Model forecast temperature up to 6 degrees lower than measured
-
The ECMWF as well as other models forecast as much as 6 degreeC lower temperatures than measured. See below. Can you guys feedback to the model developers to get them to look into this?
Betong (in Sarawak, on the north side of the Borneo Island), on lower right in the maps, measures 33 degC but modelled 27degC.
Reported Temperatures:
Forecast Temperatures:
-
Hi @hangInThere thank you for your feedback with provided screenshots. This is really difficult to solve since providers of those models have their own metrics for their forecasts. Also, we have to realize that we compare real-time data from local stations, where we don't know the localization of the station and etc., with global models, which resolution is usually about 9km.
-
@Ondřej-Šutera
Is your team able to foreward this feedback to the ECMWF modelling team? Let them look into this. It's not just spot, but many spots between the Malay Peninsula and Sarawak that have 4 degC or more of difference with the model. This is a great opportunity for them to look into where their model can improve. -
Here's another day of huge disparity btwn model and actuality. The model forecast rain in most of the Malay Peninsula. In actuality, it's all dry except in Penang Island and Langkawi Island in the northwest, and a brief squall in Kuala Lumpur, around 5PM. There is some controlling factor that the ECMWF model doesn't get right. This is an opportunity to nail it.
-
It's been years I compare forecast model as a side hobby project. Most of them are often completely wrong. Those companies/public organisations are just selling data based on their madeOfNothing-reputation. They just measure the weather conditions from 1 spot, and then use algorithms to "guess" the weather in other regions, hoping people will trust them - and it work !
Also keep in mind that ECMWF model is mainly using weather sensor in Europe. There predictions are only update twice a day, which might be the reason of the wrong weather conditions in your region.
-
@hangInThere I am not sure, if this is possible. I would recommend checking other available models for your location, for example Australian ACCESS model.
-
@Ondřej-Šutera
ACCESS is just as bad as ECMWF, both forecasting 4degC lower on Penang Island (actual 34degC) at the northwest corner of Malay Peninsula.ECMWF manages to go one step worse, in the south at Kulai, to go 6 degC lower than actual (34 degC) whereas ACCESS is 2 degC.
It's not just a individual town that ECMWF got badly wrong, but an entire area of more than 100km inlands, 100km along the west coast, in the southern state of Johor.
This temperature forecast error may be due to ECMWF's error in forecasting HEAVY rain when in actuality it's DRY all day till 4PM.
Being a paying customer, Windy.com has the right to request ECMWF to look into flaws in their product, and Windy.com has the responsibility to its paying subscribers to forward this feedback to ECMWF to point them to what needs corrected.
Improvements come from knowing what don't work.
ECMWF, 4PM:
Reported Temps, 4PM:
ACCESS, 4PM:
ECMWF, 4PM, Johor state:
Actual, 4PM, Johor state:
ECMWF forecast HEAVY RAIN, 4PM, Johor state:
Weather radar image show DRY, 4PM, Johor state:
ACCESS forecast LIGHT RAIN, 4PM, Johor state;
-
@hangInThere I went through the map in Singapore, Malaysia, and Thailand. You are right about differences in real temperatures from stations and forecasted ones from global models. However, the difference is usually about 2 degrees and we still talk about the comparison of global model forecast vs. real-time data from stations, which can be located in specific conditions that can affect this.
I also discussed this with colleagues and there is no reason to hesitate about the performance of used models.