Abstract
Local governments tend to produce disjointed planning and implementation responses to ambiguous problems. This study adapts a cognitive science framework known as active inference to investigate local government resilience planning and management processes. We contend that disconnects between planning and management reflect the inadequacy of the “knowledge infrastructure” systems which could enable more adaptive responses. Using a mixed-methods approach to study stormwater resilience in the U.S. state of Indiana, we find evidence that municipalities integrate planning and management efforts relative to experienced flooding impacts. However, they are less likely to proactively develop information on longer-term or more distant climate impacts.
Introduction
Climate change is complicating local government planning and management processes in ways which are difficult to quantify, mitigate or adaptively respond (Deslatte, Krause, and Hawkins 2022). As in other countries with aging infrastructure, increasing temperatures and extreme precipitation events are already disrupting transportation systems while increasing flooding and droughts across the United States, likely necessitating major adaptive or transformative investments in coming years (Cook, McGinnis, and Samaras 2020; Henstra, Thistlethwaite, and Vanhooren 2020; Wiechman et al. 2024). Despite the exponential increases in information available to modern societies, climate change risks are also exposing significant knowledge gaps for local policymakers and managers (Chu et al. 2023).
This study adapts a novel cognitive science framework known as active inference to investigate the information-processing which underwrites local government resilience planning and management processes (Friston 2010; Sajid et al. 2021). Active inference describes the efforts all goal-directed agents make to develop or improve predictions about their environment in order to sustain themselves (Friston et al. 2017; Parr, Pezzulo, and Friston 2022). From this perspective, the goal for any self-organizing system is straight-forward: to minimize surprise and stave off entropic decline of functionality (Clark 2023; Constant et al. 2022). Organizations, like humans, engage in an active inference process when they seek and curate information to guide budgeting, strategic, or long-range planning (Deslatte, Koebele, and Wiechman 2024). Governments seeking to sustain specific public goods or services must actively seek feedback on system conditions and adjust programs, policies or management processes when they detect unacceptable risks (Anderies et al. 2013). The challenge then becomes how accurately or proactively they can prepare for the future. Plans are artifacts of this collective inference process.
Despite the seemingly intuitive nature of this argument, ambiguous problems create difficulties linking actions to outcomes and introduce inherent uncertainty about future scenarios (Meerow, Newell, and Stults 2016). Challenges like climate change—which introduce ambiguity into infrastructure investment and planning decisions—complicate the ability of policymakers and managers to accurately identify problems and formulate efficacious responses (Berardo and Scholz 2010; Bozeman and Kingsley 1998; Bullock, Greer, and O’Toole 2019; Chun and Rainey 2005).
Local governments, in particular, tend to produce disjointed planning and implementation efforts when confronted with ambiguity (Bryson, Edwards, and Van Slyke 2022; Deslatte, Krause, and Hawkins 2022; Poister 2010). For instance, local governments typically plan for growth and development through comprehensive plans, but seldom extend these efforts beyond economic, land-use or “quality of life” goals (Brinkley and Stahmer 2024; Deslatte, Chung, and Stokan 2023). Since the early 2000s, local governments pursuing climate, sustainability or resilience goals have tended to adopt stand-alone plans, but they may not be integrated into existing capital improvement, budgeting, or policy processes (Woodruff et al. 2022).
We contend that disconnects between planning and management reflect the inadequacy of the information feedback processes—referred to in the resilience literature as “knowledge infrastructure” systems—which could enable more adaptive responses (Anderies 2015; Favero, Meier, and O’Toole 2014; York et al. 2021). While there are many definitions for resilience across academic disciplines, the urban planning literature defines resilience as the ability of a system, and all its socio-ecological components across temporal and spatial scales, to maintain or quickly return to desired functions following a disturbance (Meerow, Newell, and Stults 2016). Principles for resilience planning include setting ambitious but realistic goals, linking the strategies for achieving them to a strong fact basis or awareness of system dynamics, and the consideration of different scenarios to embrace the inherent uncertainties of dealing with complex systems (Woodruff et al. 2022). This definition imposes considerable need for different types of information, including knowledge about the history, interactions and exposures of specific physical and social components of a community (Anderies, Mathias, and Janssen 2019). Evidence suggests most local governments fail to account for uncertainty in climate adaptation, resilience and flood management planning (Roy et al. 2024; Woodruff et al. 2022), as well as who is benefiting from climate-related policies or investments (Meerow and Newell 2019). Drawing from this planning, cognitive science and local sustainability research, our study asks: what types of information and organizational capacities drive efforts to integrate resilience planning and adaptive management efforts?
We address this question using a Bayesian mixed-methods approach which combines interview, text analysis and survey data to study municipal government stormwater resilience efforts in the U.S. state of Indiana. We find evidence that local governments integrate evidence-gathering and management efforts in relation to experienced flooding impacts. However, they are less likely to develop information on longer-term or more distant climate threats and integrate them into resilience-related decision-making. They reactively rather than proactively adapt to flooding-related climate stressors.
We conclude that understanding how governments confront the chaos of environmental change requires paying greater attention to gaps in the “knowledge infrastructure” systems which allow them to assess current and future threats, and thus adapt more effectively, proactively and equitably (Anderies, Barreteau, and Brady 2019; Ostrom 2010; Walker et al. 2023). We label the iterative process of doing so collective action inference and conclude with suggested next steps for local government researchers and practitioners.
Planning and Managing for Stormwater Resilience: A Collective Action Inference Approach
Despite its ubiquity in all levels of governments, planning remains a term of art to local government administrators (Bryson, Edwards, and Van Slyke 2022; Burby and Dalton 1994; Poister 2010). An inevitable—even habitual—element of management, planning engenders critiques and differing views over its intrinsic value to public organizations and the extent to which it supports other management functions or processes (Poister and Streib 2005). A central element of this dilemma stems from an inherent mismatch between complex, ambiguous threats like climate change and the capabilities of existing governmental planning and management processes (Bovaird 2008; Ladyman and Wiesner 2020). Applying an active inference lens, we posit that more specific or localized awareness of vulnerabilities and risks from climate impacts like flooding—via knowledge infrastructure systems—will produce more integrated efforts to mitigate those risks (Anderies, Mathias, and Janssen 2019; Veissière et al. 2019).
Specifically, the resilience literature identifies at least four types of knowledge produced by networks of governmental agencies, universities, media, nonprofit organizations and community groups (for a detailed description, see Anderies, Mathias, and Janssen 2019). These types include: experience or knowledge of the past (K1, which may include indigenous knowledge, longitudinal data or “institutional” understanding of system capacities and weaknesses); knowledge about the interactions of specific environmental, economic and social features of urban systems (K2, typically drawn from the climate and social/behavioral sciences); knowledge about future events based on probabilistic characteristics of uncertainty (K3, i.e., downscaled climate modeling, which has traditionally been more difficult to generate with local specificity); and knowledge about the levels of use for specific services or resources (K4, which requires ongoing monitoring or self-reporting). Developing all these types of knowledge is advantageous because it can increase the probability that planners detect current and future risks and minimize surprises through developing response options (Walker et al. 2023). However, efforts to close gaps in these knowledge systems may tax the capacity of resource-constrained local governments and can introduce collaboration risks identified within the voluminous literature on collective action (Anderies 2015; Berardo and Scholz 2010; Lubell et al. 2017; Ostrom 1998, 2010; Yi and Cui 2019). As a consequence, the theoretical and empirical linkages between resilience threats, planning and organizational action remain underdeveloped (Deslatte, Krause, and Hawkins 2022; Friedman et al. 2024; Martín and McTarnaghan 2018).
To illustrate our theory, consider a scenario in which a mid-sized “constrained” city is experiencing impacts of climate change, which disproportionately affect historically disadvantaged communities, spatially clustered populations which the U.S. Environmental Protection Agency has labeled LIDACs, or Low-Income and Disadvantaged Communities (Holsman and Lucatello 2022; Hughes 2020; Meerow and Newell 2019). Policymakers and managers have decided to develop a green infrastructure plan which could help guide targeted investments in infrastructure and programs aimed at reducing flooding and instances of extreme heat in specific LIDAC neighborhoods (Albro 2019; Deslatte, Chung, and Stokan 2023; Meerow 2020).
Planners will need information about their own history of organizational capacity challenges (K1), due to deindustrialization, population stagnation, and/or geographic limits to growth and quality of life (see Hughes 2020, for a discussion of legacy cities). They will need knowledge about the ecosystem dynamics affecting LIDAC communities (K2), which can be overburdened with legacy pollution, degraded streets, sewers and other infrastructure, and have a relative lack of natural or green spaces which could mitigate some environmental problems (Meerow and Keith 2022; Woodruff et al. 2021). They will need some understanding of how climate change exacerbates these problems (K3), via shifting mean and extreme hydroclimatic events (e.g., increased precipitation, flooding, droughts). Lastly, they may need to inventory the conditions of existing roads, bridges, and stormwater drainage systems (K4), to better understand their future functionality and model how gray and green infrastructure interact during flooding events or heat waves (Kim and Kang 2023).
Any effort to improve the conditions in these LIDAC neighborhoods introduces considerable collective-action challenges which may truncate the information they collect (Lubell et al. 2017; Ostrom 1998; Yi and Cui 2019). Collective action challenges emerge from the inability of groups to work together to find optimal solutions to problems (Ostrom 2010). The quality of information feedback is central to resolving such dilemmas because it can help clarify the consequences of individual and group choices (Dunlop et al. 2022). In active inference, feedback is depicted as an iterative process of “predicting” the causes of sensory inputs about the environment and actively seeking new information when these prior beliefs diverge from observed (or expected) reality (Friston et al. 2017); these divergences draw attention to the (in)accuracy of our beliefs and motivate responsive evidence-gathering in what is called an exploration-exploitation tradeoff (Constant et al. 2022).
Extended to groups, exploratory evidence-gathering can provide epistemic gains which help resolve these discrepancies by facilitating experimentation and learning (Albarracin et al. 2022). Planners can make new discoveries and update their prior beliefs. But they also face information costs and limitations which tilt the effort more heavily toward exploiting existing knowledge or beliefs (Deslatte and Stokan 2020). These costs can include difficulty assessing how climate change patterns impact infrastructure over multi-decade planning horizons (Garcia et al. 2022). They may encounter bargaining costs between groups with different welfare goals and difficulties engaging impacted communities. For example, green infrastructure can mean different things to different groups, and has been justified and deployed opportunistically or to accomplish different goals, such as managing stormwater pollution, connecting ecosystem elements, reducing flooding, and improving property values (Hoover et al. 2023; Meerow 2020). Planners may also face coordination and monitoring challenges integrating goals across governmental departments or units (Hawkins, Krause, and Deslatte 2023). The fact that local governments are organized into specialized, siloed departments can create functional barriers to collaborating on goals which span traditional lanes of responsibility, such as planning, public works, and economic or community development (Krause and Hawkins 2021).
In the context of stormwater resilience, we expect that a more integrated inference process—reflecting a more diversified knowledge infrastructure system—will iteratively minimize the divergence between system goals and expected or realized outcomes over time. When the goals are longer-range and imbued with greater ambiguity, local governments will engage in more exploration to close knowledge gaps. When they are reacting to more recent events, they will be more exploitative of existing knowledge to minimize realized risks. When they manage only the latter, they are making tradeoffs between proactively planning for the future and reacting to present stressors. While the literatures on strategic management, organizational capabilities and collective action have all spoken to elements of these processes, active inference provides a potentially unifying theoretical motivation for why planning and adaptive management become more integrated over time: to minimize surprise, or resolve uncertainty, and sustain specific elements of system functionality. Local governments that set broader climate-related goals via existing planning processes—such as comprehensive planning—are more likely to iteratively enhance other analytic capacities and ultimately take broader adaptive actions because they are more likely to detect surprises. Table 1 details some existing planning and capacity-building efforts, the variables we use to measure them, their knowledge types and potential benefits.
Stormwater-related Actions, Knowledge Infrastructure Types, and Demonstrated Benefits.
In summary, local governments are making inferential tradeoffs when they integrate planning and policy implementation efforts. They may conduct broader engagement through comprehensive planning, conduct an inventory of already impacted gray and green infrastructure, or “mainstream” resilience efforts within other types of multi-hazard and climate mitigation planning (Roy et al. 2024). They may turn to several traditional analytic capacity-building approaches such as providing dedicated sustainability staffing and financial capacity. They could also examine and learn from the results of similar efforts in peer communities, via climate-focused regional compacts or collaborative networks (Deslatte, Siciliano, and Krause 2024). These examples all feature institutional designs establishing active feedback loops and balancing acts between exploring and exploiting evidence about current and future risks. Prioritizing short-term or current impacts over future potential ones may render a community less capable of assessing future scenarios and less resilient to climate change.
Data and Methods
Our analysis focuses on municipalities in the U.S. state of Indiana. As part of the Rust Belt, Indiana cities have experienced significant deindustrialization, environmental challenges and historical resistance to long-term planning, making it a “hard case” for studying climate resilience (Lindsey et al. 2017; McCabe et al. 2022; Widhalm et al. 2018). Indiana is expected to witness significantly higher average temperatures by mid-century with impacts to air quality and urban heat, along with increased flooding which will degrade infrastructure and worsen water pollution (Widhalm et al. 2018). Significant development has occurred within floodplains, increasing flooding exposure. Increased heat and freeze-thaw events are stressing roadways, bridges, and transportation systems, possibly necessitating changes to asphalt mixtures or using heat-tolerant street landscaping. Local governments will have to account for whether bridges, culverts, and gray and green stormwater infrastructure capacity is adequate (Purdue AgUrbanreport 2018). Desite these challenges, Indiana municipalities are not required to develop comprehensive land-use plans, and similar to other states, the legislature has routinely prohibited local governments from adopting novel or innovative policies (Boswell 2018; Goodman, Hatch, and McDonald 2021).
Our study uses a mixed methods approach (Mele and Belardinelli 2019). As depicted in Figure 1, the sequential, exploratory study design involved collecting planning documents, interviews, and survey data from 2020 to 2022. Our unit of analysis is cities and towns in Indiana with populations of 1,000 or greater in the 2020 Census (N = 305). The median municipal population in the sample frame (median = 3,268; mean = 15,760) is significantly smaller than in national studies which tend to feature minimum population cutoffs of 10,000 or 20,000. As such, this study presents a unique examination of resource-constrained communities which tend to be smaller, rural and more politically conservative.
Displays the sequential, exploratory mixed-methods research design, including qualitative and quantitative components.
Data
From 2020 to 2021, we collected 159 publicly available comprehensive plans by visiting city/town official government websites or contacting the public officials of each local government through emails and phone calls. As community-wide guidance for policies on land-use and development, comprehensive plans can impact development and regulatory decisions through zoning. Unlike some other states, Indiana does not require local governments to adopt a comprehensive plan but does specify that plans must contain objectives for future development and policies for land use and public spaces, utilities, buildings, or other assets. During the same period, we conducted 51 semi-structured interviews with managers and policymakers in 5 Indiana municipal governments who were involved in sustainability-related activities. These interviews were thematically coded using a two-cycle coding process to identify planning, implementation, and performance management activities (see Deslatte 2022, for a detailed discussion of the findings). The outputs were used to design a 129-item questionnaire capturing resilience-aimed efforts in planning, capacity-building, and evaluation in several resilience-related policy domains. Researchers called, emailed, and performed web searches to identify the most appropriate government official(s) to participate in the survey for each municipality (i.e., administrators who had the most knowledge of city planning and resource management practices). From February-June 2022, the questionnaire was sent online and through the mail, with a total of 174 cities and towns either completing or partially completing it (57% response rate). Due to partially incomplete responses to some survey items used in this analysis, the sample size drops to N = 151.
Outcome Variable: Resilient Stormwater Infrastructure
Our outcome measure captures the extent of efforts cities may undertake to improve flood resilience in their built infrastructure (Resilient Stormwater Infrastructure, or RSI). Infrastructure improvements can be a useful tool for responding to visible impacts of climate (increased flooding) but also to prepare for longer-term urban ecosystem challenges, high-heat, and concentrated disadvantages, depending on how and where they are deployed. Thus, they provide a potentially insightful example of urban adaptive management and the exploration-exploitation tradeoffs planners and policymakers make. Specifically, the survey asked 18 questions pertaining to whether a city/town had ever undertaken “actions related to resilience in the built environment” and “land use.” Our dependent variable was constructed from five measures which asked whether they had integrated green stormwater infrastructure (GSI) into governmental plans or processes; developed a stormwater or GSI plan; created zoning codes with stormwater retention requirements for new construction; implemented an ordinance requiring onsite stormwater management; and implemented a pilot incentive program for green infrastructure. Using these variables, we created a new ordinal dependent variable scaled from 0 to 4 based on how many progressive RSI actions the city had taken. The directionality of this scale was informed by our interviews and observations of the typical progress cities may make in adapting to increasing flood risks. Table 2 describes the scale and reports the percentage of respondents who had taken the actions.
Ordinal Measure of Resilient Stormwater Infrastructure (RSI).
Comprehensive Plan Quality
To quantify use of resilience principles in existing planning processes, we created a measure of plan quality using a dictionary-based, automated text analysis method to identify and dichotomously code 19 metrics over 6 plan evaluation principles used in the plan evaluation literature: (1) Goals, (2) Fact base, (3) Strategies, (4) Public Participation, (5) Inter-organizational Coordination, and (6) Implementation and Monitoring (Woodruff and Stults 2016). This let us assess the types of information, engagement and topics like climate adaptation/mitigation or resilience which were present in comprehensive plans (Deslatte, Chung, and Stokan 2023). Our approach involved creating a logical function to determine if keywords were used and described in a proper context and in accordance with each metric. The plans were coded “1” if the metric was present and “0” if absent. Reported in Table 3, the plan purpose (present in 94% of plans), public participation techniques (88%), and land use strategies (91%) are some of the most frequent metrics present, while strategies explicitly addressing green infrastructure (20%), energy conservation (17%), GHG reduction (9%), and water (8%) are more rare.
Plan Evaluation Metrics.
Additional Planning Efforts
We included five additional dichotomously coded survey items capturing whether cities had: identified roads or bridges more vulnerable to higher maximum temperatures or more freeze-thaw events (road temp); identified areas likely to be impacted by surface flooding (surface flooding); developed an inventory of impacted gray stormwater infrastructure (gray inventory); developed a green stormwater infrastructure inventory of investments such as rain gardens, bioswales, tree canopy, to assess climate impacts (green inventory); or developed written built environment planning strategies to decrease flooding (impact strategies).
Administrative Capacity
Research has shown that staff capacity is often a requisite resource commitment for making sustainability or resilience gains (Hawkins, Krause, and Deslatte 2023; Krause, Hawkins, and Park 2021). To capture this, one dichotomous item asked if the city had any full- or part-time paid staff dedicated to sustainability (sustainability staff).
Collaborative Capacity
Collaboration between governments on climate-related goals often depends on the integrative mechanisms—from informal networks to contracts and regional compacts—which reduce risks, build trust, and facilitate information sharing and dividing costs and benefits of collective action (Kim et al. 2022). Because collaborative efforts tend to begin with specific policy goals before expanding the goals of partnerships, we created an additive index (collaborative capacity) from four dichotomous items which asked about a range of collaboration mechanisms cities might use, including: working with other local governments, agencies, or a university institute to develop a GHG emissions inventory (GHG collaboration); joining a regional climate partnership (regional collaboration); and entering into a informal or formal bilateral agreement with another local government on climate or energy issues.
Fiscal Capacity
To capture fiscal capacity, we created an additive index (financial capacity) from seven five-category ordinal items measuring the frequency with which cities had: budgeted for new sustainability initiatives (finance budget); issued debt to finance sustainability (finance debt); applied for sustainability grants (finance grants); budgeted recurring funding for sustainability (finance recurring); offered tax/financial incentives for using carbon-reducing technologies (finance carbon incentive); offered tax/financial incentives for (re)developing green properties (finance redevelopment); or funded capital projects related to sustainability (finance capital). Each of these items was scaled 0–4, with “0” indicating they had never done so and “4” indicating they had done so “more than 10 times.”
Environmental Conditions
Finally, we compiled demographic and government data in order to approximate the levels of climate exposures and vulnerability we expect to find in constrained cities. We used the 2010 and 2020 U.S. Census and the 2017 U.S. Census of Governments to create: a town form of government measure (typically below 3,000 in population in Indiana, compared to a city with a mayor-council form of government); the percentage of the population which is Black; the percentage which are below the federal poverty level; and the percentage with four-year degrees or higher (education). We included the percentage of the municipality’s developed land area within a 100-year or 500-year floodplain in 2010, using flood hazard data from FEMA. To incorporate municipalities’ perspectives on climate change, we included a survey item on the level of climate concern reported relative to the local government leadership. The level of concern was scaled 0–4, with “0” indicating they are not concerned at all and “4” indicating they are extremely concerned. Descriptive statistics are reported in Table 4.
Descriptive Statistics for Combined Measures (N = 151).
Analytic Method
Here, the top right term in the numerator,
Bayesian inference is useful because it depicts how changing posterior beliefs (planners altering their view of the conditions of infrastructure in a neighborhood) depend on both the strength of prior beliefs and the likelihood of making an observation if their generative model of the world were true. In other words, posterior beliefs are proportional (in the reduced form of equation (1)) to the prior beliefs multiplied by the likelihood of new observations. Applied to our analysis, we estimate Bayesian ordered logit models using the “brms” package in R with weakly informative priors for all model parameters—an assumption that planners have weak priors about specific climate impacts and available responses. The R package uses what’s called a No-U-Turn Sampler (NUTS) method for Markov chain Monte Carlo (MCMC) stochastic simulation. MCMC is a class of algorithms that can overcome sample size limitations by drawing samples from a simulated probability distribution constructed from our data. Diagnostic plots indicate model convergence in a stationary distribution for all model parameters. We then use a Bayesian interval hypothesis test function to assess our theoretical expectations by estimating the posterior probability of a directional relationship between parameters of interest and our outcome. These findings are then compared with the thematic coding results from our interviews to assess the degree to which they converge or diverge with our theoretical expectations.
Results
We find constrained cities are more likely to integrate planning and resilient stormwater efforts when they have inventoried already experienced impacts, engaged in more expansive comprehensive planning, and dedicated some personnel to sustainability efforts. These efforts are largely reactive to experience (K1) and demonstrate gaps in other knowledge types which could aid in collective action. Consequently, cities are less likely to proactively prepare for future impacts, which is consistent with a core insight from the active inference framework—the tendency to exploit recent experiences or shocks to minimize uncertainty or surprise. The Bayesian model results are reported in Table 5, and include the posterior mean, credible intervals and hypothesis expectations for each parameter (that the parameter value lies above or below zero) along with the posterior probability for each test.
Bayesian Ordered Logistic Regression Results.
Interpreting the posterior estimates for planning, cities which develop built environment planning strategies (e.g., K2 and K4 knowledge on development in flood-prone areas) are 99% more likely to engage in RSI efforts. Cities which inventory “already impacted” gray stormwater infrastructure which needs modifications to handle heavier rain events (K1, K2) are 96% more likely to do so. Meanwhile, those developing an inventory of their green stormwater infrastructure networks to assess climate hazards (K1, K2) are 98% more likely to make greater RSI strides. The posterior mean coefficient for the GSI inventory is the largest (1.32), however only roughly 13% of respondent cities had made this planning effort.
While 70% of respondents have begun to identify areas likely to be impacted by surface flooding, we observe only a 56% chance—roughly a coin flip—that those cities engaged in more RSI efforts. We find a 78% chance—which we interpret as weak evidence—that cities identifying roads or bridges more vulnerable to higher temperatures or freeze-thaw events in the future make greater RSI strides. Lastly, we find a 97% chance that higher comprehensive plan “quality” makes RSI efforts more likely—although the posterior mean estimate is much smaller (0.1) than for the other planning efforts. Our text analysis found that public participation efforts were among the most frequently detailed “quality” metrics in plans, meaning detailed or inclusive public engagement likely plays a larger role in higher quality plans—and subsequently, in RSI actions, among cities with plans. This type of engagement tends to be more pro forma in land-use planning—and a public hearing is even required under state law. Nevertheless, cities with higher quality plans appear to take more RSI actions. Interview evidence corroborates these findings that evidence-exploiting behavior is more prevalent, given the difficulties acquiring and incorporating climate data (typically K3 knowledge types) into policy and programmatic decision-making. As one interviewee noted, “We must at some point [recognize] that climate change is going to cost dramatically more than it ever would if we just addressed the problem now.”
Turning to organizational capacities, we find a greater than 99% chance that cities with full-time or part-time staff dedicated to sustainability efforts (K1, K4) engage in more RSI actions. Typically, such staffing is highly fungible, meaning city leaders can direct them to focus on a variety of goals, but they also typically lack technical or scientific expertise (K2, K3) necessary to conduct or incorporate socio-environmental and climate drivers into programmatic decisions. The low percentage of cities and towns with dedicated sustainability staffing in our sample (11.3%) also likely means the management of climate impacts is being delegated to line departments which confront isolated problems rather than system-wide ones. Our interviews corroborate these results. Of the 14 interviewees who identified resilience-related coordination challenges within their organizations, shifting development and infrastructure investment patterns was a frequent concern given the long lifespans for built-infrastructure investments, backlogs in capital improvement programs, and difficulty securing new revenue streams. As one interviewee noted, “Retroactively trying to change or transform the built environment is very challenging from a planning perspective.” Internal, functional collective action across departmental silos was also identified as a challenge requiring inter-departmental coordination and monitoring. They noted transportation systems, as capital-intensive projects, are more difficult to transition for climate change, and require significantly more coordination across governmental units. “There’s this sort of political question of like, are the right folks communicating? Is this cross-cutting throughout the city’s organization? I have concerns about that,” one interviewee said.
One way cities compensate for a lack of expertise (K2, K3) or experience (K1) is through collaborations through which they can outsource some information costs. We find cities using more collaborative mechanisms—GHG inventory cohorts, regional and bilateral coordination efforts—are 88% more likely to pursue broader RSI efforts. Our interview data suggest many collaborations underway during the study period focused on climate-mitigation or sustainability efforts, such as reducing building energy use and increasing community solar installation. Nonetheless, interviewees noted the importance of participating in regional or university-sponsored climate-action “cohorts” of peer communities for exchanging information and identifying technical competencies and fiscal resources. “That is not naturally happening in other ways,” one interviewee said. However, interviewees noted that translating these efforts into sustained action was resource intensive, tended to be focused on specific projects, and was challenging to maintain over time. As one interviewee noted, “the inertia is real.”
We find cities devoting more financial resources to sustainability or carbon-reduction activities are actually 96% less likely to engage in broader RSI efforts. A possible reason for this is that cities experiencing the greatest surface flooding are less affluent and devote fewer resources to aforementioned energy and GHG-mitigation efforts. In other words, cities financing carbon reduction and sustainability efforts more broadly are less likely to be disadvantaged or have more flooding-related vulnerabilities. Because our survey item does not specify what types of sustainability initiatives cities may be funding, we cannot narrow down what knowledge gaps or barriers are implicated by this negative relationship. However, we find no evidence that heightened climate concern drives greater RSI investment.
Finally, we find that cities with greater poverty are 75% less likely to pursue greater RSI efforts, and those with more development in floodplains are 70% less likely. This finding is significant given that 66% of survey respondents indicated areas of their communities with poor or no stormwater drainage had experienced “significant flooding” in the prior five years. Cumulatively, this evidence conforms with the widely held belief relative to larger cities that more socially vulnerable populations are exposed to greater climate risks and less capable of preparing for them.
Discussion and Conclusion
Managing climate change in local governments is about managing risk-perceptions and aggregating them across an organizational level (Zhang, Welch, and Miao 2018). This study introduces the active inference framework (Friston et al. 2017) and applies it to a specific collective-action situation (resilience planning and implementation) as a potentially useful, unifying principle for understanding public organizational risk-management and change.
The intuitive appeal of this approach for scholarship is that many of the concerns over models of bounded rationality, altruism, self-interest, and cognitive biases can theoretically be subsumed by the principle of active inference, assuming that feedback systems can adequately detect prediction errors and allow for iterative belief-updating to minimize surprises. Because all self-organizing systems (including people and organizations) seek to occupy states which preserve their lifespans, functionality or performance, a host of hypotheses can be developed relative to how cognitive and structural characteristics of collectives can alter the balance of exploratory and exploitative information search and use –and thus, impact the long range resilience of communities.
A practical question then becomes what types of information rules or institutional arrangements can assist collectives develop more proactive and adaptive responses to ambiguous threats. While we find evidence that integrated feedback processes can advance efforts to address a specific, salient threat like flooding, the knowledge-based systems cities possess appear inadequate for addressing slow-moving climate risks with nonlinear characteristics. This speaks to the tendency to be more responsive to recent events and negative outcomes. However, it also speaks to the inherent ambiguity surrounding the climate crisis. Our evidence suggests there is a specific disconnect between the knowledge infrastructure systems governments develop and the needs across units of government. Few cities in our sample were gathering information which could help better mitigate future infrastructure failure and social vulnerabilities. Rather, they tended to rely on two types of knowledge—recent history or experience (K1) and service user feedback (K4). While the resilience literature notes the importance of these types of information, failure to model or account for the dynamics and inherent uncertainty of climate is likely to decrease resilience in the future.
Several potential strategies exist for addressing this mismatch. First, scholars and practitioners can begin identifying core organizational capabilities which can be reassigned to assessing physical and social vulnerabilities—shifting from exploiting existing information feedback to developing new ones. Our plan evaluation identified that only about 5% of local government comprehensive plans identified social vulnerabilities. However, compiling such information has been made easier through new tools such as the White House Climate and Economic Justice Screening Tool (CEJST) and the EPA’s Environmental Justice Screening and Mapping Tool (EJScreen). Scholars and practitioners can begin building these tools into research as well as existing planning and budgeting processes. While several communities in our study have opted to create new processes for developing climate plans and updating them periodically, our research has found that this is an avenue with low likelihood of success for communities without dedicated staffing (Deslatte, Chung, and Stokan 2023). As of 2024, only 14 Indiana cities out of the 305 in our sample frame (4.6%) had adopted a climate action plan. Incorporating adaptation goals and strategies into existing decision-making processes may be more efficient for smaller and constrained communities.
Finally, scholars and practitioners can enhance research and begin identifying opportunities for regional, cross sector or cohort-based collaborations with other motivated governments, nonprofits, firms, and universities. Even if the collaborations function merely as information-sharing networks, interview evidence suggests communities appear more willing to re-allocate resources and engage in more epistemic learning when they are aware of other peer communities doing the same.
Our study has several limitations. It is limited to one state although, we argue, likely generalizable to a large cross-stitch of smaller, understudied local governments across the country. Second, our exploratory design precluded testing specific causal mechanisms. A related caveat is that it was primarily conducted at the beginning of the implementation of the 2021 Infrastructure Investment and Jobs Act passed by Congress and prior to the passage of the 2022 Inflation Reduction Act. Both laws collectively devote more than $140 billion to local governments for climate-related activities over the next decade, but also emphasize regional coordination, public-private partnerships, or state-level involvement. Future, mixed method and longitudinal studies are needed to assess the causal impacts of these transformative opportunities. Finally, active inference is to date a cognitive framework which requires experimental or behavioral advancements to be fully extended to collective action dilemmas. Empirical and theoretical bridges must be built between the cognitive and collective levels of analysis. Yet, advancing our understanding of these exploration-exploitation tradeoffs could have profound importance in future years as local adaptation to climate change becomes an increasingly salient necessity.
Footnotes
Data Availability Statement
Data collected for this study is available from the authors upon request*.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the U.S. National Science Foundation (Award # 1941561).
