Using Socio-Economic Indicators to predict Earthquake Recovery

Although there is profound evidence that pre-existing socio-economic conditions (such as those associated with ‘disaster resilience) greatly influence the recovery process after an earthquake, attempts to measure and validate this relationship have been limited. As a result, GEM is investigating the relationship between disaster resilience and earthquake recovery through a novel framework, according to which socio-economic indicators are used to predict the evolution of recovery of the building stock following an earthquake. The city of Napa, California and the 2014 South Napa earthquake are being used as a case study for the development and validation of the aforementioned methodology. While relatively modest in intensity (Mw 6.0), this earthquake caused significant ground shaking and damage, particularly in the city’s core. Direct costs associated with damage have been estimated at USD 362 million, with economic costs to Napa County estimated at up to USD 1 billion. The figures below illustrate examples of affected residential properties and businesses, but also critical infrastructure damage.

Damage to residential properties, business and critical infrastructure. Photo credits: Napa Valley Register (J.L.Sousa), Justin Sullivan/Getty Images, KCRA (Brian Hicky), vos Iz Neias.

In order to define the relationship between disaster resilience and earthquake recovery, both concepts can be quantitatively evaluated. The disaster resilience of the city of Napa is represented through the use of a set of proxy variables that are classified into five subcomponents: social, economic, infrastructure, community capital and institutional resilience. The variables were retrieved from publically available sources at the census block group level of geography, as defined by the U.S. Census Bureau. Conversely, the spatiotemporal evaluation of the recovery in the city was accomplished using in situ observations of building damage at six-month intervals that were conducted following an initial damage assessment performed by city officials. In the aftermath of the Napa Earthquake, building damage observations (i.e. location, building type, colour-tagging information (red or yellow) and damage description) were geocoded by city officials and made available via a web Geographic Information System (GIS). This information was retrieved to build a geospatial point-level dataset of 1462 damaged buildings. After the development of the initial damage database, two separate field surveys were conducted in the city of Napa, six months and one year following the earthquake, respectively. A third field evaluation of the city’s recovery is scheduled to occur on March 2016. As part of the recovery evaluation process, a detailed inspection of a set of 356 damaged buildings was conducted for which different recovery stages were attributed on a building-by- building basis. Due to time constraints, it was not feasible to survey all the damaged buildings (1462); therefore, 356 were selected for inspection. These included all the red-tagged and a random sample of yellow tagged structures. The figure below illustrates the block groups in the city of Napa and the distribution of the 356 evaluated damaged buildings.

Evaluated yellow and red-tagged buildings in the city of Napa.

At each field evaluation, every building was assigned a binary code (0 or 1) defining the stage of the recovery, based on an exterior visual inspection. In this framework, 0 represents a “No Recovery” stage; and 1 is associated with the building’s “Full Recovery” in which the building is fully repaired/rebuilt and reoccupied. According to the first inspection, 179 of the 356 evaluated buildings were fully recovered six months after the earthquake, whilst one year later the recovery stage of 36 more buildings evolved from “0” to “1”. The figure below presents examples of the recovery progress of two buildings in the city of Napa over a period of one year. The images on the left side correspond to the first inspection (6 months following the event), whereas the images on the right side refer to the recovery stage of the same buildings at the time of the second field survey (12 months following the event).

Examples of the recovery progress for two structures in the city of Napa, 6 months (left hand images) and one year after the earthquake (right-hand images).

To determine the relationship between the collected set of disaster resilience metrics and the observed recovery outcomes over time, a parametric probabilistic model is proposed, allowing the treatment of uncertainties in a robust and statistically significant way. Specifically, a logistic regression model was calibrated to predict the probability of a “Full Recovery” occurring in each of the block groups for which resilience variables were collected. The socio-economic parameters constitute the independent variables (predictors) of the regression, while the recovery observations are the dependent (response) variables. Because the temporal evolution of the recovery process is also of interest, a variable time (t) was included as a predictor, assuming values of 6 and 12 months.

In order for the regression to be possible, each block group was assigned a recovery stage (0 or 1), to make the data analogous to the resilience variables that were collected at the block group level of geography. Here, a simulation procedure was devised. According to the simulation procedure, for each of a total of 1000 simulations, a single inspected building in each block group was selected at random and its recovery stage was attributed to the respective block group. The probability of recovery in each block group was then determined, for each of the 1000 simulations, based on the corresponding logistic model. As a result, a distribution of the probability was derived for each block group, for each time t, as a function of the selected set of independent variables. This approach reflects not only the random nature of the recovery process, but also the uncertainty associated with the limited number of assessed points in each block group. The latter is visually presented below where the median and lower bound of predicted probabilities of recovery (16% percentile) and the upper bound (84% percentile) are presented at 6 and 12-month intervals after the Napa event.

16%, 84% and 50% (median) percentile probabilities of recovery in the city of Napa, as determined by the proposed recovery model at 6 months after the event.


16%, 84% and 50% (median) percentile probabilities of recovery in the city of Napa, as determined by the proposed recovery model at 12 months after the event.

One of the main benefits of this methodology is that it allows for the prediction of recovery over time, based on a selection of proxy variables that overcomes problems arising from a limited number of recovery observations. Consequently, the method can be applied to areas outside the city of Napa, where the social structure is similar but no long-term recovery observations from an event are available. Furthermore, although not described herein, the recovery prediction model specifies the socio-economic parameters that influence the recovery speed and trajectory. This information, along with the prediction of recovery over time, can be utilised by policy makers to identify the weaknesses, as well as the strengths in the recovery potential of communities from a socio-economic perspective. Thus, early interventions and key priorities can be established during the whole cycle of disaster management, reflecting the actual needs of the affected population. The methodology can be further used by decision makers to re-define and/or enforce rehabilitation efforts since the model and its predictions can be updated as many times as additional recovery data is collected.

This study is supported by the State of California, Alfred E. Alquist Seismic Safety Commission and is part of a collaborative effort between the GEM and the University of California at Los Angeles (UCLA), Department of Civil and Environmental Engineering. The work is being conducted to support the development of Open-source software and standards for the prediction of recovery based on earthquake damage, socio-economic conditions of populations, and decision-making processes. For more details on the variables selection, modelling process and methodology, please contact us at the