8 Outline of the algorithm Interpolate (irregularly distributed) HrW product down to pixel level Apply the vector field on every pixel n times ( n results from user’s specifications plus the limitation of max. + 1 hour) Construct the predicted image / NWCSAF product by putting pixels at their predicted positions Fill gaps through nearest-neighbour or average interpolation Other post-processing, e.g. applying land-sea mask or writing NODATA at the edge of the image
9 Assumptions... …that might be worth some discussion: We take every piece of HrW information that we can get, i.e. we eventually mix HrW vectors from those channels which the user selected in the HrW configuration file. We make predictions at multiples of slot intervals, i.e. for MSG in nominal mode: +15, +30, +45 and +60. Thus, we “have” images / products earlier than without EXIM, yet we don’t produce imagery for times where we have no actual satellite images.
10 Input fields we already tried SEVIRI: WV6.2, WV7.3, IR8.7, IR9.7, IR10.8, IR12.0, IR13.4 (yes, VIS0.6-IR3.9 work technically just as well but there are doubts whether we should enable their forecasting in the first version) Cma CT CTTH_HEIGHT, CTTH_PRE PC_PR1 and PCPh_PC CRR and CRPh_CRR SPhR_BL, ML, HL, KI, LI, SHW, TPW
11 Output files we envisage One file per channel / product per forecast date Mimicking output format for analysis output as perfect as possible. Ideally, EXIM products can be distinguished from their analysis counterparts only through the filename and the directory where they are dumped “DATABUF” for satellite imagery, netCDF for products
12 Output files we envisage (2) Relevant quality flags shall be extrapolated as well and be included in the forecast netCDF The EXIM-specific quality flag (“actually extrapolated” vs. “achieved through post-processing / gap-filling”, applicable to both SEVIRI and product forecasts, hence one per forecast date) is dumped as DATABUF
13 Validation The SAFNWC PRT defines EXIM’s threshold accuracy as: on average better than persistence forecast Target accuracy is: always better than persistence forecast Hence, the validation approach is quite obvious: Compare the EXIM forecast with what was actually observed 15, 30,… minutes later and verify that the displacement actually did anything positive on the skill score (we use Peirce’s [abbr. PSS] and, for the CT product, its multicategorical variant, “R” “verification” ”multi.cont”)
14 Validation The following slides show PSS time series Extracted every 2 hours Over ~3.5 months (Sep 14 – Jan 15) from ZAMG’s operational NWCSAF v2013+ suite, meaning in particular the HrW default model configuration file was used 3-hourly NWP data, ECMWF, 1 degree resolution European area
35 Conclusions - SEVIRI Release of EXIM certainly OK for IR WV Perhaps it would be OK even for the entire spectrum, but the lack of modelling the solar radiation impact for 0.6 – 3.9 means: Risk!
37 Conclusions – NWCSAF products Release of EXIM certainly OK for CMa CT CTTH CRR
38 Conclusions – NWCSAF products Big question-mark on PC with its pronounced dependence on illumination conditions (recommended to users only during high-summer season?). For PCPh and CRPh, stringent validation difficult (uncertainties about areas where one can have 100% confidence). Statistics nevertheless indicate that there is value in the extrapolation (and things should improve with the 24h products announced by AeMet).
39 Conclusions – SPhR (future TqPh) SPhR and extrapolation with the atmospheric flow do not seem to go well together: For the stability indices sub-group, this could have been expected. For the moisture parameter subgroup, we have evidence that - to a large degree - the task is to predict the outcome of temporal NWP interpolation, at least for NWP intervals ≥ 3 hours → Leave it in the portfolio for those who wish to experiment with 1-hr NWP (actually recommended by SPhR developers)??
40 Future developments Until v1.0: Consolidate the software (in particular working on the new netCDF output format) Incorporating User WS outcome that can be accommodated fairly quickly Continue validation Afterwards: Incorporating User WS outcome that is rather challenging to be implemented Take new NWCSAF/GEO products on board (we are always “day- 2”) MTG