RWE Interview Cheat Sheet — March 24, 2026¶

1. Motivation — Why this role? Why you?¶

• Nearly a decade in operational atmospheric science
• You run NWP systems end-to-end: WRF, MPAS, variable-mesh model
• You build automated pipelines: data ingest from ECMWF/GFS → model run → ensemble output → verification dashboards
• What's missing: your forecasts have never directly driven financial decisions — you want to be where they do
• Why RWE:
• One of Europe's largest renewable operators — weather IS the core asset
• Multi-commodity (power, gas, coal, oil, LNG, freight, CO2) — broad, global meteorology
• Team of 9 meteorologists — peer challenge and collaborative forecasting
• Investing in AI weather models (HPE partnership) — aligns with your AI-assisted workflows
• What you bring: you don't just consume model output, you build the systems. Full-stack: pipeline → model → verification → interpretation

"I've spent nearly a decade building weather forecasting systems. Now I want to be where those forecasts directly drive decisions."

2. Teleconnections — What influences European weather?¶

The key principle: In Europe, westerlies = mild + windy. Anything that blocks the westerlies = cold + calm.

Four weather regimes (the actual patterns over Europe right now): - NAO+ (~32% of winter): strong westerlies, mild, windy — good for energy - NAO- (~21%): high over Greenland blocks jet, cold, calm — bad - Scandinavian Blocking (~27%): high over Scandinavia blocks jet, cold easterly flow from Russia — worst for energy (Dunkelflaute) - Atlantic Ridge (~20%): high in mid-Atlantic, northerly flow, cool — moderate

Teleconnections (remote signals that shift the probability of which regime occurs): - NAO: the pressure difference between Iceland Low and Azores High — controls jet stream strength - SSW (Sudden Stratospheric Warming): polar vortex collapses → signal propagates down over 2-6 weeks → often leads to blocking/NAO- - ENSO: El Niño late winter favours NAO- (cold Europe); early winter favours NAO+ (mild). Opposite effects flip in January - MJO phases 6-7: best tool for 2-4 week forecasting of cold outbreaks in Europe - QBO: easterly phase weakens polar vortex → more SSW events → more cold winters. Predictable over a year ahead

Use this winter as the example: - Easterly QBO background → polar vortex already vulnerable - SSW in early January → NAO flipped from positive to negative → blocking - Second SSW late January → sustained blocking into February - The teleconnection chain gave 2-3 weeks of lead time on the cold

3. Renewable Energy — New challenges for meteorology?¶

Point 1: Weather now controls both supply AND demand - Before: supply was coal/gas/nuclear (human controlled), weather only affected demand - Now: massive wind and solar → weather controls supply too - Compound event: same blocking pattern → no wind (supply down) + cold (demand up) simultaneously - Dunkelflaute is the extreme case — forecast can be perfectly correct, system just can't cope - Other direction too: too much wind → negative prices (Germany had hundreds of hours of negative prices in 2025)

Point 2: Forecast errors are amplified - Wind power ∝ v³ — small wind speed error → large power error - Cut-out threshold (~25 m/s) is especially dangerous — a couple of m/s error means "full power" vs "zero" - Storm Eowyn (Jan 2025): prices swung dramatically same day because winds crossed cut-out

Point 3: Spatial resolution - Offshore wind: wake effects persist tens of km in stable marine boundary layer - Onshore wind: terrain effects - Solar: clouds cut output rapidly, need high-resolution cloud forecasting - Temperature: for regional demand forecasting - "This is close to what I do with WRF and variable-mesh downscaling"

Point 4 (if asked): Merit order change - Renewables have zero marginal cost → power price now set by residual load (demand minus renewables) - Residual load is entirely weather-dependent

4. Multitasking — Three things at once, who first?¶

First — trader at your shoulder (30 seconds) - They left their desk to find you — give a quick direct answer - "Germany windier Thursday, above consensus. Detail in 10 minutes."

Second — forecast contribution for colleagues - Team dependency — others can't finish without your piece - The team forecast serves the whole trading floor

Third — the briefing presentation - Fixed time slot — you know your runway - By the time you've done the forecast contribution, the briefing analysis is mostly done

Principle: Serve the person waiting NOW → team bottleneck → scheduled event. Be decisive.

5. Describe This Past Winter (2025/26)¶

December 2025: NAO positive - Strong westerlies, mild, windy - Good wind generation, low heating demand, soft energy prices

January 2026 first half: SSW → blocking - SSW weakened polar vortex → Scandinavian block built - Cold everywhere including UK (-12.5°C in Norfolk, heavy snow Scotland) - Continental Europe: coldest January in about 15 years - Gas demand surged, wind power dropped, prices rose

January 2026 second half: jet trapped over UK - Jet stream strengthened but couldn't pass the block - UK: storms arrived (Goretti, Ingrid, Chandra), record rainfall, wind power surged (four-year high) - Continental Europe: STILL cold and calm — block hadn't moved - Same pattern, opposite outcomes in different countries

February 2026: second SSW - Polar vortex collapsed → blocking sustained into February - UK: rain continued (Cornwall: 55 consecutive wet days) - France: Storm Nils, record flooding, 1 million homes lost power - EU gas storage dropped to ~40% — lowest in several years

Previous winter (2024/25) for comparison: - Mostly mild, NAO positive - Big event: Storm Eowyn (Jan 2025) — bomb cyclone, 939 hPa, record winds in Ireland, 1 million lost power - Eowyn showed the cut-out problem: winds too strong → turbines shut down → price spike - November 2024 Dunkelflaute: German prices hit ~€820/MWh

Key line: "Europe is not one market — the same weather regime creates different energy outcomes in different regions."

6. Forecast Failure — When you were wrong¶

• All models (ECMWF, GFS, WRF, MPAS) underestimated a heavy rainfall event
• Ensemble probability of exceeding heavy rain threshold was low at all lead times
• Insurance client had vehicle flooding losses
• Why it failed: mesoscale convective system enhanced by topography and moisture convergence — convective parameterisation couldn't resolve it
• Communication: technical post-mortem within 24 hours, presented to client portfolio managers, transparent about the limitation
• What you changed: established a checklist of synoptic precursors (high CAPE, strong low-level convergence, high precipitable water) to trigger a high-uncertainty flag even when model QPF is low. Protocol caught a similar event months later

"They didn't expect perfection — they expected us to quantify uncertainty honestly and improve systematically."

7. Team vs Individual¶

• Currently own the full pipeline independently — you make forecast decisions and take responsibility
• But you actively seek collaboration for new requirements — kickoffs with engineering team
• What you want at RWE: a team of meteorologists to challenge your reasoning
• Peer challenge prevents groupthink and catches mistakes — you once missed a rainfall event that a colleague might have flagged
• You thrive when people push back on your thinking

8. Technical¶

• NetCDF/GRIB: xarray, cfgrib backend for ECMWF GRIB, CDO/NCO for batch processing
• Plotting: matplotlib, cartopy for maps
• Additional: pandas, scikit-learn, dashboard building
• Verification metrics: bias, RMSE, ACC — and importantly, verify by weather regime not just overall

9. Questions to Ask Them¶

About the team and daily work: 1. "How does the team divide responsibility — by commodity, by region, or by timescale?" 2. "How is the Weather Analysis team involved in RWE's AI weather model investment with HPE?" 3. "With RWE operating a large wind portfolio, how much focus is on asset-specific forecasting versus market-wide analysis for trading?"

About how weather drives trading (also useful for your own learning): 4. "When the 06Z ECMWF comes in and shows a significant change, how quickly does the market typically react? Minutes? Hours?" 5. "Which forecast changes move the market most — temperature, wind, or something else? And is it different for gas versus power?" 6. "How do you assess whether a model revision is a genuine signal versus noise? Systematic methods or experience-based?" 7. "For sub-seasonal timescales — weeks 2 to 4 — do traders actually position based on those forecasts, or is it mainly the short-range that drives decisions?"

Pick 2-3 total. Don't ask all of them.

10. If They Ask About AI Weather Models¶

• AIFS (ECMWF's own), GraphCast (DeepMind), Pangu-Weather (Huawei)
• Advantage: run on a single GPU in minutes, can generate bespoke ensembles
• Limitation: can produce physically impossible states, lack physical consistency
• Best approach: hybrid — NWP for physical grounding, ML for speed, meteorologist to interpret

11. If They Ask How You Add Value Beyond Models¶

• Model awareness: knowing each model's systematic biases in different regimes
• Cross-model interpretation: investigating why the outlier model disagrees — sometimes it's right
• Anticipation: predicting what the next forecast run will show before the market sees it — this is the edge

12. Current Situation (check morning of interview)¶

• NAO: currently positive — westerlies running, systems moving through
• UK this week: cold front passing through, colder and showery, then high pressure rebuilding end of week
• EU gas storage: very low (~29%), lowest since 2022 crisis. Refilling before next winter is the big concern
• Geopolitics: Iran situation adding energy price volatility
• Renewables milestone: wind + solar reached 30% of EU electricity in 2025, surpassing fossil fuels for first time

Key Vocabulary¶

Know About RWE¶

• Weather Analysis team: 9 meteorologists led by Nils Kaiser
• Locations: London (Threadneedle Street) and Essen (Germany)
• Trades: power, gas, coal, oil, LNG, freight, CO2
• Coverage: Europe, Asia, US — intraday to seasonal timescales
• Owns ~5 GW wind in UK — both asset owner AND trader
• AI investment: HPE partnership (Nov 2024), AI Research Lab in Bellevue, Washington
• They monitor: Brazilian hydro, Danube/Rhine river levels, US soil moisture, tropical cyclones, drought

If You Don't Know Something¶

"I haven't worked with that directly, but here's how I'd approach it..."

Never say "not much" or "I don't know." Show the reasoning, not the gap.