
Ratio of income lower than poverty level	Ratio of population with income lower than the poverty level to total population
Median household income	Median household income in the past 12 months (in 2019 inflation-adjusted dollars)
Unemployment ratio	Ratio of unemployed population to total population in labor forces
Renter-occupied housing ratio	Ratio of renter-occupied housing units to total housing units
Ratio of constructions built after 2000	Ratio of constructions built before 2000 to total housing units
Median housing value	Median value of owner-occupied housing units in dollars
Median gross rent	Median gross rent of renter-occupied housing units paying cash rent

Figure 3. Hourly ratio of power outage from Feb 11 to Feb 24 (lines, left axis) and the number of counties where > 80% area is covered by high-quality NTL pixels (bars, right axis). The red dashed line indicates the general NTL acquisition time (close to 2:45 AM Central Standard Time) of NTL images on Feb 16, 2021.

Figure 4. (a) Acquisition time in Central Standard Time (CST) and VIIRS satellite orbit from VNP46A2 on Feb 16, 2021, (b) power outage ratios (tracking data) between 2–3 AM in Texas counties, areas that did not experience extensive power outages are highlighted in blue dashed boundaries.

Figure 5. Scatterplots of radiance by zenith angle from Jan 1, 2020, to Mar 2, 2021, in random four pixels in the study area. The red box indicates the radiance values on dates that have a similar viewing zenith angle as Feb 16, 2021.

Figure 6. Comparison between NTL radiance change and snow accumulation: (a) spatial distribution of NTL radiance change ($D$), (b) snowfall accumulation from NOAA National Gridded Snowfall Analysis (72-hour snowfall accumulation) at 12:00 AM, Feb 16, 2021.

Figure 7. (a) Bin-fit scatterplot between 72-hour snowfall and NTL difference ratio (D) of pixels in the Feb 16 NTL image, (b) density plots between 72-hour snowfall and NTL difference ratio (D) of pixels in the Feb 16 NTL image. The white horizontal line (D = 0) represents no change in NTL. The blue dashed line in (a) and white dashed line in (b) indicate the actual bin fit to the target (equation and R² listed at the bottom of each graph).

Table 3. Coefficients in NTL radiance adjustment function by snow accumulation category.

Snow accumulation (inches)	Slope ($f$)	Intercept ($g$)	Coefficients of determination (R^2)	Number of points
0–1	0.9464	1.0543	0.9671	36796
1–2	0.9707	0.7268	0.9483	50420
2–3	0.7815	1.4018	0.9681	36365
3–4	0.9780	1.2139	0.9738	43730
4–6	0.9278	1.3944	0.8896	67490
6–8	1.1255	0.7386	0.9849	20701
8–12	1.0645	0.9900	0.9867	10677

Figure 8. (a) Bin-fit scatterplot between the actual NTL radiance and simulated NTL radiance on Feb 16 (bin width = 0.25 unit), (b) density plots between the actual NTL radiance and simulated NTL radiance on Feb 16, (c) bin-fit scatterplot between the actual NTL radiance and simulated NTL radiance on Feb 16 (bin width = 0.25 unit), (d) density plots between the adjusted NTL radiance and simulated NTL radiance on Feb 16. Black lines in (a)&(c) and white lines in (b)&(d) indicate the adjustment target (f(x) = x). The blue dashed lines in (a)&(c) and white dashed lines in (b)&(d) indicate the actual bin fit to the target in all figures (equation and R² listed at the bottom of each graph).

Figure 9. Changes of R² with increasing ${\beta }_1$ (left) and ${\beta }_2$ (right) in the iterative program. Blue and red curves are calculated with adjusted and original radiance, respectively.

Table 4. Highest coefficient of determination (R²) in each NTL threshold and the optimal coefficient $\beta$ (if applicable) that generates the highest R².

	$R_{threshold1}$ (Eq. (5(5) $R_{threshold1}=mean(R_{baseline})-{\beta }_1\ast \mathrm{stdev}(R_{baseline})$(5) ))	$R_{threshold2}$ (Eq. (6(6) $R_{threshold2}={\beta }_2\ast mean(R_{baseline})$(6) ))	$R_{threshold3}$ (Eq. (7(7) $R_{threshold3}=min(R_{baseline})$(7) ))
Original NTL radiance	0.2938 (${\beta }_1$ = 0.96)	0.3035 (${\beta }_2$ = 0.71)	0.2882
Adjusted NTL radiance	0.3939 (${\beta }_1$ = 1.44)	0.4157 (${\beta }_2$ = 0.26)	0.3827

Figure 10. (a) Outage pixels from NTL images in the declared disaster counties in Texas, (b) zoom-in view of Austin (Travis County) and San Antonio (Bexar County), (c) zoom-in view of Houston (Harris County), (d) zoom-in view of Dallas (Dallas County) and Fort Worth (Tarrant County).

Figure 11. (a) Power outage ratios aggregated in counties, (b) power outage ratios aggregated in census tracts.

Table 5. The statistics (p-value and Pearson correlation coefficient) of the correlation models between socio-economic variables and the NTL power outage ratio.

Population group	NTL outage (county)			NTL outage (census tract)
	p-value	Pearson’s r	Degree of freedom	p-value	Pearson’s r	Degree of freedom
Ratio of White only	0.764	0.027	123	0.013*	−0.038	4180
Ratio of African American only	0.184	−0.120	123	0.194	0.020	4180
Ratio of American Indian and Alaska Native alone	0.097	−0.149	123	0.135	−0.023	4180
Ratio of Asian	0.097	0.149	123	0.139	0.023	4180
Ratio of Latino/Hispanic	0.000***	0.334	123	0.000***	0.083	4180
Ratio of 25 years old + and hold a degree less than college degree	0.356	−0.083	123	0.546	0.009	4180
Ratio of Commute time less than 30 min	0.109	−0.144	123	0.000***	−0.171	4180
Ratio of income lower than poverty level	0.484	0.063	123	0.360	−0.014	4180
Median household income	0.331	0.088	123	0.266	0.017	4180
Unemployment ratio	0.281	−0.097	123	0.000***	−0.069	4180
Renter-occupied housing ratio	0.889	0.013	123	0.120	−0.024	4180
Ratio of constructions built after 2000	0.022*	0.205	123	0.559	−0.009	4180
Median value	0.266	0.100	123	0.018*	0.037	4180
Median gross rent	0.299	0.094	123	0.803	0.004	4180

***p < 0.001.

**p < 0.01.

*p < 0.05.

Data availability statement

The data is freely available upon contacting the corresponding author.