Comment by aleciffo
Comment by aleciffo 3 days ago
Does anyone know what are we talking about in practice in terms of weather forecast prediction improvement? Like MAE/RMSE
Comment by aleciffo 3 days ago
Does anyone know what are we talking about in practice in terms of weather forecast prediction improvement? Like MAE/RMSE
They also feature that the IR hyperspectral measurement is new -- 1700 channels in IR for a telescope in GEO seems new to me, but I'm not sure what exists now in this space.
They say they hope to retrieve trace gases at that global scale (seemingly with 30 minute cadence), which I think would be new. Also, they seem to say that this spectral resolution would enable them to retrieve temperature and humidity as a function of height -- not just surface temperature and column-integrated water content ("humidity").
Aha, here's a nice link (https://www.ssec.wisc.edu/geo-ir-sounder/) on exactly this question, pointing out the NASA IR sounders that have existed for many years (AIRS). These instruments get vertically-resolved atmospheric information, but they are not at GEO so their coverage is different. This makes them less useful for NWP.
The difference in wind speeds can be quite significant, which greatly alters some forecasts.
The question is often when it will rain, not if it will rain.
I was curious but it’s surprisingly hard to find info. These guys [1] are pretty stoked about “nowcasting”—which seems to be on sub-10-minute timescales to issue local severe weather warnings and such. It appears current sounders don’t scan as often.
This project ppt from 2011 [2] references different requirements for different areas/teams and shows the instrument spits out readings at 150 Mbit/s, which seems like a good clip. Overall it sounds like a lot of local knowledge is involved in turning this output into forecasts. Maybe there’s not a precise answer to your question.
Somebody else must know more.
[1]: https://www.eumetsat.int/features/think-global-act-local
[2]: https://www.researchgate.net/profile/Donny-Aminou/publicatio...
Replying very late, but with an actual answer.
It turns out that yes, better forecasts is a large part of what motivated the launch of this instrument.
High-spectral-resolution IR spectra at GEO allow estimation of vertically-resolved temperature and water vapor (over large spatial areas, at high temporal cadence), which are then assimilated. Forecasts and nowcasts thus improve.
These "spectra-to-get-temperature-and-water" measurements were pioneered by other instruments in LEO (e.g., NASA's AIRS, https://airs.jpl.nasa.gov/mission/overview/), but LEO does not provide enough coverage to help forecasts.
To understand the benefits of GEO IR spectra, we do "OSSE's" (Observing System Simulation Experiments) to quantify how much improvement you get. You take a "Nature Run", make simulated observations (existing and proposed), and see if there is an improvement. (Since the Nature Run, which you made, provided ground truth, you can judge if there really was an improvement.)
Thankfully, many people have already done this. See: https://www.ssec.wisc.edu/geo-ir-sounder/osse/
In particular, looking at the figure there from Li et al., compare panels:
* (d) -- (Nature Run) - (existing data) ("CNTRL")
* (e) -- (Nature Run) - (existing data + GEO IR)
which both show differences between the Nature Run (NR) and the forecast.
The RMSE improvement (on a CONUS storm) is given as RMS of 0.55 (existing) versus 0.43 (with GEO IR), in degrees Kelvin. So that's 0.12 Kelvin or 0.22 Fahrenheit. Also, and probably more interestingly, the spatial pattern changes.
There are a lot of OSSE's reported on that page for these sounders. NASA is also conducting OSSE studies for a more ambitious multi-spacecraft observing system (https://science.nasa.gov/earth-science/decadal-surveys/decad...).
Studies like this (i.e., OSSEs like the ones above) are one of the main ways we decide how to build the next instruments -- what provides the most benefits vs. cost, which system parameters to push to improve and which are good enough.
This is an improvement as it provides better data and has nothing to do with the models that are used in a separate step to forcast anything. But that is said in the article as well, with the satellite being the first hyperspectral view on Europe and North Africa.
I am not sure what to make if your question.
Been pulling in some of the newer data and concepts from open-meteo to my home weather display specifically the ensemble data, helps to provide some level of spread over the forecast, I mean I am not that sure how useful it is but I kinda got used to reading it of the display.
https://openmeteo.substack.com/p/ensemble-weather-forecast-a... https://open-meteo.com/en/docs/ecmwf-api https://open-meteo.com/en/docs/ensemble-api
This data is from the third generation of Meteosats, which are the European meteorological satellites. A lot like GOES in the north-America. The main improvement is really significant improvement in resolution. The resolution is, depending on the channel, 9 times better than in the second generation. The main improvement in forecasting comes due to better information in the initial condition of the numerical weather prediction, but it is hard to quantify in advance. I'd be surprised if MAE, over the 15 days the prediction spans, would improve more than 0.1 C, if we talk about the raw prediction. There are plenty of things that this data is used for, but I would say that improved nowcast of cloud coverage, and energy production related parameters are likely to benefit most from the improvement in resolution.