The IWEPS Foresight Paper no. 12, published in August 2026 under the title Measuring and governing climate risks: what is at stake?, marks an interesting development in the way adaptation is approached. Climate risks are described there as complex social-ecological phenomena, produced by the interactions between hazards, territories, populations, institutions and environmental transformations. The document stresses threshold effects, tipping points, cascades, long timescales and the deep uncertainties that make these risks hard to treat through a purely statistical logic.
This approach fits the reality of territories better than analyses centred on the intensity of a hazard taken in isolation. A prolonged drought alters water reserves, weakens certain ecosystems, affects agriculture and can heighten tensions between competing uses of water. A heatwave arrives in an environment whose soils, infrastructure, health services and populations are already in a particular state. Several successive perturbations can reduce the capacity to recover before a critical threshold becomes clearly visible.
The paper largely identifies this dynamic. Its operational instruments nevertheless remain far more static than the theoretical framework it defends. This difference deserves particular attention, because it bears directly on the capacity of public action to follow a trajectory that is deteriorating, rather than to note after the fact that a territory was vulnerable.
Systemic thinking still translated into static tools
The first chapter represents social-ecological risk as a relation between hazard, exposure, social vulnerability and institutional vulnerability:
Social-ecological risk = Hazard × Exposure × (Social vulnerability + Institutional vulnerability)
The authors then specify that this representation remains illustrative and that they are not seeking to develop a genuine quantitative equation, before referring a few lines later to the « result of this equation ». The difficulty goes beyond this inconsistency of formulation. A multiplication suggests relatively stable relations between components whose behaviour can vary strongly according to context, to the recent history of the territory and to the proximity of a threshold.
The same gap appears in the Walloon indicator of social vulnerability. Twenty-two variables are normalised and then added without weighting, on the explicit assumption that they carry equivalent importance. The tool offers a legible representation of territorial differences, but several variables may measure closely related dimensions, while their real influence depends on the risk under study. Age, access to care, income, tenancy status and social isolation probably do not contribute identically to every climatic situation.
A sensitivity analysis would already show whether the results withstand a change of weighting or the removal of certain variables. A map that stays broadly stable under several hypotheses carries more decisional weight than a ranking that changes profoundly as soon as the method is altered.
Temporality adds a further fragility. Some of the variables used still rest on the 2011 Census while others date from 2022 or 2024. The authors themselves note that the 2021 Census data would improve the present representation. A territory nevertheless changes considerably over fifteen years. A map built from data widely spaced in time ends up superimposing several successive states of the same system.
The classification into five levels of vulnerability likewise calls for cautious reading. The categories rest on quintiles and correspond to a relative position among Walloon districts, without any absolute threshold defining an objectively « very high » vulnerability. When the paper states that 62% of the population lives in districts classed as of high or very high vulnerability, it is describing the spatial distribution of its indicator, not the proportion of inhabitants who have individually crossed a critical threshold.
These limits do not condemn the approach proposed by IWEPS. They call rather for a change of logic: keeping the maps and indicators as diagnostic tools, then connecting them to a system able to track a few genuinely decisive variables, to test several possible trajectories and to modify the action as conditions evolve.
Sentinel variables connected to decisions
The move toward a dynamic system could begin modestly. Adding ever more indicators would reproduce the original problem in another form. A lighter architecture would select a few sentinel variables for each risk or system studied, with an explicit procedure for choosing and revising them.
This selection could combine several criteria. Causal analysis would identify the variables located at sensitive points of the system and able to transmit a perturbation to several other components. Statistical analysis would test their influence on existing diagnoses. The quality and frequency of the data would allow indicators impossible to track properly to be set aside. Field knowledge would finally bring in signals that institutional databases capture poorly or too late.
This last dimension connects directly with the IWEPS paper, which argues for integrating experiential expertise alongside scientific knowledge, while preserving requirements of adversarial debate, intersubjectivity and refutability. The choice of sentinel variables could thus be made by groups bringing together scientists, managers and local actors, then re-examined after each significant event or stress-test exercise.
Choosing what we measure remains a political act as much as a technical one. A variable set aside can keep a vulnerability in the blind spot, while a favoured variable can attract greater resources. The composition of the monitoring core should for this reason be transparent, justified and revisable.
The distinction between technical threshold and decisional threshold then becomes essential. Take a forestry example. An abnormally high canopy temperature, combined with a significant water deficit, may constitute a technical signal indicating severe vegetation stress. The decision to forbid certain uses of water, to activate a protective measure or to mobilise a reserve from that level onward is, by contrast, a decisional threshold. The first rests mainly on knowledge of how the system works; the second involves trade-offs between risks, available resources, costs and consequences for the various users.
This distinction avoids presenting as purely scientific a decision that in reality carries a political dimension. Technical thresholds can be documented and revised by the competent disciplines. The rules of intervention associated with those thresholds must be defined publicly when their effects bear directly on populations, economic activities or the distribution of a resource.
The system could thus operate with a limited core of sentinels, completed by more detailed layers of analysis when several signals converge. A variable that has become uninformative could be withdrawn. A hitherto neglected signal could enter the system after a crisis or a stress test. Monitoring would itself become capable of evolving.
Tracking trajectories and confronting scenarios
Data do not all evolve at the same rhythm. A demographic structure can be reassessed periodically, whereas the state of an aquifer, a watercourse, heavily stressed vegetation or an electricity network can change rapidly. Each variable should carry information about its age, its quality, its uncertainty and its relevant update frequency.
Decision follows yet another temporality. A heatwave may call for daily adjustments while a planning policy is built over several decades. One of the challenges of adaptive governance is to connect these rhythms without letting permanent urgency efface deeper transformations.
The paper already argues for policies able to combine a long-term strategic vision with immediate action, with decision points allowing trajectories to be redirected as circumstances change. A concrete translation could combine continuous monitoring of the fastest variables, an annual review of trajectories and scenarios, and more widely spaced stress tests intended to test the territory's capacities under severe conditions.
Monitoring alone does not, however, resolve deep uncertainty. A system detects above all what it was designed to observe. An absent variable may become decisive, a relation assumed stable may change, and a rupture may occur outside the trajectories envisaged.
Scenarios must then complete the monitoring. The paper describes scenario-building as a progression between an initial situation, several conditional pathways and different future situations. It also shows how these scenarios can reveal cascading consequences and help identify strategies of avoidance or adaptation, notably in the face of water stress.
A territorial application does not need to produce dozens of futures. A few contrasting scenarios may suffice: a central trajectory, several plausible developments and at least one severe situation intended to test the limits of the system. Their function is to check whether policies and investments remain acceptable under different conditions, and then to observe whether the real data progressively bring the territory closer to one of those trajectories.
Stress tests allow this logic to be pushed further. A city can examine the consequences of a prolonged heatwave combined with a power failure and hospital saturation. A catchment can test several years of rainfall deficit. A forest massif can be confronted with a succession of high temperatures, water deficit and sustained winds. The value of these exercises lies in the dependencies they reveal: which capacities become insufficient, which services depend on one another, which decisions ought to be prepared before the situation becomes critical?
The results then feed back into the monitoring system. An unexpected vulnerability may bring a new sentinel variable to light. An assumption that has become unrealistic can be withdrawn. A policy whose robustness depends on too narrow a scenario can be revised.
A polycentric governance resting on existing institutions
Climate risks rarely follow administrative boundaries. The paper stresses that their consequences can propagate well beyond the place where the hazard appears, and that a relevant analysis must sometimes articulate biophysical, social and administrative territories.
Water follows a catchment. A fire depends on vegetation continuity, relief, water status, winds and access. Urban heatwaves follow the structure of the built environment, mineral surfaces, vegetation and air circulation. Health and electricity networks have geographies of their own.
A dynamic architecture should be able to work at these different scales without creating a new body charged with centralising everything. The paper already acknowledges the fragmentation between adaptation, civil protection, security, social protection and transition, with different instruments and time horizons. Adding a further structure without clarifying responsibilities would risk reinforcing the problem.
A polycentric organisation offers a more coherent route. Local actors would keep their capacity for observation, participation and experimentation. The regional level could provide the data infrastructures, the common methods and part of the expertise. The federal level would coordinate the risks and networks that exceed the Regions and would guarantee general interoperability. The European scale would remain essential for certain data, certain funding and phenomena that far exceed national borders.
Coordination would rest on common protocols, responsibilities established before crises, and compatible data formats. Heavy capacities, such as computing infrastructure, certain models or the production of scenarios, could be pooled. Territories would keep the freedom to add their own indicators where their characteristics require it.
Funding should follow the same logic. Shared infrastructures require stable financing rather than a succession of temporary projects. Territorial interventions would meanwhile benefit from incorporating mechanisms of solidarity, since the most exposed or most vulnerable authorities are not necessarily those with the greatest budgetary capacity.
The paper insists precisely on the differences in resources, adaptive capacities and exposure between territories and populations. A decentralised governance without equalisation would rapidly produce a multi-speed adaptation. Resources should therefore be distributed according to vulnerability and available capacity, rather than according to local ability to co-finance measures alone.
Local participation completes this architecture. It becomes particularly useful when choices carry distributive consequences, when territorial knowledge is missing from the models, or when triggering rules directly affect uses. Participatory mapping, networks of observers, territorial committees or citizens' juries can then bring situated expertise without seeking to replace scientific measurement.
Connecting each item of data to an action and each action to an evaluation
The principal risk of a sophisticated monitoring system remains bureaucratisation. Maps, indicators, dashboards and reports can multiply without genuinely reducing the vulnerability of a territory.
One simple discipline would limit this drift: each piece of information tracked should have an identifiable decisional function. A variable may serve to trigger an in-depth analysis, modify a scenario, activate an intervention, re-examine an investment or evaluate a policy. A costly item of data that contributes to none of these functions over several cycles deserves re-examination.
This rule should apply to the arrangements themselves as well. A report regularly produced but never used can be simplified or dropped. An indicator that does not distinguish a real improvement from a statistical variation loses its interest. A monitoring meeting from which no decision follows must be open to rethinking.
Policy evaluation closes this loop. A measure tested on a limited territory should have from the outset observable objectives, a baseline situation, an evaluation period and conditions allowing its adaptation or its abandonment. The results then feed back into the scenarios and the variables tracked.
This capacity to recognise an error is probably one of the most important dimensions of adaptive governance. An institution that knows only how to add new policies ends up accumulating arrangements without genuinely learning.
Conclusion: five developments to make foresight genuinely operational
Le IWEPS Foresight Paper no. 12 already supplies much of the necessary intellectual framework: a systemic approach, cascading effects, tipping points, foresight, territorialisation, participation and adaptive governance. Its principal gap appears between this dynamic reading and instruments still largely built as snapshots.
An operational development could rest on five principles:
Select a small number of sentinel variables, chosen on the basis of causal relations, sensitivity tests, data quality and territorial knowledge, then revise them when experience reveals new signals.
Combine monitoring and scenarios, so as to follow observed trajectories while regularly testing plausible or severe futures through territorial stress tests.
Separate technical thresholds from political decisions, letting expertise characterise the physical or biological limits while submitting the rules of intervention and the margins of caution to a transparent governance.
Organise a polycentric governance, in which the different levels of government share data, methods and responsibilities while retaining a capacity for action suited to functional territories.
Continuously evaluate the usefulness of data and policies, with the possibility of removing a useless indicator, modifying a strategy or abandoning an experiment that does not produce the expected effects.
This architecture would remain imperfect and would not make the future predictable. It would, however, bring the instruments closer to the systemic logic the paper already defends. The maps would continue to show where certain vulnerabilities lie, while monitoring, scenarios and evaluation would help us understand how they evolve and at what moment a strategy begins to lose its effectiveness.
The capacity to adapt then depends less on the quantity of information accumulated than on the speed with which a society manages to turn relevant information into a decision, to measure the effect of that decision, and to change trajectory when conditions require it.