Precipitation Forecast (GFS 0.25 res) a Bone of Contention

Hi Tom,

I know I have raised this before, though recent rain events have driven me to raise it again.
Back in the (GFS 0.5 res) days the Precipitation Forecast for my local was near on perfect (from a timeliness and Amount point of view).
I do acknowledge your past statements on Precip being (a poop shoot) or rather hit and miss to a point, however…
Since transitioning to (GFS 0.25 res) the forecasts have been extremely inaccurate. Purely I feel due to the GFS 0.25 modeling.
We have had so much rain recently and GFS/WxSim continually forecast <2mm or 2mm (when we actually receive near 28mm on a recent example.
Today the forecast was for Clear and a bit of Breeze, nothing else mentioned (no rain - nothing).
So far since midnight we have had 9.6mm and near non stop showers and full Overcast all day.

In order to question the GFS 0.25 Modelling I have tried to visualize it by adjusting one of my GFS Maps scripts and adding our Location on the Map in the form of a red dot (converted today’s 18z into an animated gif = attached below). Admittedly the maps are 3hrly Totals but at least tries to highlight my issue. (or dare I even go outside the box and ponder may the 3hrly data be an issue).

What this shows is that WxSim has forecast correctly what GFS has supplied (no rain for our local. Tonight is saying <2mm and if you look at the last map yep rain all around us, but only about 2mm colour on our local. To tell the truth (laughingly) I have wondered at times whether the apcp is a decimal point out or even an imperial vs metric issue, though looking at the GFS Maps that is obviously not the case.

I am at a loss as what to do, as WxSim can’t create something that is not injected into it. (GFS Modelling seems to be pulling up precip at the western mountainous areas, our big recent cold fronts have and always do, sweep in from the west). This can be seen on the maps.

I would like to know if any other users have experienced this issue since the inception of GFS 0.25 res. Or is it just my location that the modelling is getting way wrong.

Aside from that the Temp Forecasts have gone from Good to Outstanding, though I credit that to WxSim-Light etc… (compare image below), my MAE hit a low 1.08 C yesterday http://hrvistaweather.com/weather/wxsimcompare.php

Kind regards,


gfsmap18z20160725.gif

Hi Tom,

Just a bit more info comparing Australian BoM Beta Forecast for Franklin (our town) and WxSim, both below are from same 20160726 00z run time.

Big difference in Precip and % chance of, obviously the BoM have access to more refined data and local inputs.

regards,


I have collected accuracy-stats for temperature for about all freely available European models + the Amercian GFS and CMC-models for a while now for northern Europe. GFS have rather been worse than better in the “upgrades” they have did to it and its one of the worst of the whole accuracy-test. While European models have as average 1-1.25

Hi Henkka,

Greatly appreciate the response, wow your verifiaction-page is brilliant. It depicts some large differences for sure.
When you say “The values sounds not big but there are a clear difference.” I agree especially when it does not need to be out by much at all to have some or no precipitation, this is obviously what I am experiencing here in Tassie. At least know I suppose I feel better that it’s not just me that see’s it.

And at the end of the day there is nothing I can do, so will just have to live with it.

Kind regards,

For Europe have i a couple of times offered Tom data from the European models as i have them dumped to mysql but he haven’t taken the plunge at least yet. European data are available down to 0.01

Hi Henkka,

Appreciate the feedback, would be nice if the equivalent modelling where available for us down here in the South.

I meant to note it in my initial post, that I had posed the concept to Tom ages ago being, is there a remote possibility that the Optimization/Modelling/Data from a Geo point of view is not taking into account Realtime (or rather be it actual) Tectonic Plate movement within it’s calculations. Which in turn, in my opinion could have potential to result in the finite discrepancies we see over a small area. I realize this is an (out of the box) question, though one I always ponder. We would all assume/hope that the super computer modelling process does adjust. The reason I question this is primarily due to the move from a higher res model to lower res model which as you have also proven, highlights significant difference over a small area. I suppose a bit like a Lightning detector which is not truly aligned and has no additional true triangulating correlation stations feeding into the final output (not to mention the underlying software computations to adjust the changes in inputs), over distance it can be way off. Possibly not the best as-simile, but hopefully you get my drift.

Kind Regards,

Henkka, thanks about that European data suggestion/offer. I don’t know if I’m up to implementing it yet, but would like to correspond more on the size of the files, and how this might be done. Thanks! :slight_smile:

Tom

Tom, email sent :slight_smile:

// Henkka