Human Mobility as a Response to Inequality in Community Disaster Impact During Snowstorm Uri

doi:10.21203/rs.3.rs-2314179/v1

Natural disasters may cause extensive damage to local communities. In 2021, the historically low-temperature snowstorm Uri hurt Texas by disrupting business and activities, constraining energy distribution and consumption, and preventing residents from accessing critical resources. To mitigate the adverse impacts of disasters and improve the preparedness of vulnerable communities, this study incorporates SafeGraph data to investigate mobility challenges during the snowstorm by aggregating foot traffic to measure mobility change and examining the varied impacts of Uri on people of different socioeconomic statuses. The results suggest: (1) when the snowstorm occurred, everyone suffered the same level of mobility constraint; (2) human mobility was constrained with extremely cold weather, and gradually recovered when the temperature raised back; (3) households of lower socioeconomic status have more loss of mobility; at the same time, they have a higher mobility recovery rate; (4) elderly people were less resilient to the snowstorm in mobility recovery; (5) road users altered from highways to arterial routes after the occurrence of Uri. The findings serve to enhance or restore critical resources to foster greater adaptability in all aspects of community resilience, provide evidence for offering additional care to vulnerable groups, and build well-prepared emergency management programs.

Disaster Impact

Resilience

Mobility Change

Mobility Recovery

Equity

Texas experienced an unprecedented terrible snowstorm Uri and extremely cold weather in mid-February 2021, resulting in no power and constrained mobility for multiple days. Millions of Texans were constrained from access to food, water, and heat. Reinforced with the coronavirus pandemic, life was extremely difficult for Texans during that week.

Due to climate change, extreme weather has become more frequent and unpredictable in recent years. Vulnerable groups are unlikely to be prepared for hazards, and the effects of climate change are making matters worse among socially disadvantaged people. Vulnerable groups live on the margins of society, in unstable structures, and in areas most susceptible to disasters. Despite that emergency shelters provide secure and heated refuge during extreme conditions, vulnerable groups may not be able to access those resilience hubs due to limited mobility services during disaster events. Therefore, ensuring mobility services to vulnerable groups and helping them access critical resources is of utmost importance for emergency management.

Snowstorm Uri that slammed Texas extreme cold weather demonstrated the need to enhance resilience. Achieving resilience depends on a better understanding of growing vulnerabilities and identifying solutions to respond to them. As resilience comes into play at multiple scales, all disciplines jointly play roles in mitigating the harm created by such extreme weather events. To enhance resilience, local agencies need to understand and manage extreme event risk, and protect vulnerable groups from harm (Bergstrand et al., 2015). To narrow down the research interests, this study examines community disaster impact using mobility as a metric, and asks three research questions.

During the winter storm, what factors are associated with mobility?
During the winter storm, which neighborhoods are more impacted by the snowstorm and constrained from mobility? What factors are correlated with the sudden change in mobility?
After the disaster, which neighborhoods recovered faster in mobility? What factors are correlated with their mobility recovery rate?

Answering these three questions can help decision-makers better plan for man-made and natural disasters and prepare local communities to be more resilient for future disasters. This study contributes to the field from the following aspects. First, this study measures the community using transient mobility change and recovery, which is interdisciplinary research intersecting transportation engineering and environmental planning. Second, this study addresses equity issues, which is the core of community vulnerability and resiliency.

2.1 Disaster impact: A perspective from human mobility

Resilience and vulnerability reflect two connected but different approaches to understanding the response to disasters. Within the field of disaster management, many organizations and institutions have defined the concept of resilience. For example, National Research Council (2012) defined that “resilience is the ability to prepare and plan for, absorb, recover from, and more successfully adapt to adverse events.” Similarly, the United Nations International Strategy for Disaster Reduction (UNISDR, 2011) has defined resilience as “the ability of a system, community or society exposed to hazards to resist, absorb, accommodate to, and recover from the effects of a hazard in a timely and efficient manner.” Moreover, when applying the concept of resilience, the adaptative capacity and recovery capability have been used by global government organizations and research institutions the International Panel on Climate Change (Field et al., 2012), the Economic and Social Commission for Asia and the Pacific (United Nations, 2013) and Asian Development Bank (ADB, 2013), and the Department for International Development (UK) (DFID, 2011). Moreover, a growing number of scholars also highlighted the capacity to sustain the current state and recover from disasters (e.g., Cutter et al., 2008; Longstaff, 2005; Pasteur, 2011; Pelling, 2003; Twigg, 2007). Comparatively, vulnerability is more focused on the susceptibility to be disrupted by disasters, while resilience largely relies on a community’s adaptative and recovery capacity to withstand disasters (Manyena, 2006). Therefore, in brief, vulnerability is grounded in the existing conditions that are susceptible to disasters, while resilience reflects the community’s adaption capacity to withstand external stressors, such as disasters. While the focuses are slightly divergent, Cutter et al. (2008) suggested that vulnerability and resilience are not mutually exclusive and because they have overlapping. For example, a family has a house within a high-risk flood zone that makes increases this vulnerability, while they can enhance resilience by purchasing flood insurance to financially recover from the disaster (Yuan et al., 2021a).

Assessment methods of vulnerability include qualitative approaches (Anderson and Woodrow, 2019) or place-based assessment (Cutter et al., 2003). The Social Vulnerability Index (SoVI) is an additive model reflecting the community’s vulnerability with a set of preselected (Cutter et al., 2003). While the popularity of SoVI, it also receives substantial critiques from the perspectives of validation, scale, disaster context, and so on (Preston et al., 2011; Gallina et al., 2016). For the measurement of resilience, no one-fits-all method exists. Conceptual frameworks of community resilience are abundant and involve a variety of approaches that have been established to operationalize the resilience of communities, regions, and countries. Specifically, the resilience measure can be defined as a set of engineering functionality (Bruneau et al., 2003), community capitals (Miles et al., 2011), community capacity index (Foster, 2012), place-based (Cutter et al., 2010) or pattern-based measurements (Lam et al., 2016). Despite these efforts, the debate on characteristics of the empirical measurement of community resilience is continuing (Cutter et al., 2014; Schipper, 2015). However, these measurements are developed based on the characteristics during non-crisis times, without real-world changes during disasters.

To understand impact in real-world disasters, several studies have applied satellite imagery and aerial images from drones for rapid disaster impact assessment (Akshya and Priyadarsini 2019; Popescu et al. 2015; Qiang et al. 2020; Skakun et al. 2014). Zhai and Peng (2020) applied the Google Street View images to estimate the damage perceived by human eyes. However, these imagery data have some primary limitations, such as their relatively coarse spatial and temporal resolution and higher computation costs. To this end, researchers have examined the effectiveness of community-scale big data in measuring disaster impact (Yabe et al. 2020; Lu et al. 2016). The emerging social sensing technologies and “data for good” programs of some technology companies have increased the availability of community-scale big data (Neelam and Sood 2020), with a fine spatiotemporal resolution of human activities, such as daily activity indexes, transaction activities, and mobility change (Yuan et al., 2021a). Hence, variations in human movability in a community can signal disaster impacts (Yabe et al. 2020; Farahmand et al. 2021). For example, Podesta et al. (2021) explored the changes in visits to points of interest (POIs) during the Hurricane Harvey period and found that such short-term changes can reflect flood impacts. Yuan et al. (2021b) evaluated the impacts of Hurricane Harvey using the changes in credit card transactions. Kryvasheyeu et al. (2016) found that disaster-related tweets can be used to estimate damage in Hurricane Sandy. Using continuous changes of mobility during Hurricane Harvey, Hong et al. (2021) considered community resilience capacity as a function of the magnitude of impact and time-to-recovery and evaluated the community resilience using mobility decrease and recovery rate. Thus, the existing studies have provided adequate evidence that changes in mobility can also provide signals about resilience. Despite these efforts in measuring resilience from the perspective of human mobility, little is known about what factors are correlated with the sudden change in mobility during and in the immediate aftermath of the disaster.

2.2 Mobility change and social inequity

Disasters will generally change commuters’ mobility temporarily due to strong precipitations, hurricanes, tornados, snowstorms, and extreme heat (Koetse and Rietveld, 2009). For example, some scholars found that disasters can cause abrupt decreases in transit ridership and the variation is the largest during weekends and evenings (Singhal et al., 2014; Arana et al., 2014). Singhal et al. (2014) found a modal shift from cars to public transportation during snowstorms. Khattak and De Palma (1997) suggested that this mode switch may be attributed to individuals’ perceptions about road safety or other concerns about disasters (Khattak and De Palma, 1997). In addition to changing an individual’s mobility, disasters can also have an influence on transit system operation by reducing average bus frequency and increasing average headway and trip duration (Hofmann & O'Mahony, 2005). In the meantime, Cools et al. (2010) found that natural disasters, such as snowstorms, can reduce trip distances and a reduction in traffic volumes. Moreover, climate disasters can impact the frequency and duration of an individual’s movement and this effect is the most noticeable during peak hours (Koetse and Rietveld, 2009; Böcker et al., 2013). It is apparent that human mobility can change during disasters. However, reviews by Koetse and Rietveld (2009) and Böcker et al. (2013) argued that the existing studies are fragmented and sometimes contested in how individuals’ sociodemographic factors influence them to change their mobility choices and plans.

The strand of mobility research has shown that, during non-crisis times, an individual’s mobility differs because of social groups and, more specifically, income levels (Pucher and Renne, 2003; Olvera et al., 2004; Wixey et al., 2005). While acknowledging a potential accumulation of social disadvantages, occupations, etc., poverty is commonly considered to have a direct impact on individuals’ everyday mobility (Jouffe, 2019). For example, low-income people have shorter and less frequent trips because they have less access to automobiles (Dargay and Hanly, 2007; Giuliano and Dargay, 2006; Pucher and Renne, 2003). For instance, in the United States, car ownership is significantly associated with income, regarding that 27% of poor families do not have a car while only 2% of rich families do not (Jouffe, 2019), not to mention that the vehicle type, maintenance frequency, and age of the car is also related to income levels (Bhat et al., 2009). These inequalities in daily mobility are highly likely to occur in the poorest communities and minority communities (Priya and Uteng, 2009). During disasters, such inequity in mobility could exacerbate the disaster impacts, indirectly reflecting community vulnerability and resilience (McMahon, 2007). However, in the context of disasters, the association between social inequity and mobility change is yet to be specifically examined. Even though many studies have revealed that poor individuals are more vulnerable to disasters due to infrastructure and warning systems (Sastry et al., 2013; Sawada and Shimizutani, 2007; Thomas et al., 2010), unsafe houses (Cutter et al., 2006), fewer stocks of liquid assets (Wainwright and Newman, 2011), lower perception of the disaster (Lachlan et al., 2009), lack of relief services (Zakour and Harrell, 2004; Elliott and Pais, 2006; Fussell et al., 2010) and so on, the stream of disaster research, usually ignores the inequity of mobility.

3.1 Study Area

To understand the disaster impact on community using human mobility, this study takes the snowstorm Uri for empirical analysis. Snowstorm Uri greatly hurt the US for a week from February 15th, 2021. Since February 13th, 2021, various winter weather alerts were sent out across the nation, Texas seemed not well prepared for this disaster. Despite the snowstorm Uri resulting in nationwide damage, more than 90% of death were from Texas. As Houston is the largest city in Texas, this study chooses Harris County as the study area.

3.2 Data

This study employs data from various sources, aggregated at the census block groups. The first source is SafeGraph, which provides accurate points of interest (POI) across Harris County. This study employs one-month SafeGraph’s mobility data to measure individuals’ mobility inside the analytical units throughout February 2021. As individuals use smartphones, their locations are tracked by various location-based apps and synthesized by Google Maps. The second source is a collection of sociodemographic factors that derive from the American Community Survey 2014–2018. The third source is Harris County Open Data. All built environment features, including land use and road network, as well as annual average daily traffic counts, are public data. Census block groups with no individual’s mobility data are excluded from the analysis. The fourth source of data is the weather, spanning rainfall and temperature, which are obtained from Harris County’s public source and National Weather Service’s database.

3.3 Research Design & Model Specification

This study sets three steps to implement the analyses, which correspond to three independent models. The first step is to identify factors that are correlated with mobility. Mobility here is aggregated movement counts at different time occasions throughout February 2021.

The second step narrows the research to disaster impact on community during the snowstorm, which is measured by mobility change of each community due to the sudden event of snowstorm Uri. The event day was February 15th, 2021, Monday. Human mobility was greatly changed in that week; the average movement counts on other Mondays during that month was aggregated as the reference. Then disaster impact is measured by 1 minus the changed percentage of mobility, which equals the movement counts on February 15th divided by the average movement counts on other Mondays during that month. Eq. (1) presents how mobility change is obtained.

$$\begin{array}{c}M=1-\frac{{V}_{date}}{{V}_{Monday}}\#\left(1\right)\end{array}$$

where $M$ is the percentage of mobility change,${V}_{date}$ is the foot traffic counts on a specific date, i.e., ${V}_{02-15}$ is the counts on February 15th, 2021, and ${V}_{Monday}$ is the average counts on the other Mondays during February 2021.

The third step is to explore the mobility recovery after the snowstorm, which is presented by movement counts after the snowstorm divided by the average counts on the other corresponding days of normal conditions. As shown in Fig. 1, there are at least three days experienced quick recovery after snowstorm Uri, noted as February 16th, 17th, and 18th, 2021. Eq. (2) presents the method to calculate the recovery rate.

$$\begin{array}{c}{R}_{date}=\frac{{V}_{date}}{{V}_{days}}\#\left(2\right)\end{array}$$

where ${R}_{date}$ refers to the percentage of the recovery of movement counts after the snowstorm, specifically for February 16th, 17th, and 18th, 2021, ${V}_{date}$ is the counts on corresponding dates, and ${V}_{days}$ denotes the average counts on other days during that month, corresponding to ${V}_{Tuesday}, {V}_{Wednesday}$, and ${V}_{Thursday}$ in this study.

3.4 Methodology

The first dependent variable, movement counts, is aggregated at different time occasions at the level of the census block group. Therefore, a longitudinal model is applied to analyze the data. The mobility decreases and mobility recovery in the immediate aftermath of disaster are also aggregated at the level of block groups. In consequence, a set of cross-sectional models are needed to analyze what factors affect mobility decreases or recovery rates. To examine the joint effects, and to capture spatial dependence and temporal autocorrelations, this study employed the generalized additive model (GAM).

For the longitudinal analysis, the explanatory variables include time-varying variables (rainfalls and temperatures), geocoordinates, land use, transportation, and sociodemographic factors. For the cross-sectional models, the dependent variables are vulnerability (mobility change) on February 15th and resiliency (recovery rate) on February 16th, 17th, and 18th. The specified explanatory variables are similar to the longitudinal model except for the exclusion of time-varying variables. Due to the three steps having different ways of model specification, the GAMs are specified as Eqs. (3) ~ (6).

where ${T}_{i}$ is the vector of foot traffic counts in the block group $i$; ${M}_{i}$ is the vector of mobility change rate in the block group $i$; ${R}_{data, i}$ is the vector of mobility recovery rate after the snowstorm Uri in the block group $i$; ${\beta }_{0}$ is the intercept for all models; $K$ is the total number of fixed effects; ${X}_{ik}$ refers to the k^th covariate, which includes factors of socio-demographic, land use, transportation, temperature and rainfalls; $L$is the total number of covariates that present nonlinear features; ${f}_{i}\left(\bullet \right)$ is a smooth function and ${X}_{il}$ refers to the l^th covariate that contributes to nonlinear effects; ${S}_{i1}$ and ${S}_{i2}$ refer to the geocoordinates of block groups $i$; ${f}_{i1}\left(\bullet \right)\times {f}_{i2}\left(\bullet \right)$ fits the geocoordinates utilizing the same smooth function; $\epsilon$ is the vector of random errors following normal distribution.

4.1 Descriptive Analysis

To evaluate the community recovery during the snowstorm Uri, this study explores the mobility change and recovery of each neighborhood. Table 1 describes the included variables, and Table 2 presents the data summary. The monthly average movement counts were 157.82, while the average foot traffic counts on the disaster day (February 15th ) were only 41.60, and then gradually increased to 113.63 on February 18th, 2021, as shown in Figs. 1, 2, and 3. The third quartile had more fluctuations during the disaster week compared with the first and second quartiles. These results are consistent with our assumptions. There was a sudden decline and then gradually recovery, and areas with more traffic reduced more during the snowstorm season. It is worth mentioning that mobility capability was not decreased, but increased in several block groups on February 15th. Such neighborhoods are located in downtown Houston or other suburban centers, marked by white areas in Fig. 2. It indicates that people may go shopping or perform critical activities on the disaster day to obtain food and other essential resources, in preparation for the snowstorm Uri.

People of different groups have different levels of access to critical resources, and equity is a major concern. Looking at the distribution of poverty based on the American Community Survey 2018, the percentage of poor households in some block groups was 100%, while in some rich areas, this percentage was 0%. The percentages of minority groups, including Asian and Black, had large variances, indicating spatial clustering of non-white racial groups. Densities also greatly vary. The densest block group had 8.64 times higher population density in comparison with the mean. Such variation was even greater for job density. For example, the standard deviation of job density was about 2.67 of the mean, and the maximum of job density was 68.71.

Among transportation-related features, the access to highways and transit great varies across different block groups. Their variances are about 2.01 and 1.45 of their means, accordingly. For land use, commercial, office, and industrial all show great variations across block groups, indicating clustering of different land use in space except for residential.

Table 1

Variable description
Variables	Description
Mobility	Smartphone-enabled movement counts on a daily basis in block groups in February, 2021.
Mobility change	The changed percentage of movement counts in the disaster day compared with foot traffic counts in normal.
Mobility recovery rate	The comparison between movement counts after the disaster with movement counts under normal conditions.
Sociodemographic
Female	The percentage of females in block groups
Age 17-	The percentage of people aged 17 or younger in block groups
Age 65+	The percentage of people aged 65 or older in block groups
Black	The percentage of Black in block groups
Asian	The percentage of Asian in block groups
Poverty	The percentage of households in poverty in block groups
Population	The number of people in block groups
Job	The number of jobs in block groups
Land Use
Residential	The percentage of residential land in block groups;
Commercial	The percentage of commercial land in block groups;
Office	The percentage of offices in block groups;
Industrial	The percentage of industrial land in block groups;
Transportation
AADT	Annual average daily traffic: the average number of cars passing through a road segment
Local	The density of local streets in block groups; length of streets/ the area of the block group
Highway	The density of highways in block groups; length of highways/ the area of the block group
Major	The density of major arterials in block groups; length of arterials/ the area of the block group
Bus Lines	The number of bus lines in block groups
Temporal Factors and Weather
Day	The numeric order of days from February 1st to 28th, 2021
Week	The day of the week (from Monday to Sunday)
Rainfall	The rainfalls in block groups for each day in February, 2021
Temperature	The average lowest temperature for all days in February, 2021 (℉)
Geocoordinates
Latitude	A geographic coordinate that specifies the north–south position of each block group center.
Longitude	A geographic coordinate that specifies the east–west position of each block group center.

Table 2

Data summary
	Mean	St. D.	Min.	Max
Objective Variables
Mobility	157.82	276.89	0.00	5,670.00
Mobility on Feb. 15th	41.60	61.15	0.00	942.00
Mobility on Feb. 16th	72.53	111.59	0.00	1,216.00
Mobility on Feb. 17th	81.83	129.07	0.00	1,577.00
Mobility on Feb. 18th	113.63	182.50	0.00	2,238.00
Mobility change	0.63	0.36	-4.31	1.00
Mobility recovery rate on Feb. 16th	0.55	0.59	0.00	16.33
Mobility recovery rate on Feb. 17th	0.61	0.68	0.00	25.29
Mobility recovery rate on Feb. 18th	0.76	0.72	0.00	26.25
Explanatory Variables
Sociodemographic
Female (%)	50.21	3.26	11.99	63.78
Age 17- (%)	26.70	7.36	0.00	48.99
Age 65+ (%)	9.33	5.96	0.00	54.85
Black (%)	19.51	23.36	0.00	97.32
Asian (%)	5.52	7.46	0.00	58.75
Poverty (%)	17.88	11.99	0.00	100.00
Population	1,919.41	1,310.90	25.00	16,578.00
Job	1,081.35	2,891.96	0.00	74,299.00
Land Use
Residential (%)	42.04	18.10	0.00	92.62
Commercial (%)	6.08	7.45	0.00	77.85
Office (%)	1.67	4.12	0.00	41.74
Industrial (%)	6.12	9.53	0.00	61.06
Commute
AADT (/1000)	6.94	3.64	0.32	22.21
Local	12.10	6.25	0.00	38.35
Highway	1.31	2.63	0.00	23.20
Major	2.93	2.35	0.00	19.76
Bus Lines	2.07	3.00	0.00	48.00
Time and Environment Factors
Day	14.50	8.08	1.00	28.00
Week	4.00	2.00	1.00	7.00
Rainfall	0.06	0.20	0.00	1.72
Temperature	42.29	14.72	13.00	70.00

4.2 Inferential Analysis

4.2.1 Mobility models (longitudinal)

Two longitudinal models are estimated for human mobility before Uri vs. in and after Uri. Both models are well fitted, with adjusted R square of 0.45 and 0.49. The modeling outcomes are largely consistent. There are several general findings. In block groups with a higher percentage of females, we observe more movements. In block groups with a greater percentage of children or elderly people, we notice a lower level of mobility. However, this finding may involve sampling bias in the data because children and elderly people generally have a lower level of smartphone usage, which relates to how the mobile device data is solicited. This finding may not be true if the data collection method is changed. In block groups with a higher percentage of black people, we observe a lower level of mobility. Densely developed block groups are correlated with a higher level of foot traffic. No matter population density or job density, they are both correlated with a higher level of human activities. More movement is observed in zones with a higher percentage of commercial land use, while less is observed in zones with a higher percentage of residential land use. The number of bus lines and the annual average daily traffic both suggest positive associations with mobility.

There are several differences between the modeling outcomes (before vs. after Uri) and deserve our attention. First, the percentage of poverty households in block groups suggests a negative association with human mobility before Uri, while this association is no longer significant for the model with data collected after Uri. This finding indicates that low-income households are more mobility constraints. That is to say, when the disaster occurs, everyone suffers the same level of mobility constraint. Disasters have justice for all Texans because they are all human beings. Second, the percentage of industrial land use suggests a positive association with mobility before Uri, while changes are insignificant after Uri. This is an indication that economic development is greatly impacted for a certain period after the disaster, and business owners may need help for recovery. Third, the density of arterial routes suggests a negative association with foot traffic, while changing to a positive association after Uri. This indicates that road users may perceive arterial routes are safer for driving after the disaster in comparison with highways. Lastly, the temperature does not suggest any significance for the model before Uri, possibly due to limited variation in temperature for the first half month before snowstorm Uri. However, the temperature shows a positive association with foot traffic after the occurrence of Uri. This is confirming evidence that human activities can be constrained after extremely cold weather, while becoming normal when the temperature is rising back. As shown in Table 3, as well as Figs. 3 and 4, there are great spatial variations in mobility across the whole county, indicating that the spatial autocorrelations should be captured using advanced models.

Table 3

Results of longitudinal models (Unit: block groups, # of Units: 10,460 for each model)
Variable	Pre Snowstorm Uri			In and Post Snowstorm Uri
Variable	Estimate	P-value	Pr(>\|t\|)	Estimate	P-value	Pr(>\|t\|)
Intercept	3.920	< 0.001	***	3.575	< 0.001	***
Female	3.095	< 0.001	***	2.894	< 0.001	***
Age 17-	-1.057	< 0.001	***	-1.112	< 0.001	***
Age 65+	-0.593	0.004	**	-0.876	< 0.001	***
Black	-0.638	< 0.001	***	-0.477	< 0.001	***
Asian	-0.121	0.442		0.139	0.328
Poverty	-0.292	0.015	*	-0.046	0.658
Population/1000	0.151	< 0.001	***	0.156	< 0.001	***
Jobs/1000	0.035	< 0.001	***	0.037	< 0.001	***
Residential	-2.058	< 0.001	***	-1.605	< 0.001	***
Commercial	3.921	< 0.001	***	4.304	< 0.001	***
Office	0.130	0.537		-0.064	0.747
Industrial	0.223	0.037	*	0.124	0.202
AADT	0.015	0.001	***	0.011	0.007	**
Local	-0.009	0.002	**	-0.009	< 0.001	***
Highway	0.003	0.437		-0.003	0.403
Major	-0.006	0.286		0.015	0.001	**
Bus Lines	0.037	< 0.001	***	0.032	< 0.001	***
Temperature	0.005	0.107		0.025	< 0.001	***
Smooth terms	edf	p-value		edf	p-value
s(Rainfall)	1.015	0.069	.	6.371	0.006	**
s(Week)	1.898	< 0.001	***	1.960	< 0.001	***
s(Latitude, Longitude)	28.108	< 0.001	***	27.812	< 0.001	***
Signif. codes: 0 ‘*’ 0.001 ‘’ 0.01 ‘*’ 0.05
	${R}_{adj}^{2}$	0.452		${R}_{adj}^{2}$	0.492
	REML	-29,195		REML	-26,257

4.2.2 Mobility change models (cross-sectional)

As Texans’ mobility became more constrained when Uri landed (February 15th, 2021), and most block groups experienced reductions on that day. As shown in Table 4, for the modeling results in mobility change, all land-use variables suggest significance. The percentages of commercial, offices, and industrial suggest positive associations with mobility change. These findings indicate that human activity concentrated spaces, no matter shopping centers, offices, and manufacturing areas, suffered different levels of impacts made by Uri. In contrast, the percentage of residential land use shows a negative correlation with mobility change, indicating most people stayed at homes when Uri arrives in Texas. The percentage of poverty households suggests a negative correlation with changed mobility. This result indicates that low-income households had limitations to move in space or receive mobility services. This result also indicates households of low socioeconomic status were at greater risk than other groups, as well as barriers to disaster preparedness and other adverse situations or experiences they may face during the phases of disaster impact, response, and recovery. The adjusted R square is only 0.07. No significant spatial difference can be identified regarding the mobility change zones.

4.2.3 Mobility recovery (cross-sectional)

After Uri occurred, especially on February 16th and 17th, 2021, the disaster hurt all Texans equally. Almost no difference is identified regarding people’s income, race, gender and age. The percentages of commercial and industrial land use show negative associations with recovery. This result indicates that the mobility was more constrained in these areas during the two days. Till February 18th, 2021, the situation was started to recover. The percentage of residential land use shows a positive association with the recovery rate, indicating people were better off at home. The percentage of poverty households shows a positive association with the recovery rate, indicating areas with a higher proportion of low socioeconomic status households quickly recovered in comparison with other areas. Being consistent with our assumptions, low socioeconomic status households were more constrained from accessing resources. However, they were more vulnerable; they were also more resilient to disasters. The percentage of elderly people shows a negative association with the recovery rate. This result indicates that elderly people were less resilient to disasters, and more attention should be assigned to help this age group. Overall, the differences across the three cross-sectional models are negligible. There were some spatial differences regarding the mobility recovery rate on February 16th, while such spatial autocorrelations disappeared for the following two days regarding the modeling outcomes.

Table 4

Results of cross-sectional models (Unit: block groups)
Variable	Vulnerability		Resiliency
	Feb. 15th (2,088)		Feb. 16th (2,088)		Feb. 17th (2,088)		Feb. 18th (2,088)
	Estimate	p-value	Estimate	p-value	Estimate	p-value	Estimate	p-value
Intercept	0.456 **	0.003	0.563 *	0.027	0.530	0.072	0.467	0.133
Female	0.357	0.252	-0.039	0.940	0.089	0.881	0.593	0.346
Age 17-	0.240	0.123	-0.132	0.622	0.003	0.991	-0.278	0.399
Age 65+	0.098	0.582	-0.004	0.990	-0.141	0.682	-0.751 *	0.039
Black	-0.057	0.182	0.067	0.359	0.066	0.433	0.082	0.367
Asian	-0.019	0.879	-0.082	0.702	0.075	0.763	0.189	0.476
Poverty	-0.303 ***	< 0.001	0.186	0.202	0.276	0.103	0.376 *	0.037
Population/1000	0.007	0.368	0.001	0.910	-0.020	0.174	-0.022	0.161
Jobs/1000	-0.001	0.832	-0.001	0.870	0.000	0.948	-0.002	0.771
Residential	-0.255 ***	< 0.001	0.170	0.114	0.198	0.114	0.373 **	0.005
Commercial	0.659 ***	< 0.001	-0.437 *	0.026	-0.522 *	0.022	-0.123	0.610
Office	0.524 *	0.018	-0.564	0.123	-0.757	0.075	-1.147 *	0.011
Industrial	0.224 *	0.031	-0.371 *	0.030	-0.480 *	0.016	-0.391	0.062
AADT	0.004	0.472	-0.008	0.408	0.008	0.498	0.007	0.559
Local	0.000	0.853	0.001	0.640	0.002	0.641	-0.001	0.801
Highway	-0.003	0.355	0.004	0.553	-0.004	0.561	0.007	0.361
Major	0.003	0.411	-0.004	0.581	0.001	0.923	-0.001	0.897
Bus lines	0.007	0.069	-0.008	0.221	-0.007	0.302	-0.007	0.328
Smooth terms	edf	p-value	edf	p-value	edf	p-value	edf	p-value
s (Latitude, Longitude)	8.270	0.578	14.650***	< 0.001	14.250	0.136	15.880	0.064
Signif. codes: 0 ‘*’ 0.001 ‘’ 0.01 ‘*’ 0.05
	${R}_{adj}^{2}$	0.0721	${R}_{adj}^{2}$	0.0488	${R}_{adj}^{2}$	0.0391	${R}_{adj}^{2}$	0.0423
	REML	-831.54	REML	-1,862.2	REML	-2,174.2	REML	-2,282.1

Our findings shed new light on community resilience research at least from the following perspectives. First, our study demonstrated the feasibility of applying mobility change and mobility recovery to measure disaster impact during Snowstorm Uri, which are empirically meaningful and can be acted upon in future studies. Our findings show that the metrics can fill the aforementioned gaps of measuring disaster impact at the high spatiotemporal resolution that was argued in the existing literature (Yabe et al. 2020; Farahmand et al. 2021), i.e., the metrics can be measured at the block group level on a daily basis using the mobility changes. For instance, by tracking the daily mobility change of individuals, we are able to not only evaluate the seriousness of the snowstorm but also easily identify the recovery process of a specific block group.

Second, beyond the measurement of disaster impact, this study contributes to the knowledge about human mobility in the context of snowstorms, which notably enables future exploratory research into these similar disaster events and the mitigation of the adverse impacts. Interestingly, our analysis shows that when the snowstorm occurs, everyone suffers the same level of mobility constraint, suggesting that higher-income neighborhoods may not be able to move anyplace. This finding is a little counterintuitive at the first glance since it challenges the conventional wisdom that high-income individuals often have higher resilience capacities in the face of a disaster (Zhai et al., 2020), in part due to their access to automobiles (Dargay and Hanly, 2007; Giuliano and Dargay, 2006; Pucher and Renne, 2003). However, the finding is particularly true in our case study considering that snowstorm often brings traffic to a halt on highways, meaning that car ownership may not an advantage under such a situation. Moreover, our results show that the mobility level of poverty households could recover to the original state more quickly, again suggesting that conventionally vulnerable populations may not always have a low resilience capacity.

Third, disaster impact assessment using our method indicates that all land-use variables are statistically significant, which can help land-use planners understand which types of land use are the least resilient, as well as how they can improve the adaptive capacity of different types of land use to respond to disasters (Ahern, 2011; Hung and Chen, 2013). It further asks for guidelines to allocate land resources and transportation infrastructures, and strengthen zoning regulations and environmentally sensitive area protection (Godschalk, 2003; Haimes, 2009). Thus, a remaining challenge for future studies is how land use policy-making can be made more operable and applicable to enhance individuals’ mobility during a disaster. Lastly, in our research, the temperature does not suggest any significance for the model before Uri; however, the temperature shows a positive association with human mobility after Uri occurred, confirming that colder weather could decrease travel volumes (Shih and Nicholls, 2011).

This study provides a block-level analysis to comprehensively describe the disaster process, to quantify and aggregate mobility change, to measure the disaster impact, and to identify the spatial and temporal variations of Uri’s impact across different communities. This study provides rich insights into the variation of how people of different socioeconomic statuses and age groups suffered from the disaster. This study draws conclusions on how to advance our emergency management programs to provide timely help for low-income neighborhoods and vulnerable groups.

Yet, this study fails to capture how individuals were impacted by Uri. Future research may interview local residents to better understand how their mobility challenges were addressed during the event. Also, this study may involve digital divide, i.e., a biased understanding of the mobility challenges of different socioeconomic groups during the disaster, because the results are analyzed based on location-enabled app-solicited data, while children and elderly people generally have a lower level of smartphone usage. The findings serve to enhance or restore critical resources to foster greater adaptability in all aspects of community resilience, provide additional care to vulnerable groups, and build well-prepared emergency management programs.

Funding: The authors have not disclosed any funding.

Ethics declarations

Conflict of interest: The authors state that they have no conflict of interest.

Ahern, J. (2011). From fail-safe to safe-to-fail: Sustainability and resilience in the new urban world. Landscape and urban Planning, 100(4), 341-343.
Akshya, J., & Priyadarsini, P. L. K. (2019, February). A hybrid machine learning approach for classifying aerial images of flood-hit areas. In 2019 International Conference on Computational Intelligence in Data Science (ICCIDS) (pp. 1-5). IEEE.
Anderson, M. B., & Woodrow, P. J. (2019). Rising from the ashes: development strategies in times of disaster. Routledge.
Arana, P., Cabezudo, S., & Peñalba, M. (2014). Influence of weather conditions on transit ridership: A statistical study using data from Smartcards. Transportation research part A: policy and practice, 59, 1-12.
Asian Development Bank. (2013). Investing in resilience: Ensuring a disaster-resistant future. Asian Development Bank.
Bergstrand, K., Mayer, B., Brumback, B., Zhang, Y. (2015) 'Assessing the relationship between social vulnerability and community resilience to hazards', Social indicators research, 122(2), pp. 391-409
Bhat, C. R., Sen, S., & Eluru, N. (2009). The impact of demographics, built environment attributes, vehicle characteristics, and gasoline prices on household vehicle holdings and use. Transportation Research Part B: Methodological, 43(1), 1-18.
Böcker, L., Dijst, M., & Prillwitz, J. (2013). Impact of everyday weather on individual daily travel behaviours in perspective: a literature review. Transport reviews, 33(1), 71-91.
Bruneau, M., Chang, S. E., Eguchi, R. T., Lee, G. C., O'Rourke, T. D., Reinhorn, A. M., ... & Von Winterfeldt, D. (2003). A framework to quantitatively assess and enhance the seismic resilience of communities. Earthquake spectra, 19(4), 733-752.
Cools, M., Moons, E., Creemers, L., & Wets, G. (2010). Changes in travel behavior in response to weather conditions: do type of weather and trip purpose matter?. Transportation Research Record, 2157(1), 22-28.
Cutter, S. L., Ash, K. D., & Emrich, C. T. (2014). The geographies of community disaster resilience. Global environmental change, 29, 65-77.
Cutter, S. L., Barnes, L., Berry, M., Burton, C., Evans, E., Tate, E., & Webb, J. (2008). A place-based model for understanding community resilience to natural disasters. Global environmental change, 18(4), 598-606.
Cutter, S. L., Boruff, B. J., & Shirley, W. L. (2003). Social vulnerability to environmental hazards. Social science quarterly, 84(2), 242-261.
Cutter, S. L., Burton, C. G., & Emrich, C. T. (2010). Disaster resilience indicators for benchmarking baseline conditions. Journal of homeland security and emergency management, 7(1).
Cutter, S. L., Emrich, C. T., Mitchell, J. T., Boruff, B. J., Gall, M., Schmidtlein, M. C., ... & Melton, G. (2006). The long road home: Race, class, and recovery from Hurricane Katrina. Environment: Science and Policy for Sustainable Development, 48(2), 8-20.
Dargay, J., & Hanly, M. (2007). Volatility of car ownership, commuting mode and time in the UK. Transportation Research Part A: Policy and Practice, 41(10), 934-948.
Department of International Development, United Kingdom (DFID). Defining Disaster Resilience: A DFID Approach Paper (2011). available at last accessed on July 15, 2021. https://www.fsnnetwork.org/sites/default/files/dfid_defining_disaster_resilience.pdf
Elliott, J. R., & Pais, J. (2006). Race, class, and Hurricane Katrina: Social differences in human responses to disaster. Social science research, 35(2), 295-321.
Field, C. B., Barros, V., Stocker, T. F., & Dahe, Q. (Eds.). (2012). Managing the risks of extreme events and disasters to advance climate change adaptation: special report of the intergovernmental panel on climate change. Cambridge University Press.
Foster, K. A. (2012). Resilience Capacity Index-Site-Resilience Capacity Index. University of California, Berkeley.
Fussell, E., Sastry, N., & VanLandingham, M. (2010). Race, socioeconomic status, and return migration to New Orleans after Hurricane Katrina. Population and environment, 31(1-3), 20-42.
Gallina, V., Torresan, S., Critto, A., Sperotto, A., Glade, T., & Marcomini, A. (2016). A review of multi-risk methodologies for natural hazards: Consequences and challenges for a climate change impact assessment. Journal of environmental management, 168, 123-132.
Giuliano, G., & Dargay, J. (2006). Car ownership, travel and land use: a comparison of the US and Great Britain. Transportation Research Part A: Policy and Practice, 40(2), 106-124.
Godschalk, D. R. (2003). Urban hazard mitigation: creating resilient cities. Natural hazards review, 4(3), 136-143.
Haimes, Y. Y. (2009). On the definition of resilience in systems. Risk Analysis: An International Journal, 29(4), 498-501.
Hofmann, M., & O'Mahony, M. (2005, September). The impact of adverse weather conditions on urban bus performance measures. In Proceedings. 2005 IEEE Intelligent Transportation Systems, 2005. (pp. 84-89). IEEE.
Hong, B., Bonczak, B. J., Gupta, A., & Kontokosta, C. E. (2021). Measuring inequality in community resilience to natural disasters using large-scale mobility data. Nature communications, 12(1), 1-9.
Hung, H. C., & Chen, L. Y. (2013). Incorporating stakeholders’ knowledge into assessing vulnerability to climatic hazards: application to the river basin management in Taiwan. Climatic change, 120(1), 491-507.
Jouffe, Y., Caubel, D., Fol, S., & Motte-Baumvol, B. (2019). Dealing with inequality in mobility: Tactics, strategies and projects for poor households on the outskirts of Paris. Cybergeo: European Journal of Geography.
Khattak, A. J., & De Palma, A. (1997). The impact of adverse weather conditions on the propensity to change travel decisions: a survey of Brussels commuters. Transportation Research Part A: Policy and Practice, 31(3), 181-203.
Koetse, M. J., & Rietveld, P. (2009). The impact of climate change and weather on transport: An overview of empirical findings. Transportation Research Part D: Transport and Environment, 14(3), 205-221.
Kryvasheyeu, Y., Chen, H., Obradovich, N., Moro, E., Van Hentenryck, P., Fowler, J., & Cebrian, M. (2016). Rapid assessment of disaster damage using social media activity. Science advances, 2(3), e1500779.
Lachlan, K. A., Burke, J., Spence, P. R., & Griffin, D. (2009). Risk perceptions, race, and Hurricane Katrina. The howard journal of communications, 20(3), 295-309.
Lindell, M. K., & Perry, R. W. (2003). Communicating environmental risk in multiethnic communities. Sage Publications.
Longstaff, P. H. (2005). Security, resilience, and communication in unpredictable environments such as terrorism, natural disasters, and complex technology. Center for Information Policy Research, Harvard University.
Lu, X., Wrathall, D. J., Sundsøy, P. R., Nadiruzzaman, M., Wetter, E., Iqbal, A., ... & Bengtsson, L. (2016). Unveiling hidden migration and mobility patterns in climate stressed regions: A longitudinal study of six million anonymous mobile phone users in Bangladesh. Global Environmental Change, 38, 1-7.
Manyena, S. B. (2006). The concept of resilience revisited. Disasters, 30(4), 434-450.
Maze, T. H., Agarwal, M., & Burchett, G. (2006). Whether weather matters to traffic demand, traffic safety, and traffic operations and flow. Transportation research record, 1948(1), 170-176.
McMahon, M. M. (2007). Disasters and poverty. Disaster Management & Response, 5(4), 95-97.
Miles, S. B., & Chang, S. E. (2011). ResilUS: A community based disaster resilience model. Cartography and Geographic Information Science, 38(1), 36-51.
N. Lam, N. S., Reams, M., Li, K., Li, C., & Mata, L. P. (2016). Measuring community resilience to coastal hazards along the Northern Gulf of Mexico. Natural hazards review, 17(1), 04015013.
National Research Council. (2012). Disaster resilience: A national imperative. The National Academies Press, Washington, D.C., USA (2012)
Neelam, S., & Sood, S. K. (2020). A scientometric review of global research on smart disaster management. IEEE Transactions on Engineering Management, 68(1), 317-329.
Olvera, L. D., Mignot, D., & Paulo, C. (2004). Daily mobility and inequality: the situation of the poor. Built Environment, 30(2), 153-160.
Pasteur, K. (2011). From Vulnerability to Resilience, a framework for analysis and action to build community resilience. Practical Action Publishing.
Pelling, M. (2003). The vulnerability of cities: natural disasters and social resilience. Earthscan.
Podesta, C., Coleman, N., Esmalian, A., Yuan, F., & Mostafavi, A. (2021). Quantifying community resilience based on fluctuations in visits to points-of-interest derived from digital trace data. Journal of the Royal Society Interface, 18(177), 20210158.
Popescu, D., Ichim, L., & Caramihale, T. (2015, October). Flood areas detection based on UAV surveillance system. In 2015 19th International Conference on System Theory, Control and Computing (ICSTCC) (pp. 753-758). IEEE.
Preston, B. L., Yuen, E. J., & Westaway, R. M. (2011). Putting vulnerability to climate change on the map: a review of approaches, benefits, and risks. Sustainability Science, 6(2), 177-202.
Priya, T., & Uteng, A. (2009). Dynamics of transport and social exclusion: Effects of expensive driver's license. Transport policy, 16(3), 130-139.
Pucher, J., & Renne, J. L. (2003). Socioeconomics of urban travel. Evidence from the 2001 NHTS.
Qiang, Y., Huang, Q., & Xu, J. (2020). Observing community resilience from space: Using nighttime lights to model economic disturbance and recovery pattern in natural disaster. Sustainable Cities and Society, 57, 102115.
Sastry, N., & Gregory, J. (2013). The effect of Hurricane Katrina on the prevalence of health impairments and disability among adults in New Orleans: differences by age, race, and sex. Social Science & Medicine, 80, 121-129.
Sawada, Y., & Shimizutani, S. (2007). Consumption insurance against natural disasters: evidence from the Great Hanshin-Awaji (Kobe) earthquake. Applied Economics Letters, 14(4), 303-306.
Schipper, E. L. F., & Langston, L. (2015). A comparative overview of resilience measurement frameworks. Analyzing Indicators and Approaches; Overseas Development Institute: London, UK, 422.
Shih, C., & Nicholls, S. (2011). Modeling the influence of weather variability on leisure traffic. Tourism Analysis, 16(3), 315-328.
Singhal, A., Kamga, C., & Yazici, A. (2014). Impact of weather on urban transit ridership. Transportation research part A: policy and practice, 69, 379-391.
Skakun, S., Kussul, N., Shelestov, A., & Kussul, O. (2014). Flood hazard and flood risk assessment using a time series of satellite images: A case study in Namibia. Risk Analysis, 34(8), 1521-1537.
Thomas, T., Christiaensen, L., Do, Q. T., & Trung, L. D. (2010). Natural disasters and household welfare: evidence from Vietnam. World Bank Policy Research Working Paper, (5491).
Twigg, J. (2007). Characteristics of a disaster-resilient community: A guidance note. Department for International Development (DFID).
UNISDR, G. (2009). Global assessment report on disaster risk reduction. United Nations International Strategy for Disaster Reduction Secretariat Geneva, Switzerland.
UNITED NATIONS. (2013). Building resilience to natural disasters and major economic crises. United Nations Economic and Social Commission for Asia and the Pacific, ESCAP.
Wainwright, F., & Newman, C. (2011). Income shocks and household risk-coping strategies: evidence from rural Vietnam. Institute for International Integration Studies Discussion paper, 358.
Wixey, S., Jones, P., Lucas, K., & Aldridge, M. (2005). Measuring accessibility as experienced by different socially disadvantaged groups. User needs literature review. EPSRC FIT Programme, Working paper, (1).
Yabe, T., Zhang, Y., & Ukkusuri, S. V. (2020). Quantifying the economic impact of disasters on businesses using human mobility data: a Bayesian causal inference approach. EPJ Data Science, 9(1), 36.
Yang, Y., Li, Q., & Mostafavi, A. (2021). Unraveling the Temporal Importance of Community-scale Human Activity Features for Rapid Assessment of Flood Impacts. arXiv preprint arXiv:2106.08370.
Yuan, F., Li, M., Liu, R., Zhai, W., & Qi, B. (2021a). Social media for enhanced understanding of disaster resilience during Hurricane Florence. International Journal of Information Management, 57, 102289.
Yuan, F., Yang, Y., Li, Q., & Mostafavi, A. (2021b). Unraveling the Temporal Importance of Community-scale Human Activity Features for Rapid Assessment of Flood Impacts. arXiv preprint arXiv:2106.08370.
Zakour, M. J., & Harrell, E. B. (2004). Access to disaster services: Social work interventions for vulnerable populations. Journal of Social Service Research, 30(2), 27-54.
Zhai, W., & Peng, Z. R. (2020). Damage assessment using google street view: Evidence from hurricane michael in mexico beach, florida. Applied geography, 123, 102252.
Zhai, W., Peng, Z. R., & Yuan, F. (2020). Examine the effects of neighborhood equity on disaster situational awareness: Harness machine learning and geotagged Twitter data. International Journal of Disaster Risk Reduction, 48, 101611.

Human Mobility as a Response to Inequality in Community Disaster Impact During Snowstorm Uri

Status:

Journal Publication

Version 1

Abstract

Figures

1. Introduction

2. Literature review