The Netherlands All editions · No. 1 ENNL

After The Forecast

The weather forecast, reviewed after the weather.

Published monthly by VeriSky

Verification report · 1 July 2026 to 31 July 2026

ECMWF IFS came closest on temperature. NOAA AIFS read the wind best.

11 forecast models, 2 Dutch weather stations and 840 common observation hours. This edition of After The Forecast narrows the evidence until every model faces the same independent test.

Edition No. 1 · The Netherlands · 9,240 forecast-hour comparisons
840
Common observed hours per model
11 models
Each tested on the same hours
2 stations
Rotterdam and Schiphol, complete days only

Across 35 complete station-days at Rotterdam and Schiphol, we compared every model one day ahead with the temperature and wind later measured by the station. ECMWF IFS had the smallest temperature error, 1.08 degrees. NOAA AIFS had the smallest wind error, 4.38 km/h.

That sample is deliberately smaller than the scorer's full Dutch archive. Locations backed by a weather-model analysis were excluded, as were station-days with even one missing hour. Rain is absent too: the current rain truth is a model analysis over the Netherlands, not an independent instrument record. A shorter honest table says more than a larger self-scored one.

July 2026 at the two stationsFig. 1 · one day ahead
Model Temperature score Typical error Wind score Typical error Hours
ECMWF IFS Europe
83.4
1.08 °C
62.2
5.29 km/h 840
ICON Germany
82.7
1.13 °C
66.0
4.75 km/h 840
ECMWF AIFS Europe, AI
81.9
1.18 °C
67.8
4.51 km/h 840
Arpege France
81.0
1.23 °C
68.4
4.42 km/h 840
NOAA AIFS United States, AI
79.3
1.34 °C
68.7
4.38 km/h 840
DMI Harmonie Denmark
76.7
1.51 °C
66.7
4.66 km/h 840
GFS United States
74.5
1.66 °C
66.8
4.65 km/h 840
UKMO United Kingdom
74.1
1.68 °C
64.9
4.92 km/h 840
KNMI Harmonie Netherlands
72.7
1.77 °C
66.5
4.69 km/h 840
JMA Japan
71.6
1.85 °C
65.6
4.82 km/h 840
CMA GRAPES China
71.5
1.85 °C
63.6
5.10 km/h 840
One-day-ahead forecasts on 840 common hours at Rotterdam and Schiphol. Every row uses exactly the same 35 complete station-days. Scores run from zero to 100; the measured root mean square error is printed beside each score.

KNMI Harmonie placed 9 of 11 on temperature and 6 on wind. That is a result for Rotterdam and Schiphol during July 2026, not a verdict on the whole Netherlands. High-resolution local models can gain ground away from airports or in a different weather regime.

Distance separates the models

The table above asks the easiest question: what did the model know one day before the observation? The chart keeps the same station sample and moves the forecast farther away. A line stops when its model no longer supplies at least 95 percent of the common hours.

Accuracy against distanceFig. 2 · temperature by days ahead
Temperature forecast score by days ahead Five model lines. Exact values and observation-hour counts are in the table below. 405060708090 day 123456 score ECMWF IFSICONECMWF AIFSNOAA AIFSGFS
View the chart values as a table
Temperature score and observation hours at each lead time
ModelDay 1Day 2Day 3Day 4Day 5Day 6
ECMWF IFS83.4 840 h81.2 840 h80.2 840 h75.0 840 h68.9 840 hNot available
ECMWF AIFS81.9 840 h81.6 840 h78.8 840 h73.9 840 h71.7 840 h66.4 840 h
ICON82.7 840 h79.8 840 h77.9 840 h68.8 840 hNot availableNot available
NOAA AIFS79.3 840 h77.1 840 h76.1 840 h70.5 840 h64.5 840 h64.4 840 h
GFS74.5 840 h72.4 816 h67.1 816 h59.3 840 hNot availableNot available
Temperature score at each lead time, on the same station-days as Fig. 1. Every visible point carries at least 95 percent of the 840-hour headline sample; its exact hour count is shown in the data table and interactive tooltip.

The sample, not just the score

A monthly number can look national while resting on a narrow set of instruments. These are the two station records that survived every gate. Publishing their dates and hours is part of the result, not administrative detail.

Where the evidence came fromFig. 3 · common station sample
StationDays HoursFirst included Last included
Rotterdam EHRD 16 384 7 July 2026 31 July 2026
Schiphol EHAM 19 456 4 July 2026 31 July 2026
Only station-days with all 24 temperature and wind observations enter the candidate pool. The final 35 station-days are the intersection also served completely by every ranked model. Partial station-days and analysis-backed locations are left out.
A forecast is a claim about the future. Somebody should check, afterwards, whether the claim was true.Why this report exists

What to take from this

At day 6, the best remaining line belongs to ECMWF AIFS at 66.4, 17.0 points below the best day-one score. Individual lines can flatten or tick upward from one day to the next, but the broad loss of information is clear. A forecast six days away is a different kind of claim from tomorrow's forecast.

Next month's edition covers August 2026 and publishes after the final station-day has arrived and passed the same checks. The sample can grow, but the rule does not change: no model is allowed to supply its own truth.

What was observed

Temperature and wind come from NOAA Aviation Weather Center METAR reports at Rotterdam and Schiphol. Raw reports are snapped to the nearest whole hour within 20 minutes. Only days with all 24 hours enter the sample.

Why rain is absent

The current rain accumulator uses Open-Meteo's best-match analysis, which over the Netherlands is KNMI Harmonie output. Ranking KNMI against that source would be self-scoring. Rain stays out until independently collected radar truth backs the published month.

Common hours

A model first needs at least 95 percent of the available station hours. The final table is then reduced to the exact intersection served completely by every qualifying model. That is why every row has 840 hours.

What qualifies

Commercial providers are omitted while their public-comparison terms are under review. VeriSky's blend is held out. MET Norway's seamless row is omitted because at day one in the Netherlands it is byte-identical to ECMWF IFS, not an independent forecast.

Limits

Two airport stations are not a country. This edition describes Rotterdam and Schiphol in July 2026. It does not measure inland minima, coastal waters or every Dutch weather regime, and one month is not a permanent verdict on any model.

Independence

We take raw model output from open national and international archives, compare it with independent station observations, and publish the result whether or not it flatters anyone. Data is at data.json, the full method at methodology and corrections at corrections.

The live version of this report is an app

After The Forecast comes out once a month. VeriSky keeps the rolling scores current for the places you care about and shows how models change with forecast distance.

Download on the App Store Get it on Google Play

Free standings · No account · No cookies