What's changing
Europe does not lack climate data. Companies lack an operating layer.
Signal is cheap. The company's own model of itself is not.
Who this is for Operations and supply-chain leaders in industrial groups whose exposure runs through rivers, energy and suppliers.
28 August 2026
I started looking at climate resilience as a fairly conventional business problem. Europe is getting hotter, rivers are running lower, and industrial companies depend on water, energy, transport corridors and suppliers that increasingly behave outside their historical ranges. The commercial question seemed obvious: what can businesses do about it?
The deeper I looked, the less this appeared to be a climate-data problem. It looks like an execution problem.
The physical risk is already operational
Consider what happened in Europe this summer.
Romania shut down both reactors at the Cernavodă nuclear plant because the Danube no longer carried enough water to cool them. Unit 1 went offline in late July; the controlled shutdown of Unit 2 began on 13 August, after the government had already declared a nationwide state of alert in the energy sector for the month. Before that, the state allocated more than two million euros to emergency works aimed at keeping the plant operating — blasting a rock that was restricting flow and sinking four barges loaded with stone to redirect the river. A country intervened in the physical course of the Danube to protect a fifth of its electricity supply.
On the Rhine the consequences ran through industry instead of the grid. Barges could carry less. Freight shifted towards rail and road, but those alternatives could not fully offset the lost capacity. Covestro declared force majeure on polyether polyols from its Dormagen site, stating that the lost transport capacity could not be made up. BASF declared force majeure on deliveries of certain surfactants from its European assets, citing low Rhine levels and a substantial reduction in the availability of key raw materials. LyondellBasell declared force majeure on butadiene from its 170,000-tonne Wesseling unit, effective 16 July, after low water restricted feedstock deliveries. Across chemical producers, utilities, steelmakers and agricultural traders, companies reported higher costs, bottlenecks and, in some cases, reduced output.
These are not environmental problems in the abstract. A river level becomes a transport-capacity problem. Transport capacity becomes an inventory problem. Inventory becomes a production problem. Production becomes a margin problem.
The scale is not marginal. ECB work estimated that surface-water scarcity could put nearly 15% of euro-area output at risk under an extreme but plausible 25-year drought scenario. A subsequent ECB Occasional Paper put exposure at up to 24% under a 100-year event.
Companies know something is changing. The European Investment Bank’s 2025 survey found that 68% of European firms had already incurred costs from extreme climate-related events, while only 53% of EU firms had taken action on physical climate risks. That gap is more interesting than the weather.
The information is already available
My first assumption was that companies lack the information required to manage these risks. That assumption does not survive much investigation.
Europe has built an extensive environmental data infrastructure. Copernicus provides satellite observation, historical climate data, soil conditions, drought indicators and flood information. ECMWF provides forecasts and ensemble models. The European Drought Observatory tracks soil moisture, GloFAS models river discharge, EuRIS and national hydrological services publish actual river levels and navigation conditions, and ENTSO-E covers the electricity system. Commercial providers go further, aggregating weather models, calculating asset-level exposure and monitoring supply-chain disruption.
The external world is becoming observable. It is possible to know that rainfall is below normal, that the Rhine is falling, that drought and heat and flood conditions are forming — and increasingly, to forecast them.
So collecting another layer of climate data is unlikely to be the interesting business. That layer is already crowded.
Nor is Europe short of recommendations. As of 2025, all 32 EEA member countries have adopted national adaptation policies. The Commission has strategies for water resilience and climate adaptation, alongside an EU framework for critical-entity resilience. Companies have risk departments, business-continuity plans, supplier monitoring, operating procedures, insurers who model physical risk and consultants who produce resilience assessments. And yet the European Environment Agency still reports persistent gaps between planning and implementation, along with unclear risk ownership, uncertain financing, weak monitoring and difficult coordination across governance levels.
That is where the problem becomes interesting.
Knowing that the Rhine is falling is not a decision
Imagine a chemical manufacturer receives a warning that the Rhine may fall below a critical level in ten days. The information is useful. It does not answer any of the questions the company actually needs answered.
Which suppliers depend on the river?
Which materials do they supply?
Which shipments are already in transit?
How much inventory sits at each plant?
Which lines consume those materials?
Can another supplier replace them?
Can freight move by rail, and how much capacity is available?
Which customer orders depend on that production?
What margin is at risk, and when does somebody have to act?
The climate model cannot answer those questions. The answers sit inside the company — and rarely in one place. Part of the picture is in ERP. Part in inventory systems. Part in procurement spreadsheets. Part in logistics platforms. Part in email. Part in messaging channels. And an uncomfortable amount exists only in the heads of the people who know how the operation actually runs.
The missing model is the company itself
The useful system is not one that says Rhine level: high risk. It needs to hold something closer to a chain:
Rhine level → transport lane → Supplier A → Material B → Plant C → six days of inventory → Production Line D → Customer Orders E and F → the contribution margin those orders carry.
Only then does an environmental signal acquire operational meaning. And only then can the system move one step further, to: reserve rail capacity by Tuesday, raise safety stock on Material B, pull Batch 184 forward, ask Supplier G to confirm alternative capacity.
That is not climate analytics. It is an operating layer connected to climate analytics — and the distinction is the whole argument.
Why the ERP does not already do this
The obvious objection is that this model already exists. Supplier, material, plant, order, margin: that is precisely what an ERP holds, and what supply-chain planning suites are sold to do. Enterprises have spent decades and considerable money building exactly this representation.
The objection is fair, and it is where the real problem shows itself.
What those systems hold is the documented company. Master data often describes the supplier relationship as it was contracted, not as it is currently being worked around. The bill of materials describes the approved substitution, not the one quality actually accepts under time pressure. Lead times reflect the agreed terms, not the fact that this particular supplier has been running four days late since spring, which everyone in procurement knows and nobody has recorded.
The gap between those two versions is where operations live. It shows up as the exception: the case the system could not close, which someone resolved by hand, over a call, in a thread, and closed without leaving anything behind. The written process becomes a fiction, and the real one runs in a shadow loop that no system observes.
A dashboard cannot cross that gap either. Identifying exposure and executing a response are different functions. A risk platform can report that an asset is drought-exposed or a supplier flood-affected. Neither knows that Supplier A provides three materials of which one is critical, that it feeds two plants, that one holds 28 days of stock and the other four, that the second can substitute the material but only after quality approval, that approval takes three days and the alternative supplier needs five.
This is where generic risk intelligence reaches its boundary. The next layer has to understand the company.
How that layer gets built
It cannot be built reliably from interviews alone. Ask an organisation to describe its own process and it will usually describe the documented one, in good faith. The model has to be assembled from work as it happens — from the transactions, the messages, the approvals and the corrections — and resolved against a common set of business objects: suppliers, materials, assets, orders, commitments.
The mechanism that keeps it current is the exception. Whatever the layer cannot close is routed to a human, and the human’s answer returns as precedent the system reuses. Coverage grows exactly where the system was weakest, and it grows from the work rather than from a documentation project that is stale before it ships.
Autonomy is then earned rather than assumed. Work moves to an agent only when its outcome is objectively verifiable, its actions are reversible or contained, and there is enough precedent behind it. The human decides first, then confirms, then audits a sample. The switch stays with a person.
The sequence is simple to state and hard to build:
Signal → dependency → consequence → decision → action → evidence.
The difficult part is not the signal. It is everything after it.
Climate is making the problem easier to see
For a climate-resilience product, almost none of the external data infrastructure is worth building. Buy the weather models. Use Copernicus, public hydrological data, existing flood and drought models, and specialised risk intelligence where it beats public sources. The proprietary layer starts inside the company, and it starts with dependencies, consequences and owned actions.
But the architecture problem is not a climate problem. Companies now have access to extraordinary volumes of external intelligence — weather, logistics, energy, commodity prices, supplier risk, sanctions, political events. The constraint is no longer observation. It is the absence of an internal operating layer capable of interpreting those signals against the current state of the business.
Without it, every new agent has to reconstruct the company from fragments. With it, the climate agent becomes one application, the procurement agent another, the inventory agent another. The expensive part stops being each agent and becomes the environment in which any of them can work.
Climate resilience is a useful place to see this because the causal chain is unusually short and visible. The river falls, the barge carries less, the material does not arrive, the factory slows, revenue disappears. Elsewhere the same chain runs through the same missing layer, only slower and with less obvious attribution.
Climate change did not create that problem. It is making the cost of it easier to measure.
Sources. Cernavodă shutdown, the state of alert and the emergency river works: PBS, Al Jazeera. Rhine disruption and the Covestro declaration: Reuters, via The Globe and Mail and C&EN. BASF and LyondellBasell declarations: Platts / Chemical Week, with the LyondellBasell declaration first reported by ICIS. Euro-area output at risk: ECB, The ECB Blog, May 2025 and ECB Occasional Paper 380. Firm-level exposure and adaptation: EIB Investment Survey 2025. National adaptation policies and implementation gaps: EEA, Climate resilience in Europe, 2025 and the accompanying EEA press release. Data infrastructure: Copernicus, ECMWF, the European Drought Observatory, GloFAS, EuRIS and ENTSO-E. The supplier and inventory figures used to illustrate a dependency chain are examples, not observed data from a specific company.