Modeling Demographic Relocation in Response to Climate Risk Factors and Gentrification Displacement Pressures

Abstract

Climate risk factors, including wildfire, sea level rise, inland flooding, and extreme heat, as well as gentrification displacement pressures will be primary drivers of migration in the coming years. Travel demand modeling relies on reasonable and appropriate forecasts of demographic totals at the detail of travel analysis zones. Methodologies for developing scenarios in response to individual and combined climate risk factors are described, drawing on work undertaken for the Southern California Association of Governments SoCal Regional Climate Adaptation Framework. Methodologies for developing scenarios in response to gentrification displacement pressures of low-income workers are described, drawing on work carried out for the California Statewide Freight Forecasting and Travel Demand Model. These methodologies leverage modeling tools that are readily available to agencies, allowing for rapid testing of scenarios and integration with other planning processes. Climate adaptation and housing policy, respectively, are currently in need of greater integration and coordination. Future directions are explored to integrate these methodologies and create a combined demographic relocation model, sensitive to both climate risk factors and the affordability and gentrification displacement pressures arising out of shifting demand–supply dynamics and population–job balance in high growth areas.

Keywords

classification description: planning and analysis transportation planning policy and processes investment prioritization long-range plan megaregions sustainability and resilience transportation and sustainability transportation and economic development accessibility and mobility economic impacts environmental impact land development social impacts sustainable growth transportation equity transportation systems resilience natural hazards and extreme weather events climate change strategic relocation vulnerability and resilience assessment transportation infrastructure protection and preparedness community resilience hazard mitigation preparedness and protection resilience and risk management vulnerability and threat assessment

For this paper, the authors drew together their experiences modeling the relocation of future demographics in response to climate risk factors and gentrification pressures, respectively. The former draws on work undertaken for the Southern California Association of Governments (SCAG) SoCal Regional Climate Adaptation Framework; the latter on work carried out for the California Statewide Freight Forecasting and Travel Demand Model (CSF2TDM). Both projects performed travel demand modeling in their respective areas (Southern California and the entire state of California). Critical to such modeling was the development of scenarios as input data, namely future population, household, and employment in travel analysis zones (TAZs). This paper focuses on the different methodologies developed for the creation of these scenarios. In the case of climate risk factors, distinct scenarios were developed to model displacement and relocation from wildfire risk, sea level rise, inland flooding, and extreme heat health impacts. In the case of gentrification, scenarios were developed to model attraction into high-demand and transit-oriented districts and the resulting displacement of low-income workers owing to rising rents and costs.

Each of these projects, considered independently, offers valuable approaches and lessons learned for the best ways to model climate change- and gentrification dynamics. The authors are keenly aware that combining these approaches would lead to more accurate portrayals of future scenarios. After describing the relocation models developed for climate change and gentrification independently, we consider the limitations of each and the ways to combine the two for future work.

Literature Review

California is facing a severe crunch in affordable housing supply ( 1 ). Coupled with the widening wage gap, over 1.3 million households lack access to an affordable home. Between 2016 and 2019, California saw a net decrease of 13% in the number of affordable homes. As a result, over 70% of low-income households are “cost burdened,” paying over half of their incomes toward housing costs. Out of 1.3 million households vulnerable to displacement statewide, over 70% live in the Bay Area or Greater Los Angeles. Past studies have focused on the relationships between factors driving shifts in density across regions. In two separate studies ( 1 , 2 ) researchers predicted shared socioeconomic pathways, calling for mitigation and adaptation to climate change through changes in demographic-, economic-, policy-, technological-, and environmental indicators. With respect to mitigation, associated costs strongly depend on three factors: policy assumptions, socioeconomic projections, and the range of scenario targets.

A study by Berkhout et al. laid out representative heuristic tools to aid assessment of climate-change impacts on moving populations and the supporting infrastructure, such as employment centers ( 3 ). Studies have also concentrated on estimating potential biases caused by ignoring climate-induced changes when predicting demographic shifts (4, 5)—a limitation that this analysis specifically targets to isolate and address. Scenario analyses have evolved to be powerful tools in integrated assessment and policy analysis for climate change. Socioeconomic and climate scenarios are often combined to assess climate adaptation impacts and vulnerabilities across different sectors and to inform risk management strategies ( 6 ). Such combinations of scenarios also play an important role in enabling interactions between experts and other stakeholders, like developers, utilities, emergency responders, and public health officials, outlining issues and providing means for furthering policy interventions.

Some studies have explored the impacts of socioeconomic growth from the effects of climate change to identify the primary drivers of stress on the availability and adaptability of unique natural resources, for example, water resources ( 7 – 9 ), and biodiverse species of flora and fauna ( 10 – 12 ). These studies build a matrix of relationships between common socioeconomic drivers and climate-change factors germane to future growth scenarios. A quantitative approach ( 13 ) to modeling these relationships chose a 25 km² grid as a compromise between a larger-scale climate setting and more detailed spatial socioeconomic resolution.

It is already being reported that people residing in extreme heat conditions in several central American, African, and Indian communities have begun to migrate to places with more tenable climatic resources ( 14 ). Areas deemed climatically unlivable are projected to grow from 1% in 2020 to 19% in 2070. Economic models suggest that political responses to both climate change and migration could lead to drastically different futures. The integrated roles of scenarios, error propagation in linked models, model validity, transparency, and transportability were examined. Future scenario development involves regional and sectoral extensions of new generations of earth system climate models ( 3 ). Societal understanding of the equity implications of climate-change impacts and adaptation actions remains limited, however, studies have discovered geographic bounds within which vulnerable populations are likely to bear a disproportionate cost of adaptation ( 15 ). Disadvantaged communities are the first to be affected by many climate-related calamities. Martinich’s analysis, which accounted for the economic efficiency of adaptation approaches, found that ∼20% of people that will potentially be affected by sea level rise are socially vulnerable and much more likely to be abandoned than protected in the face of rising sea levels. In the Gulf region of the United States, ∼99% of the most socially vulnerable people live in areas that are unlikely to be protected from inundation, in stark contrast to the least socially vulnerable group: only ∼8% live in areas unlikely to be protected. Key improvements to address this include ensuring active participation by representatives of Disadvantaged Communities in adaptation planning processes, expanding adaptation to jurisdictions with low financial/infrastructural capacity, and integrating justice into urban design ( 16 , 17 ). For a nuanced understanding of the diverse makeup of communities and their associated susceptibilities, regionally specific climate-change impacts need to be included in evacuation/relocation planning ( 18 ).

The impact of mass migration on transportation is expected to be substantial. Beyond disproportionate displacement of disadvantaged communities in the face of climate adaptation, gentrification too plays a role in displacing vulnerable low-income populations. Transit-oriented development (TOD), neighborhood safety enhancement, and other social upliftment initiatives ( 19 ) can potentially lead to decreases in affordable housing supply. Since gentrification pushes traditional communities out of most newly developed urban cores, vehicle miles traveled (VMT) associated with them increase radically. Tradeoffs between antidisplacement and VMT reduction goals need to be studied in regional contexts and incorporated into travel models, especially since TOD seeks to reduce VMT. Subsidizing and mandating affordable units in new TOD construction could offset the resultant increase in demand. However, zoning laws and public outcry against dense developments hinder this. Withdrawal of private services from neighborhood structures, conversion of low-cost housing into condominiums, foreclosures, natural disasters, partition sales, redlining, and expensive rehabilitation of older units contribute to the dislocation of vulnerable households.

Estimating Migration Driven by Climate-Change Adaptation

California’s Fourth Climate Change Assessment ( 20 ) reports that as global temperatures continue to rise, Southern California expects to face extreme storms and temperatures, more widespread wildfires, and rising sea levels. These will all trigger displacement of population and employment. Related factors such as landslides, extreme wind, drought, agricultural pests and ecological hazards, and worsening air quality will further contribute to displacement. The SoCal Regional Climate Adaptation Framework developed several tools to support local jurisdictions’ and agencies’ efforts to adapt to climate change. As a regional metropolitan planning organization (MPO), this SCAG project focused on the six-county area encompassing Los Angeles, Orange, Ventura, Riverside, San Bernardino, and Imperial counties. As part of the framework, we conducted significant transportation and land-use analysis and modeling. We mapped four climate risk factors (wildfire, sea level rise, inland flooding, and extreme heat health impacts), overlaid on demographic data and a summary of demographics under various levels of risk for these factors. Scenarios of relocation for each risk factor were then created and input into the regional travel demand model. For this study, the base year was set as 2030 and risk projections were derived for 2045. We hypothesized that climate-adaptation-driven migration would begin at the base year and would continue over this period.

Data and Assumptions

We developed socioeconomic demographic (SED) relocation targets for the top four climate risks that warrant mass-scale migration: wildfire, sea level rise, inland floods, and extreme heat. We also simulated a fifth scenario: a combination of all four, modeling relocation while remaining sensitive to overlapping catchments for these risk factors, careful to not double count any populations and analyzing areas that would be newly stranded when faced with multiple risk factors. The zonal structure used for this analysis was the SCAG travel demand model TAZ system. Fractions of the population projected to migrate were developed based on subjective knowledge of climate resiliency, since this is a futuristic study for which we sought to model scenarios that had little to no historic precedence. The assumptions made to develop evacuation targets for these scenarios were:

Sea level rise—Areas affected by a global mean sea level rise of 1 m, under average storm conditions, were sourced from the Coastal Storm Monitoring System dataset. This dataset detailed estimated depths to which zones within the affected area are expected to inundate. We assumed 100% evacuation for flooded areas where the inundation depth was greater than 1 m (Figure 1). Further, any similarly affected transportation assets were removed from the network and any zone depending on said route would be deemed “stranded.”

Flood risk—Regions likely to be flooded more frequently than once in 100 years according to the Federal Emergency Management Agency were identified (Figure 2). Based on area ratios of TAZs flooded, we assumed a 10% evacuation from affected zones. No transportation assets were taken offline for this scenario.

Extreme heat health event days—The annual number of days considered heat health events (HHEs) based on temperature, humidity, tree cover, urban heat island impacts, elderly population, and historical hospital admissions and fatalities data were sourced from the California Heat Assessment Tool. Zones with more than 30 annual HHE days were considered candidate zones for demographic changes (Figure 3). This threshold was ascertained in consultation with SCAG as a reasonable number of days after which a community-scale response to the extreme heat conditions would be expected to be triggered. The rationale for this judgment was based on empirical evidence (e.g., mortality rates and hospitalization rates caused by HHEs such as stroke and rapid dehydration) dating back to heat events since 2006 ( 22 ). Based on empirical indications and in absence of a calibration measure since there is little recorded, measurable precedence of heat-prompted regional migration, the threshold of tolerable heat events was set at 30 days a year before any noticeable impacts on residential or commercial location decisions could be felt. It was reasoned that heat conditions could be endured up to a certain number of days (30 in this case) until they started affecting public health: this had to be captured in our estimation of how frequencies of annual heat events correlate with relocating populations/employment. This proportion of evacuation was assessed on a graduated scale equal to number of HHE days minus 30, calculated as a percent. For example, in zones with 45 HHE days a year, we would assume a 15% (45 − 30) evacuation. This metric was introduced assuming that the threshold of tolerable HHE days in a year should be close to a month (30 days) and any increase in HHE days would be linearly associated with migrating shares of the total population of the TAZ. The methodology/framework we propose allows for an easy update to this threshold considering new literature, data, or evidence relating migration threats or patterns to extreme heat risk.

Wildfire risk—Area boundaries expecting moderate, high, and very high risk of wildfire were sourced from the United States Forest Service database (Figure 4). Corresponding evacuation proportions were assumed to be 25%, 50%, and 75% respectively, based on actual fractions of Traffic Analysis Zones (TAZs) affected. Transportation assets were removed from the network only for the “very high” risk-category and any zone depending on said route would be deemed “stranded.”

Figure 1.

Sea level rise scenario color-coded by predicted inundation depths for 2030. Offset map is zoomed into Long Beach Port Area.

Figure 2.

(a) Flood plains within the SCAG region and (b) SCAG zones predicted to be impacted by floods in 2030.

Figure 3.

Annual heat health event days predicted in 2030; zones with more than or equal to 30 annual HHE days.

Figure 4.

Zones in SCAG region predicted to be impacted by wildfire in 2030 by severity.

SED Relocation Algorithm

Assumptions

All relocating populations, employment, and households remain within the SCAG region.

Zones receiving relocating demographics must be “safe” under the climate risk scenario being considered. For combined scenarios, receiving zones must be “safe” under all four scenarios.

Spatial units of finer resolution than TAZs, taken from SCAG’s scenario planning module zones, otherwise called subTAZs, were used to estimate the fraction of residents/workers/households contained in the affected zones. Since demographics are not uniformly distributed within TAZs, this allows more granular assessment of the population required to relocate. For each risk scenario, the fraction of people assumed to be relocating was thus calculated. This was to restrict the migrating population to originate solely from the scenario-affected portion of the TAZs.

As a simple stochastic approach to ascertain where the relocating population would presumably migrate to, we assumed these two criteria to probabilistically determine the receiving zones’ attractiveness for the displaced population.

Proximity of “receiver” TAZs from affected TAZ (sourced from TAZ-to-TAZ distance skims); and

SED growth potential of receiver TAZs between base year (2030) and target year (2045).

The odds of where relocating people would migrate to was assumed to be affected equally by both these factors. This assumption can easily be updated to account for various testing conditions for which one might need to simulate how distance or growth potential would play a more dominant role in determining the likelihood of resettlement by multiplying each probability with a suitable weight. The proximity probability is the ratio of distance between each TAZ pair to the sum of distances between all eligible donor–receiver TAZ pairs associated with the affected TAZ multiplied by −1 to represent the decreasing likelihood of receiving populations with increasing distance from the affected TAZ. The growth-potential probability denoted the proportion of growth potential of each receiver TAZ to the sum of growth potentials of all eligible receiver TAZs associated with that risk-afflicted TAZ.

The SED growth potential of each receiver TAZ was calculated by finding the corresponding difference between SCAG-estimated 2045 and 2030 populations. TAZs with a decline in population/households between 2030 and 2045 were not made eligible to receive any relocating populations.

A “probability sum,” obtained by adding the two calculated probabilities, was assigned to all TAZ pairs where this sum was greater than 0 (Equation 1). A negative sum would imply that the TAZ pairs are located at a great distance from one another and that the predicted growth-potential when compared with other receiver zones is insufficient to offset distances between affected- and receiver zones. Equations 1 and 2 mathematically represent these definitions.

Growth - potential Probability = \frac{Δ Estimated Population (2045 - 2030) for TAZ A and TAZ B}{\sum i, j \to n Δ Estimated Population (2045 - 2030) for TAZ i and TAZ j}

(1)

Proximity Probability = \frac{Distance between TAZ A and TAZ B}{\sum i, j \to n Distance between TAZ i and TAZ j}

(2)

An all-or-none approach would be a different and slightly more deterministic technique to reassign migrating populations—this approach would assume that receiver zones from donor–receiver pairs with the highest “probability sum” will absorb all the population originating from corresponding donor zones until they have no remaining growth potential. The process would then seek to exhaust the growth potential of receiver halves of the second highest donor–receiver “probability sum” and so on until all relocating populations get allocated among receiver zones in the top-ranking donor–receiver pairs. More straightforward to formulate and visualize, this would be an effective hypothesis to test and spatially isolate the predicted hotspots of human migration in the event of climate emergencies. However, the purpose of this study was to assess the impact of such an exodus on the transportation infrastructure of the affected zones—both donor and receiver. To draw insights on how travel network links expected to be included among both evacuation routes and in the concerned regions would be affected, we needed a more randomized distribution of the migrating population that most closely mimicked the real-life movement of people.

In contrast to the all-or-none approach, the stochastic method of reassignment, in this context, assumes that people could potentially relocate to even distant receiver zones given high enough growth incentives. Similarly, we theorized that receiver TAZs with marginally lower employment or household growth-potential would still be attractive enough for migrants, albeit at a quantifiably lower propensity. The probabilistically ranked assignment approach used in this analysis was expected to account for this uncertainty. This is not unlike a gravity model used for generating trips in a traditional four-step model based on trip attraction and production components. In an improved version of this model, variability introduced by other heuristic factors like occupational availability and real estate prices could also be captured using a process analogous to gravity models—generating utility functions with appropriate weights assigned to the explanatory variables.

As receiver zones get updated with the reallocated population in an incremental fashion according to the rank of their probability scores in this first iteration of assignment, it is possible that the sum of population assigned to the receiver zones featuring in multiple donor–receiver pairs exceed their growth potential or “receiving” margin. In such cases, the subsequent iterations of the reallocation algorithm calculate the sum of excess population from all receiving zones, sending them back for reassignment to the remaining unfilled eligible receiver zones. Figures 5 to 9 show where populations relocated from and reallocated to, for the five different scenarios.

Figure 5.

Population added: (a) total and (b) as a percentage of existing zonal population, owing to extreme heat conditions predicted for 2030.

Figure 6.

Population added: (a) total and (b) as a percentage of existing zonal population, owing to wildfire conditions predicted for 2030.

Figure 7.

Population added: (a) total and (b) as a percentage of existing zonal population, owing to floods predicted for 2030.

Figure 8.

Population added: (a) total and (b) as a percentage of existing zonal population, owing to sea level rise predicted for 2030.

Figure 9.

Map represents reallocated (added) population to TAZs (each point ~100 people added) when all four scenarios (wildfire, sea level rise, extreme heat conditions, and floods) are assumed to occur simultaneously in 2030.

Table 1 summarizes total predicted relocating population within and for the SCAG region under four separate scenarios shown in Figures 5 to 8 and the combined scenario, in which all four climate risk scenarios occur simultaneously (Figure 9). For zones with overlapping risks in the combined-risk scenario, the risk type warranting the highest number of people/employment/households to migrate was identified and associated with that respective zone. This prevents repeated counting of the relocating population and overestimation.

Table 1.

Summary of Reallocated Population for Four Scenarios and the Combination Scenario Predicted for 2030 in the Southern Coast Region

	Reallocated population	Reallocated employment	Reallocated households
Independent risk scenarios
Moderate risk wildfire	140,835	45,546	46,662
High-risk wildfire	273,644	73,497	90,747
Very high-risk wildfire	46,405	10,562	17,482
Total wildfire	460,884	129,605	154,891
Flood risk	31,115	15,809	10,517
Sea level rise (1 m) risk	46,053	28,888	21,593
Extreme heat risk	306,432	112,744	103,333
Total	844,484	287,046	290,334
All risk scenarios combined
Moderate risk wildfire	136,339	44,363	45,072
High-risk wildfire	272,104	72,857	90,186
Very high-risk wildfire	46,116	10,369	17,395
Total wildfire	454,559	127,589	152,653
Flood risk	19,544	11,197	6,234
Sea level rise (1 m) risk	45,058	28,137	21,157
Extreme heat risk	192,198	71,801	63,280
Total	711,359	238,724	243,324

Displacement of Low-Income Households Owing to Unaffordable Housing

This section describes methodology to simulate displacement of low-income households in high growth urban areas in the Greater Los Angeles area and the Bay Area. With these regions expected to continue as drivers of economic growth and drawing an ever-growing number of white-collar and blue-collar jobs over next couple of years, substantial new “blue-collar” households will be unable to afford housing in these areas and will be pushed to lower density, more affordable regions on urban peripheries. As a result, workers in these households will need to travel longer distances for work trips, adding additional VMT and congestion on transportation infrastructure that was stressed to begin with. As urban areas push to reduce and restrict industrial development, wholesale and transport jobs (such as warehousing, trucking, shipping) are also shifting to sparse territories on urban outskirts. Together, these forces are expected to influence job and population balance in these regions.

State policy makers have several tools to test the impacts of such scenarios, including the statewide travel demand model. California has a very detailed travel demand and freight-forecasting model that models activity patterns and freight movements in the state at a disaggregated temporal and spatial level. For this study, the base year was set as 2015 and the displacement impacts were modeled on new households projected to reside in the two regions between 2015 and 2040. The study areas were selected because

Both are economic centers in the state;

The cost of living is disproportionately higher than other parts of California ( 23 );

They account for the bulk of blue-collar jobs in California;

Urban areas are expected to reduce and restrict industry development for the job sectors under consideration, contributing to large-scale relocation; and

The demographic profiles of blue-collar workers are different in the two regions.

The methodology described in the following sections is tailored to the specific planning scenario of interest to Caltrans (region definitions, sociodemographic and employment characteristics of migration candidates, size of migration pool, restrictions on migration patterns, etc.). However, the methodology is robust and flexible so it can be adapted to scenarios in which one or more of these constraints/assumptions are relaxed/modified.

Data, Assumptions, and Methodology

There are several steps in modeling this scenario.

Synthetic Database Generation

The main inputs to this scenario development exercise were the 2015 and 2040 synthetic population databases developed as a part of the CSF2TDM. These synthetic “actors” have detailed sociodemographic attributes (worker status, household income, home location TAZ, etc.) assumed to be representative of actual sociodemographic profiles within the population. The team used an open-sourced software called PopGen ( 24 ), based on an iterative heuristic algorithm that draws actors from a seed dataset until the population and household attributes are matched to the actual distributions observed in the population (control totals). The seed dataset used for this analysis was the 5-year American Community Survey (ACS) Public Use Microdata Sample (2011 to 2015), and for control totals, the ACS 2015 5-year estimates were used. This detailed information allows a disaggregated approach to selecting candidate households when modeling migration.

Geographic Scope of Migration

Figure 10 shows the two regions modeled in this study, both of which account for over 70% of households facing the risk of displacement because of insufficient affordable housing in California. We assumed that migrating households residing in donor counties would move to locations within the recipient counties based on a probabilistic approach described in the next section. We also assumed no transfers between the two regions, so residents of one region relocated only to the recipient counties identified for that region:

Greater Los Angeles area

Donor counties: Los Angeles and Orange

Recipient counties: San Bernardino and Riverside

Bay Area

Donor counties: Alameda, Contra Costa, San Mateo, Santa Cruz, and Santa Clara

Recipient counties: Merced, Stanislaus, San Joaquin, Sacramento, Solano, Yolo.

Figure 10.

Donor and recipient counties for the Bay Area and Greater Los Angeles area.

Sociodemographic Characteristics of Migration Candidates

This study considered the impacts of the housing crisis on projected growth between 2015 and 2040, therefore only the new households added to the region between the two model years were considered in the migration pool. Within this pool, the only households considered as potential migration candidates had at least one member of the household working in a blue-collar occupation, which are listed with Standard Occupational Classification codes:

• Farming, fishing, and forestry occupations (45)

• Construction and extraction occupations (47)

• Installation, maintenance, and repair occupations (49)

• Production occupations (51)

• Transportation and Material Moving (53).

Displacement likelihood is highly correlated with household income: lower income households are more likely to be unable to keep pace with escalating housing costs and forced to relocate to areas with more affordable housing. This is evident in the statistics provided by the National Low-Income Housing Coalition ( 25 ): the proportion of cost-burdened households (who pay over 50% of their income on housing costs) for extremely low-income households (0% to 30% of area median income [AMI]) and very low-income households (31% to 50% of AMI) is over 85%, around 65% for the next income group (51% to 80% of AMI), and this drops to around 38% for middle income households (81% to 100% of AMI). The number drops precipitously moving toward the other end of the income spectrum.

To reflect this cost-burden curve in the displacement likelihood, households were classified into three tiers based on their income in 2010 dollars: low income (less than $35,000), middle income ($35,000 to 75,000), and high income (more than $75,000). In a simplified approach to calculate the number of households from each income tier that are displaced, we assumed 100% of low-income households in the migration pool (blue-collar worker households added between 2015 and 2040) being forced to relocate. For the middle and high income groups, the displacement percentages assumed were 50% and 25% respectively. These percentages were selected based on deliberation with Caltrans staff to reasonably reflect the likelihood of displacement based on household income. They were comparable to the proportion of cost-burdened households in each income bracket.

Table 2 shows the migration candidates flagged based on their home location, household income tier, percentage of migration pool in each class selected for relocation, number of households in 2015 and 2040, and number of households relocated after applying the rules on the right-hand side of the table. A total of 52,175 households were selected for migration to the recipient counties.

Table 2.

Migration Flags for Household Groups, % of New Households Relocated, and Number of Households Moved

Home location	Household income tier (2010 dollars)	Migration flag	Percent of new HHs moved (2040 to 2015)	HHs in 2015	HHs in 2040	No. of HHs moved—% of new HHs moved × (HHs in 2040 – HHs in 2015)
Bay Area	Low income	1	100	5,428	6,339	911
Bay Area	Middle income	2	50	10,745	15,947	2,601
Bay Area	High income	3	25	25,709	39,043	3,334
Greater LA area	Low income	4	100	86,698	113,443	26,745
Greater LA area	Middle income	5	50	126,040	146,423	10,192
Greater LA area	High income	6	25	105,197	138,764	8,392
			Total	359,817	459,959	52,175

Note: HHs = households; LA = Los Angeles.

With the criteria in place, first a sampling space for each household group (Migration Flags 1 to 6) was generated using the synthetic households database and then random draws of size N (where N is number of households moved) yielded the households selected for the relocation. This resulted in a fairly uniform selection of household types (single attached, single detached, multifamily, etc.) and structures (number of members, presence of children), though in reality certain households may have a greater impedance than others, for example, households with children may bear higher costs to not relocate to the outskirts where schools and childcare facilities may be limited. A more stratified approach to sampling was considered, but not used for this sketch-planning-level exercise.

Recipient Location Receptibility and Relocation

The home locations of the households were tagged to the CSF2TDM TAZ (different from the SCAG TAZ structure used in the climate-related movements). Since this study used a statewide model, the spatial resolution was coarser than what a typical MPO such as SCAG would use. Once the migrating households were selected, the next step was to allocate them to a TAZ in the recipient counties. The receptibility of a TAZ, or the relative attractiveness of a TAZ to the incoming households, can be calculated using a range of methods of varying complexity: from simply apportioning based on population, employment, or both, to more complicated gravity models, which can additionally include factors such as land-use, network connectivity, and growth potential.

Housing costs in the recipient TAZs would be an ideal predictor of receptibility. For families being displaced because of high costs in the urban core, realistic relocation would target TAZs with a higher share of affordable housing. However, there might be other barriers to entry in neighborhoods with higher average incomes, for example, wealthier communities often restrict the construction and maintenance of affordable housing, which compounds the problems for migrating households. Low-income families also typically move with very little lead time, making it more difficult to locate housing outside of existing low-income neighborhoods.

However, there were no data on average housing costs at the disaggregated level used in the model. In the absence of such a dataset, the closest proxy variable was the proportion of low-income households (income less than $35,000 in 2010 dollars). A probabilistic approach was used for which the receptibility of the TAZS was calculated based on the percentage of low-income households in the TAZ. Probabilistic draws of recipient TAZs the following probability weights (odds) were applied based on the following 2040 proportion of low-income households (These thresholds were arrived at based on discussions with modeling staff at Caltrans):

Less than 25%: 0.1

Between 25% and 50%: 0.2

Between 50% and 75%: 0.3

Between 75% and 100%: 0.4

Once TAZs were mapped to the percentage of low-income resident categories, for each migrating household group, recipient TAZs were drawn from the sampling space consisting of all TAZs in the recipient counties of that region: one TAZ was drawn per household, and the household home location was updated to the new TAZ.

Job Relocation

Relocation of the workforce is only half of the picture; in 2040, it is also anticipated that certain industries such as wholesale, warehousing, trucking, and so forth, will move to the lower density areas outside the urban cores for a litany of reasons—space restrictions, regulatory restrictions, and availability of workforce (which is a direct impact of the migration of low-income workers). In particular, industries tied to the e-commerce industry are expected to dominate in peripheral areas, with cities such as Moreno Valley clearing way for megawarehouses ( 26 ). For the purposes of this study, wholesale and transport jobs in Bay Area donor counties (Alameda, Contra Costa, San Mateo, Santa Cruz, Santa Clara) and Los Angeles area counties (Los Angeles, Orange) in 2040 were reduced by 50%. These jobs were then apportioned to the recipient counties using a proportion of 2040 jobs before the reshuffle. Since the modeling framework predicts the work location for every worker, some of the displaced workers were designated a new workplace closer to their new homes—therefore, dampening the congestion and VMT impacts attributed to longer commutes and altered travel patterns.

Results

Table 3 shows the migration flows of households at the county level. Figures 11 and 12 show the change in populations at the TAZ level in the Bay Area and Los Angeles area (zoomed into regions with the biggest flows). As expected, the new households were relocated in and around major population centers in the recipient counties in the Bay Area (Stockton, Sacramento, Modesto, Davis, Vacaville/Fairfield) and for the Los Angeles area (City of San Bernardino and Riverside, Palm Springs, Corona, Ontario). East Bay (Alameda County) and the eastern flank of South Bay (Santa Clara County) have high concentrations of blue-collar workers in the Bay Area. In the Los Angeles area, the highest concentration of displaced population areas is in South Central LA, Compton, Downey, and Santa Ana. It is clear that the proportion of households that met the criteria for migration was much higher in the Los Angeles area than the Bay Area. The Los Angeles area also had a significantly larger proportion of the blue-collar jobs (that were expected to relocate because of escalating costs in the core areas) owing to its proximity to two large ports (Los Angeles and Long Beach) and the international border with Mexico.

Table 3.

Original Home Locations and Changed Home Locations of Relocated Households (County Level)

Old home county	New home county
	Merced	Sacramento	San Joaquin	Solano	Stanislaus	Yolo	San Bernardino	Riverside	Total
Alameda	155	1,050	415	148	185	150	-	-	2,103
Contra Costa	76	545	224	70	135	81	-	-	1,131
San Mateo	65	421	156	41	78	56	-	-	817
Santa Clara	189	1,199	520	186	260	209	-	-	2,563
Santa Cruz	22	115	44	18	21	12	-	-	232
Los Angeles	-	-	-	-	-	-	18,132	18,755	36,887
Orange	-	-	-	-	-	-	4,183	4,259	8,442
Total	507	3,330	1,359	463	679	508	22,315	23,014	52,175

Figure 11.

Population change in 2040 by TAZ after migrations are simulated: Bay Area.

Figure 12.

Population change in 2040 by TAZ after migrations are simulated: Los Angeles area.

Table 4 shows the changes in daily auto and truck trips and VMT for 2040, calculated between the post- and premigration scenarios. There was a net increase in daily auto trips and VMT, which was attributed to some of the migrated workers traveling further to reach their workplace. This was counteracted, to some extent, by the relocation of some blue-collar jobs to the recipient counties, which allowed for some of the migrated workers to have a workplace closer to their new home locations ( 21 ). The net decrease in truck VMT and trips was attributed to how trucks could now operate in less restrictive traffic conditions (avoiding going through congested core areas), carry more commodities in larger vehicles, and leverage the increased capacity in warehouses and factories.

Table 4.

Daily Trips and VMT Impacts of Estimated (Gentrification-Related) Migration for Southern and Northern California Counties

	County	Change in 2040 metrics (post-migration–pre-migration)
		Daily auto	Daily auto	Daily truck	Daily truck VMT
		Trips	VMT	Trips
Donor counties	Alameda	−45,000	−309,200	−1,500	−128,100
	Contra Costa	−19,900	−94,800	−800	−32,700
	San Mateo	−31,900	−166,200	−800	−29,900
	Santa Clara	−43,100	−178,300	−1,700	−52,500
	Santa Cruz	−4,400	−9,800	−100	−3,600
	Los Angeles	−545,800	−2,027,000	−15,000	−774,600
	Orange	−134,700	−597,800	−6,100	−256,000
	Total	−824,800	−3,383,100	−26,000	−1,277,400
	Merced	14,600	52,200	100	−46,200
	Sacramento	68,700	408,400	1,000	8,600
	San Joaquin	19,200	12,600	100	−40,100
Recipient counties	Solano	14,700	69,500	300	−11,100
	Stanislaus	14,800	56,800	200	−28,400
	Yolo	15,500	118,100	200	−5,800
	San Bernardino	402,500	2,479,600	3,900	−187,800
	Riverside	331,800	2,181,900	3,400	−88,300
	Total	867,200	5,326,900	9,100	−352,900
Net change		42,400	1,943,800	−16,900	−1,630,300
		Base year totals (overall travel in regions)
Donor counties		77,850,200	474,075,100	651,600	23,520,800
Recipient counties		37,807,300	253,974,100	312,000	30,671,000

Note: VMT = vehicle miles traveled

Conclusions and Future Directions

The methodologies used in each of these approaches to relocation modeling, for climate change and gentrification pressures respectively, can benefit each other if integrated. The algorithms outlined for demographic relocations in response to climate factors took into account proximity to eligible receiving zones as well as the gravity factor of the attractiveness indicated by later horizon year growth-potential forecasts. A weakness of this approach, however, is not incorporating market-based real estate dynamics. In reality, relocation choices will be heavily influenced by the income levels of those evacuating high-risk zones and the affordability of eligible receiver zones. This dynamic is iterative in nature—as climate risk evacuees increase demands on lower risk areas, housing and office prices, in both the purchase and rental markets, will rise in response. As mentioned, additional factors can be incorporated into the prediction algorithm in the form of weighted utility functions when data pertaining to real estate price points become available. It should be kept in mind that not all of those seeking to relocate to particular zones will be able to. And those who do relocate to their zones of choice will displace others who now find that area too expensive/overcrowded, thereafter contributing to possible secondary and tertiary waves of relocation. These displacement dynamics were captured in the relocation modeling focused just on gentrification pressures, which, by definition, were tied more closely to people’s propensity to congregate in highly desirable living/working zones, market dynamics, and the availability of suitable occupations. However, the gentrification displacement modeling did not incorporate any consideration of climate risk. In reality, the displacement factors rooted in gentrification will worsen in future years as climate factors increase overall migration, both intra- and interregional. A combined relocation model would potentially start with the climate evacuation module and then move into the gentrification displacement module while maintaining the limits on eligible receiver zones determined by the climate evacuation module. The focus of this research was to arrive at a practicable framework to design inputs for more inclusive travel demand models that are sensitive to such often overlooked explanatory factors that bring significant changes in land use over extended time periods.

Although gentrification displacement is more widely studied and has a larger body of data available, data accessibility for modeling climate risk evacuation is relatively sparse and primitive. The ratios of relocation posited for given risk factors and severity categories in our climate-relocation modeling methodology here were an educated hypothesis for outlining possible best practices. These ratios of relocation for various factors and severities, as well as patterns by which we determined where climate evacuees relocated, had to be grounded in real data as these migration patterns will manifest to greater degrees in the coming years. We know that certain regions of the world are already experiencing severe climate impacts more than others and we can look for ways to transfer patterns between contexts. We could potentially look for and apply before and after patterns from recent wildfire events, for example.

Both analyses are fundamentally different—the former attempted to very conservatively model a postfactual future using assumptions based on subjective empirical evidence, whereas the latter sought to model gentrification displacement across various demographic groups using forecasts of sociodemographic and employment trends. The magnitudes of migration, were of course, also sensitive to the assumptions and tools used for the respective scenarios, and thus cannot directly be compared across the two case studies presented in the paper—an integrated module that allows for interdependencies between the two displacement pressures would be the obvious next step.

The housing policy implications of this research are significant. Within California, different departments focus on climate adaptation and housing policy, including affordable housing. The state mandated Regional Housing Needs Allocation process requires each MPO to determine their needs for future housing supply given the forecasts about population growth and the conditions of existing housing supply. Out of each MPO’s regional targets, local jurisdictions are consulted and then assigned their fair share of new housing supplies, development of which to zone and plan for could be defined as one of the major impetus for undertaking this study. However, this whole process happens largely independent of climate adaptation analysis and planning. Many coastal communities tasked with significant growth of future housing supply are at high risk of sea level rise; the existing population should be proactively evacuated from such locations, never mind prohibited from additional growth. Strong coordination between climate- and housing policy is critical to regional and state planning. Building integrated models as described in this paper will be important for this coordination.

Footnotes

Acknowledgements

The authors thank Southern California Association of Governments and Caltrans for agreeing to share the findings in the two case studies. We also express thanks to Ramesh Thammiraju, Ishraq Ahmed, and Mobashwir Khan of Cambridge Systematics, Inc. who helped run the travel demand models.

Author Contributions

The authors confirm contribution to the paper as follows: study conception and design: S. Roy, P. Vinayak, D. Von Stroh; data collection: S. Roy, P. Vinayak, D. Von Stroh; analysis and interpretation of results: S. Roy, P. Vinayak, D. Von Stroh; draft manuscript preparation: S. Roy, P. Vinayak, D. Von Stroh. All authors reviewed the results and approved the final version of the manuscript.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

ORCID iDs

Sneha Roy

Pragun Vinayak

David Von Stroh

References

California Housing Partnership Corporation. Housing Needs Dashboard https://chpc.net/housingneeds. Accessed July 30, 2020.

Riahi

Van Vuuren

D. P.

Kriegler

Edmonds

O’Neill

B. C.

Fujimori

Bauer

Calvin

Dellink

Fricko

Lutz

The Shared Socioeconomic Pathways and Their Energy, Land Use, and Greenhouse Gas Emissions Implications: An Overview. Global Environmental Change, Vol. 42, 2017, pp. 153–168.

Berkhout

Hertin

Jordan

Socio-Economic Futures in Climate-Change Impact Assessment: Using Scenarios As ‘Learning Machines. Global Environmental Change, Vol. 12, No. 2, 2002, pp. 83–95.

O'Neill

B. C.

Kriegler

Ebi

K. L.

Kemp-Benedict

Riahi

Rothman

D. S.

van Ruijven

B. J.

van Vuuren

D. P.

Birkmann

Kok

Levy

The Roads Ahead: Narratives for Shared Socioeconomic Pathways Describing World Futures in the 21st Century. Global Environmental Change, Vol. 42, 2017, pp. 169–180.

Mjelde

J. W.

Sonka

S. T.

Peel

D. S.

The Socioeconomic Value of Climate and Weather Forecasting: A Review. Research Report 89-01. Illinois State Water Survey, Champaign, 1989.

Berkhout

van den Hurk

Bessembinder

de Boer

Bregman

van Drunen

Framing Climate Uncertainty: Socio-Economic and Climate Scenarios in Vulnerability and Adaptation Assessments. Regional Environmental Change, Vol. 14, No. 3, 2014, pp. 879–893.

Fant

Schlosser

C. A.

Gao

Strzepek

Reilly

Projections of Water Stress Based on an Ensemble of Socioeconomic Growth and Climate-Change Scenarios: A Case Study in Asia. PLoS One, Vol. 11, No. 3, 2016, p. e0150633.

Rogers

Assessing the Socioeconomic Consequences of Climate-Change on Water Resources. Climatic Change, Vol. 28, No. 1–2, 1994, pp. 179–208.

Sabler

D. G.

Shortle

J. S.

Climate-Change and Agriculture in the Mid-Atlantic Region. Climate Research, Vol. 14, No. 3, 2000, pp. 185–194.

10.

Alcamo

Leemans

Kreileman

Global Change Scenarios of the 21st Century: Results from the IMAGE 2.1 Model. Elsevier, Oxford, 1998.

11.

Brown

J. L.

Weber

J. J.

Alvarado-Serrano

D. F.

Hickerson

M. J.

Franks

S. J.

Carnaval

A. C.

Predicting the Genetic Consequences of Future Climate-Change: The Power of Coupling Spatial Demography, The Coalescent, and Historical Landscape Changes. American Journal of Botany, Vol. 103, No. 1, 2016, pp. 153–163.

12.

Doak

D. F.

Morris

W. F.

Demographic Compensation and Tipping Points in Climate-Induced Range Shifts. Nature, Vol. 467, No. 7318, 2010, pp. 959–962.

13.

Holman

I. P.

Rounsevell

M. D.

Shackley

Harrison

P. A.

Nicholls

R. J.

Berry

P. M.

Audsley

A Regional, Multi-Sectoral and Integrated Assessment of the Impacts of Climate and Socio- Economic Change in the UK. Climatic Change, Vol. 71, No. 1–2, 2005, pp. 9–41.

14.

Lustgarten

. The Great Climate Migration. https://www.nytimes.com/interactive/2020/07/23/magazine/climate-migration.html.

15.

Martinich

Neumann

Ludwig

Jantarasami

Risks of Sea Level Rise to Disadvantaged Communities in the United States. Mitigation and Adaptation Strategies for Global Change, Vol. 18, No. 2, 2013, pp. 169–185.

16.

Shi

Chu

Anguelovski

Aylett

Debats

Goh

Schenk

Seto

K. C.

Dodman

Roberts

J. T.

Roadmap Towards Justice in Urban Climate Adaptation Research. Nature Climate-Change, Vol. 6, No. 2, 2016, pp. 131–137.

17.

Maru

Y. T.

Smith

M. S.

Sparrow

Pinho

Dube

O. P.

A Linked Vulnerability and Resilience Framework for Adaptation Pathways in Remote Disadvantaged Communities. Global Environmental Change, Vol. 28, 2014, pp. 337–350.

18.

Zsamboky

Fernández-Bilbao

Smith

Knight

Allan

Impacts of Climate-Change on Disadvantaged UK Coastal Communities. Joseph Rowntree Foundation, UK, 2011.

19.

Chapple

Waddell

Chatman

Zuk

Loukaitou-Sideris

Ong

Gorska

Pech

Gonzalez

S. R.

Developing a New Methodology for Analyzing Potential Displacement. California Air Resources Board and the California Environmental Protection Agency, 2017.

20.

Kalansky

Cayan

Barba

Walsh

Brouwer

Boudreau

California’s Fourth Climate Change Assessment. California Natural Resources Agency and California Energy Commission, 2018.

21.

Cambridge Systematics – Location Based Services Suite (Locus). www.camsys.com/locus.

22.

California Heat Assessment Tool Whitepaper – California Energy Commission Technical Report. https://www.energy.ca.gov/sites/default/files/2019-07/PublicHealth_CCCA4-CNRA2018-012.pdf Accessed July 2019.

23.

Cost of Living in California. https://www.salary.com/research/cost-of-living/ca.

24.

Konduri

K. C.

You

Garikapati

Pendyala

R. M.

Enhanced Synthetic Population Generator That Accommodates Control Variables at Multiple Geographic Resolutions. Transportation Research Record: Journal of the Transportation Research Board, 2016. 2563: 40–50.

25.

National Low-Income Housing Coalition. 2020California Housing Profile. https://nlihc.org/sites/default/files/SHP_CA.pdf. Accessed July 30, 2020.

26.

Streetsblog. Targeting a Southern California Mega-Warehouse Project. https://earthjustice.org/cases/2015/targeting-sprawling-southern-california-mega-warehouse-project. Accessed July 30, 2020.