Rain Value Meaning In Meteogram?
nwindy last edited by
I had someone ask me (while I was looking at my weather app, Windy) what's the percent chance for rain today? And, didn't have an excellent answer, just that it is predicted to rain X in. between the 1 and 4 o'clock hours.
Led me to doing some reading... came up a little short with specifics in Windy... So...
Short version: In the meteogram, what do the rainfall values mean in regards to probabilities/confidence and area (if at all)?
Starting with what I know:
- The values are for every three hours (not premium).
- When I move the pin/location indicator on the map, the data in the meteogram changes, even for the smallest movements.
- The values are absolute amounts (0.2 in, 4 mm, etc.)
- In meteorology there is a concept of Probability of Precipitation (PoP) which is calculated for a point, but sometimes also for an area. From this weather channel's post:
'' PoP = C x A where "C" = the confidence that precipitation will occur somewhere in the forecast area, and where "A" = the percent of the area that will receive measurable precipitation, if it occurs at all. ''
And this ^ is from the National Weather Service. Both C and A are themselves probabilities - values between 0 and 1 - so multiplying them together gives another value between 0 to 1. (Not what is displayed in the meteogram data)
- Is the ''location'' of the map in windy actually a point and the meteogram only shows the data for the weather happening at that point every three hours?
- Does the meteogram rain value imply a 100% chance of rain for the point-only location of the pin? Meaning there's no concept of area in that rainfall value (the A in the PoP formula) and that there's also no concept of confidence (C in the PoP formula) that actual rainfall will be the displayed rainfall value for that point?
- Is it true/common practice (as the link mentions) that rain amounts less than 0.01 in. are not included in the meteogram?
It seems that PoP provides a probability of rainfall over a rainfall threshold (0.01 in), with no indicator of how heavy the rain will be beyond that threshold. Whereas, Windy provides the heaviness/rainfall amount in absolute values, but only for a point. And if we want to communicate to other weather app users what the percent chance of rain is, our response should be: It's complicated, what is your app's definition of percent rain? And of corse, other apps don't say what the percentage actually means. Maybe windy could have an option to display meteogram values averaged over a user defined area... and then the values will change to %'s. Anyway...
You have 2 types of weather forecast :
The first one is the one proposed by Windy models and gives you one forecast at a specific day and specific hour. There is no probability associated with this forecast because this is one single forecast calculated by the model deterministic.
Now, when we talk about ensemble weather forecasting, the model here doesn't calculate only one single forecast but a group of around 20 forecasts, each obtained by changing some parameters. From this ensemble of 20-21 forecasts at a specific day and specifc hour, you can extract data in this case with probabilities like 40% chance of rain for July the 15th at 3.00PM. This is very useful to be able to quantitify the incertitude on data
This ensemble approach looks great but because of the increase number of calculations (20-21), grid will be much bigger and therefore data will be less accurate
nwindy last edited by
So, Windy takes the model data (once for each of ECMWF, GFS, NAM, etc.), runs it through deterministic models, and that spits out windy's 'layers,' generated on a 'high' resolution grid. And the results are not probabilities, they are straight up predictions for what's going to happen at that geo-temporal point. Correct me if that's wrong.
With the ensemble method; because it uses 20-21 forecasts for a larger area (lower resolution grid), with each input perturbed, and some of those numerus forecasts result in rain and some result in less rain, that is how that method generates a probability?
Is the reason you can't do this at a higher resolution just a computational limitation, or are the 20-21 points representative of sub-pixels in different geo-areas of one of the larger main output pixels (and so with more sub-pixels, the result is more accurate but at the same time covers a larger resulting land area)?
Yes, the way the ensemble model calculates the probability is as you explain : for instance for the rain, each dot of the 21 dots can give a different level of rain (1mm, 4mm, 10mm,...). Based on that set of different levels of rain, the model can calculate several probabilistic values such as the average, variance,... This will allow you to quantify the risk of rain
Now, if the model has to adopt the same approach for many other parameters, such as pressure, temperature,... And for each parameter do the 21 calculations (do not forget that each calculation is the equivalent of a single deterministic calculation), computional limitation will be the issue if the model keeps the same grid. To overcome this problem, grid is bigger, calculation of the 21 dots for the different parameters becomes ok but accuracy will decrease.
Computational calculation is key. In average, weather forecast improves by one day every 10 years thanks to the computer (super calculator) progress