KINGDOM

Netherlands-wide climate simulator. Changing outcomes of the future.

Comment

My story

I started coding in a drive towards to create something unique. However, there was a boundary which I was unable to cross due to myself. I believed that only experts can code to solve problems. So I remained limited within my own circle that I developed. Once I came to highschool I understood that anyone can solve problems. I got interested in modeling the world, and did not know where to start. As a result, I started with the climate. I modeled the Netherlands and that is what had pulled me to this area.

My ask

I ask for feedback, ideas, and features. I want to improve upon this MVP to be used in policy decisions worldwide.

Elevator pitch

Kingdom is a Netherlands-wide climate flight simulator. Kingdom lets decision‑makers stress‑test policies and investments across water, energy, land use, infrastructure and health, before paying for expensive, specialist modelling. Through open Dutch data and calibrated response functions, we transform hazard maps into fast, transparent scenarios in minutes. For example, how does flood risk, drought, heat stress, emissions and economic exposure change if we rewet peat, redesign polders, electrify transport, or tighten building codes? Cities and water boards use it to run mandated climate‑risk dialogues; infrastructure owners (grid, ports, airports) use it to de‑risk multi‑billion‑euro assets; banks and insurers use it to explore physical‑risk pathways. We start with paid pilots, then annual subscriptions and support, and scale through partnerships with consultancies and high‑faithful tool providers (3Di) for targeted “escalation” studies.

Opportunity

Climate risk in the Netherlands is significantly different from a stable default baseline. Annual mean temperature has risen ~2.3°C since 1907, outpacing the global warming over the same period [1, 2]. KNMI reports that temperatures have increased by over 2°C since 1901, with the last three decades seeing a rise of ~1.1°C, more than twice the global rate [1, 2]. Total annual precipitation has increased by ~21% from 1906–2020 [3]. Thus, the space now requires multi-scenario stress testing. As a result, KNMI’s 2023 projections show all plausible futures bring continued sea‑level and temperature rise, wetter winters, and drier summers [4, 5]. Additionally, predicted by KNMI's 2023 projections, in 2086–2115, the average maximum precipitation deficit rises ~13–79% (≈181–286 mm) relative to 1991–2020 [4, 5]. Exposure is structural with ~26% of Dutch land below sea level, ~59% is flood‑prone, and ~70% of residents living in flood‑prone areas [6]. CBS estimates that in 2022 ~€525 billion of GDP (≈53% of the economy) was generated in flood‑prone areas (their extremely low risk class, ≤1:100,000 annual probability), up from ~€340 billion in 2010 [7]. The restriction point is analytics. There is a lack of expert‑driven, and trustworthy modeling because it is costly and therefore used selectively. For example, a professional 3Di license starts at €7,200/year (250 calculation hours), with unlimited hours +€6,200/year and “3Di Live” +€6,250/year, before model‑building or consultancy [8]. Research shows why emulation matters: surrogate approaches report ~15.9–52.2× speedups versus conventional 2D hydrodynamic solvers [9]. Even dissemination is selective: 3Di’s 2024 overview lists one scientific paper and two theses/internship reports [10]. Our MVP closes this gap as a “flight simulator." It is fast, and has low‑cost screening that determines when deeper specialized modeling is economically justified, used to make decisions quicker and solve issues faster. We expand this more on our 1-additional document [1.1 3.5, 1.2].

Validation & Progress

Validation for KINGDOM is staged around Dutch decision practice first, and ML accuracy second. We began by validating problem–output fit against the stress‑test and risk‑dialogue workflow defined in the Delta Programme for Spatial Adaptation, explicitly cyclic and enters its second national cycle from 2025, requiring repeatable stress tests and follow‑up implementation planning rather than one‑off studies [11]. To ensure institutional compatibility, we fixed the scenario backbone to KNMI’23, so every run is reproducible under nationally recognized wet/dry futures and aligned time horizons [4].

On technical progress, we have completed the full mathematical and systems specification of KINGDOM as a causal, multi‑fidelity decision investigator [E1–E53]. This includes formal intervention‑query semantics, traceable spatial and temporal aggregation, a structural causal model, and a value‑of‑information‑based escalation rule. We have locked the canonical discretisation for national screening [1 km grid, annual time step] and explicitly defined when higher resolution is invoked [250 m for pluvial screening, 100 m for urban heat, event‑time for hydrodynamic escalations]. The emulator architecture is fully specified and audit‑ready: a multi‑encoder, attention‑fusion, multi‑head model with explicit parameter counts [reference configuration ≈ 38.2 M parameters], heteroscedastic uncertainty outputs, and an additive multi‑fidelity coupling so that learned corrections remain bounded and interpretable on top of constrained screening logic.

All input layers are versioned, hashed, and QA‑checked before entering the feature store. Automated checks cover CRS alignment, unit consistency, temporal coverage, missingness, and spatial completeness. These checks are deterministic and logged, ensuring that every output can be reproduced from manifests.

Screening response functions [for example expected annual damage, people‑at‑risk, peat subsidence and emissions] are calibrated against reference layers and monitoring data where available. Emulator performance is evaluated on held‑out regions and regimes [coastal vs river, peat vs sand, dense urban vs rural]. Core quantitative metrics include:

-MAE_Δ on baseline‑versus‑intervention deltas, -Negative log‑likelihood (NLL) for probabilistic outputs, -Expected calibration error (ECE) and empirical coverage at 50 / 80 / 90 percent intervals, -p95 inference latency to ensure interactivity.

In the reference comparison, gated‑fusion screening achieves MAE_Δ ≈ 0.11 with sub‑second latency, attention‑fusion emulation improves accuracy to MAE_Δ ≈ 0.09 with ECE ≈ 0.024, and GP‑residual mode further tightens calibration [ECE ≈ 0.012] at higher computational cost.

For flood hydraulics, escalations explicitly target high‑fidelity 1D–2D engines such as 3Di which enable direct comparison of flood depths, arrival times, hazard ratings, and damage estimates in regimes where local hydraulics dominate outcomes [8][10].

Every output includes uncertainty bands and causal decomposition. Escalation is triggered automatically when uncertainty to thresholds exceeds predefined limits. Reliability diagrams, regime‑sliced error reports, and escalation logs ensure fast screening is used where appropriate and specialist tools only where the expected value of additional information is positive.

Market

Our primary customers are Dutch organisations that must plan, finance, insure, or operate long‑lived assets under rising flood, heat, and drought volatility driven by climate change [1][2]. The first core segment is the public climate and water domain: 342 municipalities and 21 waterschappen responsible for regional flood safety and freshwater management [15]. From 2025 onward, the Delta Programme for Spatial Adaptation enters its second cyclic phase of stress tests, risk dialogues, and implementation agendas, creating recurring demand for scenario‑based decision tools rather than one‑off studies [11].

What matters most to these buyers is avoiding expensive misallocation—selecting the right adaptation measures and timing under uncertainty, and demonstrating that decisions are evidence‑based. The scale of investment is substantial. The Deltafonds allocates long‑term funding toward 2050, while total Delta Programme costs remain higher, reinforcing the need for prioritisation and screening [12]. At the same time, the next expansion wave in infrastructure and utilities makes integrated climate‑risk screening equally critical: TenneT plans approximately €200 billion in grid investments by 2034 [13], and Gasunie plans roughly €12 billion in energy‑infrastructure investments by 2030, with a strong focus on sustainable gases [14].

Buyer, payer, and user roles often differ. Procurement decisions typically sit with programme owners and boards (municipal adaptation budgets, water‑board executives, utility CAPEX and risk committees, CRO/ESG functions), while daily users are planners, engineers, risk modellers, and consultancies. The ecosystem already provides strong data entry points, including the Kennisportaal Klimaatadaptatie as a national gateway [17], hazard and impact layers via the Klimaateffectatlas [5], and finance‑oriented tooling such as the Dutch Climate Risk Portal, launched in November 2025 [16]. KINGDOM’s differentiator is a Netherlands‑scale “flight‑simulator” layer that converts these inputs into fast, cross‑sector scenarios and explicitly identifies when it is economically justified to escalate into specialist models and detailed studies.

References

  1. Royal Netherlands Meteorological Institute (KNMI). De staat van ons klimaat 2022, 2022.
  2. Intergovernmental Panel on Climate Change. Ar6 working group i: Summary for policymakers. Technical report, IPCC, 2021.
  3. Royal Netherlands Meteorological Institute (KNMI). Klimaat in nederland: Neerslag, 2021.
  4. Royal Netherlands Meteorological Institute (KNMI). Knmi’23 klimaatscenario’s – gebruikersrapport. Technical report, KNMI, 2023.
  5. Climate Adaptation Services. Neerslagtekort – knmi’23 scenario-uitwerking, 2023.
  6. Government of the Netherlands. Facts and figures about water management in the netherlands, 2023.
  7. Statistics Netherlands (CBS). Impact of climate change on economic development: Size of the economy in flood-prone areas, 2025.
  8. 3Di Water Management. 3di pricing and licenses, 2024.
  9. Y. Liu, P. Bates, and J. Neal. Rapid flood inundation modelling using deep learning. Journal of Hydrology, 610:127866, 2022.
  10. 3Di Water Management. Scientific publications and theses, 2024.
  11. Climate Adaptation Netherlands. The delta programme for spatial adaptation is cyclic, 2024.
  12. Nationaal Deltaprogramma. Nationaal deltaprogramma 2024 — financial outlook. Technical report, Government of the Netherlands, 2024.
  13. Reuters. Dutch grid group tennet plans to invest about €200 billion by end of
  14. Reuters, 2025.
  15. Reuters. Dutch gasunie expects to invest around €12 billion through 2030. Reuters, 2025.
  16. Government of the Netherlands. Municipalities and water boards in the netherlands, 2025.
  17. Dutch Climate Risk Portal. Dutch climate risk portal launch, 2025.
  18. Kennisportaal Klimaatadaptatie. Kennisportaal klimaatadaptatie — entry point for adaptation data and tools, 2026.

Built With

Share this project:

Updates