RE: PoP and Sky fall way short

Forums (Forecast Builder)

RSS (Opens New Window)

Greg Mann, modified 9 Years ago.

PoP and Sky fall way short

Youngling Posts: 2 Join Date: 9/25/12 Recent Posts

There are two very nagging issues with SuperBlend, PoP and Sky. Both are a result of woefully inadequate methods or non-scientific / non-statistical approaches. Because these two elements are insufficient and will not improve enough with each successive cycle, edits need to be made over and over again just to correct the solution to the guidance advertised result, let alone any target of opportunity. This is unnecessary janitorial work imposed each forecast cycle.

1- PoP should only be calculated from ensemble based systems (including time lagged) using at least a simple head count or statistical corrections from single model sources. Raw guidance sources using QPF as a proxy for PoP must be abandoned. The method is not even close to valid. The inclusion of these raw sources of PoP result in an inability to forecast values in excess of 85-90%. Verification statistics show that the 100% has never been offered, even in period 1.

2- Sky will not honor extensive clouds at lower elevations because moisture below 925 mb is ignored. Furthermore, moisture above 500 mb is also commonly ignored. Finally, the blending of sources that use model sky (usually as a binary) and RH produce outcomes that have little spatial or temporal integrity.

This mismatch of application for each element in the smart inits pose big problems with blending given the lack of consistent sound community accepted solutions.

Andy Just, modified 9 Years ago.

RE: PoP and Sky fall way short

Padawan Posts: 99 Join Date: 6/2/15 Recent Posts

Greg,

This is also a good post for the National SmartInits Config project (https://collaborate.nws.noaa.gov/trac/nwsscp/wiki/Gfe/Smartinits/NwsInitsConfig), and probably a good check with the National Blend of Models.

On PoP, we realize the PoP from QPF is not the greatest scientific approach. There is some literature on the topic, and realistically when it comes an ensembling approach a true neighborhood probability idea would be preferred - especially in the short term when there are a lot of high resolution models present. Unfortunately what we are doing in the current framework can already max out our PX servers, and those neighborhood probabilities techniques can be very computationally intensive. Now I have looked at some PoP reliability diagrams for September and October @ ARX, which encompass both wet and dry months, and it's surprising how reliable the PoPs are both from our official forecast and SuperBlend. These diagrams will be shown on the FB webinar today. Note: individual data sources can produce 100% PoP when QPF >= 0.5" I have also been working with the SmartInit group to incorporate stratiform precipitation from models when it's available as past winters we have found the PoP/QPF relationship to be underdone, i.e. we need to have higher PoPs for lower QPF, and correlates that models can do better at prediciting stratiform events than convective.

On Sky, this is just a tough grid in general. We need to have like cloud fraction or something like that from the models, rather than trying to derive a Sky grid from RH. There are checks in place vs PoP, but still, there's room for improvement and we need to get the cloud output from the models. Even Obs Sky from RTMA and that can be lacking.