Graphics Products from the NCEP Medium Range Ensemble (16 November 1995) Note that the graphics files described here are in a state of continuous development, and that when we make changes to the graphic files in N-AWIPS we will update this document accordingly. Please remember that these products cannot be used in any way for commercial purposes. SUMMARY OF CURRENT PRODUCTS: The following products are produced twice a day in files named ens_yymmdd_00 and ens_yymmdd_12: Spaghetti diagrams (12, 36, 60, ... 252 hr for 00z runs): 500mb height: 5220m (cool season only, Oct. thru Apr.) 5400m 5520m 5640m 5760m 5820m 5880m (warm season only, May thru Sep.) 500-1000mb thickness: 5400m 850 temperature: 0 C (cool season, Oct. thru Apr.) 15 C (warm season, May thru Sep.) MSL pressure: 992 mb (cool season, Sep. thru Apr.) 1000 mb 1008 mb (warm season, May thru Aug.) High and Low centers 1024 mb (warm season, Mar. thru Nov.) 1032 mb 1040 mb (cool season, Dec. thru Feb.) 250 mb winds: 50 m/s (varies by month from 25 to 50) 850 mb winds: 15 m/s (varies by month from 12 to 15) 700 mb RH: 90% Note that the seasonal contours are chosen based on the month in which the forecast starts, and that warm and cool season months are defined separately for each variable. The following products are prepared at 00Z only, in files named ens_yymmdd_00_cluster: Cluster products (84, 108, 132 hr): Ensemble mean: 500mb ht, 1000mb ht, 500-1000 thickness Ensemble spread: 500mb ht, 1000mb ht MRF: 500mb ht, 1000mb ht, 500-1000 thickness Each Cluster: 500mb ht, 1000mb ht Each map except spread includes the anomaly in dashed lines. Probability products (72, 96, 120, 144 hr): 500mb height: 5580m, 5640m, 5700m 500mb height 24hr tendency: +30m, -30m 500-1000 thickness: 5400m, +60m anomaly, -60m anomaly 850 mb temperature: 0 C 700 mb RH: 70% AVAILABILITY The jobs that prepare data for these graphics are not yet fully operational, so the products may be delayed or unavailable from time to time. The most recent two days of GEMPAK graphics metafiles are generally available. They may be viewed with ntrans (available from Computing Development Branch of NCEP). LOCATIONS OF METAFILES: N-AWIPS servers: In ntrans, bring up the model selection panel and select the model ENS. nic server: The metafiles can be downloaded by anonymous ftp from nic.fb4.noaa.gov in the directory /pub/nadata/meta/model/ens by the following procedure: ftp to: nic.fb4.noaa.gov For userid, type: anonymous for password, use your full ip-address. Then cd pub/nadata/meta/model/ens. There you will find 3 files for each day. Two files contain "spaghetti" plots as described below. The other file (_cluster) contains additional plots for the ensemble mean, cluster means and probabilities derived from the ensemble. Experimental versions at NCEP: Experimental versions of these products are available to users within NCEP in the directory /export/sgi80/wd20rw/meta . The easiest way to view these files is to cd to this directory, then enter ntrans (note that if you want to print these graphics, you should instead be in the directory where you want PostScript print files written). Then choose the entry for user-defined files in the ntrans open-file menu. If you did not change to the metafile directory first, you will need to enter the directory name followed by /* in the upper entry area, then choose search. Categories of metafiles can be identified by one of the following suffixes: _t test: current version of code to be implemented in N- AWIPS, serves as a backup to N-AWIPS. _u alternate: contains out-of-season contours corresponding to the test version. _x experimental: contains whatever new products we are testing at the moment. _v verification products, currently rerun 3, 4, 5, and 10 days after the forecast start time. In the very near future the experimental products will be completely reorganized. After they have been tested and commented upon, some of the new set of products will become operational. Each file will be identified by a series of 2 or more suffixes: region: _us _na _sa _nh _sh _car _atl product: _pr _en _cl _sp _xx (probability, ensemble mean and spread, clusters, spaghetti, experimental) data source: _e ecmwf ensemble _i international, using multiple centers (none) ncep ensemble verification: _v verification _vn verification through n days (none) no verification contours: _a alternate, all seasonal contours (none) standard, varying by month version: _x experimental _t test (none) operational example: ens_951103_00_us_pr_v3_t test version of probability on US map with verification through 3 days. Note that in the new scheme experimental versions of current products (_x) are in different metafiles from experimental products (_xx). DESCRIPTIONS: Spaghetti diagrams: Each diagram shows the same one contour of one meteorological field from each of the 17 runs making up the ensemble. In addition to the T126 MRF and its T62 extension (shown in yellow) and the 12Z T126 AVN run with its T62 extension (shown in green), there are 10 perturbation runs (shown in red) based on adding and subtracting 5 perturbation fields from the CNT run (T62 MRF, shown in orange), and 4 perturbation runs (shown in blue) based on 2 perturbation fields applied to the 12Z AVN. The "Control" runs are plotted on top of the corresponding perturbations, and all runs from the most recent cycle are plotted on top of runs from 12 hours earlier. The principal control run in the current cycle (MRF or AVN) is plotted with a wider line wherever it does not cross over other lines. Counting the MRF, the T62 MRF (CNT), and AVN, there are 17 runs in all. Probability: The probability fields are generated by counting at each point the number of runs in which, for example, the 500mb-1000mb thickness exceeds 5400m, rescaled to a percentage. For fields like 700mb RH, for which the spaghetti diagram becomes almost unreadable after a day or two, this is basically a more legible way to present some of the same information. Currently, for historical reasons, the probability products are produced for 00Z verifying times while all the other products are produced for 12Z. This will be changed in the near future. The generation of probabilities is unweighted; that is, all the runs have equal weight. Ensemble means and spreads: The ensemble means are weighted, using weights recalculated every few weeks. The weights are empirically determined for 5 groups of ensemble members. The anomaly fields shown in dashes on the ensemble mean maps are currently based on Climate Prediction Center databases covering the years 1979 thru 1988 (500mb) and 1979 thru 1985 (1000mb). We expect to switch to a longer period database. The spreads (standard deviation of the ensemble members about the mean) are shown without normalizing for the climatological variance. This means the spreads tend to increase with latitude along with the amplitude of climatological variance of height. We expect to add a product with normalized spreads in the future. MRF maps: The MRF run is shown on maps analogous to those on which the ensemble mean and the clusters are shown. The anomaly fields shown in dashes on the MRF maps are currently based on CPC databases covering the years 1979 thru 1988 (500mb) and 1979 thru 1985 (1000mb). Clusters: The 17 runs are grouped into up to 6 clusters. Currently the clustering is done separately for each of 3 forecast hours and 2 levels, but in the near future it will be changed so that all products will be based on clustering of 500mb heights at 132 hours, which will make it possible to track the evolution of each cluster. The anomaly fields shown in dashes on the cluster maps are currently based on CPC databases covering the years 1979 thru 1988 (500mb) and 1979 thru 1985 (1000mb). Clustering is based on a correlation procedure. The first and second clusters are based on the two runs that are most unlike each other (the extremes within the ensemble), and other runs similar (anomaly correlation greater than 0.6) to these are grouped into these two clusters. The remaining forecasts are grouped into up to 4 additional clusters. Each run can be in only one cluster. A cluster may consist of only 1 run. When the anomaly correlation between the most unlike runs is greater than 0.6, all the runs are in cluster 3. The cluster means are formed from unweighted averages of the runs. We use 17 runs rather than the 46 runs (5 cycles) used for extended-range prediction because at short to medium range the runs are often grouped into clusters by cycles. For short and medium range it makes more sense to cluster one day's runs and compare these clusters with those made on previous days. The cluster calculations are performed on a rectangular area surrounding the 48 states (25N to 65N and 60W to 130W). In the future we plan to make different clusters based on different areas, generally corresponding to the areas of the maps they are shown on, and also separately for the eastern and western USA. The cluster plots are labelled with somewhat cryptic titles that show which runs are included in each cluster. The title ends with two strings of letters and a number. Each string of letters represents the runs in one forecast cycle (00Z and the preceding 12Z). Each letter in the two strings represents one run in the ensemble: m=T126 MRF, c=T62 MRF, a=AVN, p=positive perturbation of the preceding control run, n=negative perturbation of the preceding control run. Each capitalized letter represents a run present in this cluster, while lower-case letters are placeholders for runs not present. The number following the two strings is the number of runs included in this cluster. For example: McpNpnPnpnpN aPnpn 5 indicates that this cluster includes the T126 MRF, 3 MRF perturbations, and one AVN perturbation, making 5 runs in all. LOCATION OF GRIB DATA FOR INDIVIDUAL ENSEMBLE MEMBERS: How to access NCEP global ensemble products in grib format: ftp to: nic.fb4.noaa.gov For userid, type: anonymous for password, use your full ip-address. Then cd pub/ens. There you will find 15 files. The filenames will identify the variables contained in the files. We have: Height at 1000 700 500 250 mb Winds at 10 m 850 mb 250 mb Temperature at 2 m 850 mb Rel. Humidity 700 mb Pressure reduced to mean sea level PRMSL Accum. Precip. Right now the precip data is for 24 hr non-overlapping periods for the 00z and 12z runs. There may be problems with the close to surface variables, too. These files do not contain in the record headers the ensemble extension description yet. Each file is for one variable, with 17 model runs. We will put out a description of the files and some software to read the ensemble grib extension headers in the near future. The first data record of the file contains the T126 control; the second is the t62 control; records 3-12 are perturbations from the 00z cycle (neg-pos pairs); record 13 is the 12Z (previous day) control and 14-17 are 4 perturbed runs from 12Z. The first set of 17 data records is for lead time 0 hr (12 hr for the 12Z runs), the next is for 12 hr, 24 hr etc. out to 10.5 days, after which it is given only every 24 hrs. Please give us your feedback regarding the format of the files. We are also interested learning about your experience with the use of the data. General information on decoding and unpacking grib data, along with some software can be found in readme or README files on: / /pub/info For information and software on how to read grib format files, using different platforms: cd /pub/info NOTES ON INTERPRETATION It sometimes happens, especially for the AVN cycle, that all the perturbation fields have the same local shape; in that case, the shorter range forecasts (in which the response to the perturbations is fairly linear) may show the perturbations grouped in pairs or 5s. Occasionally the grouping into 2s or 5s is different in different active regions, for example, western Atlantic and eastern Pacific, because the signs of each major feature of each perturbation field may vary at random from one area to another. In such cases we may get 3 or 4 clusters which represent the various combinations of eastern and western patterns. For this reason we are planning future cluster products in which the clustering is done separately for the eastern and western USA. If you believe that our ensemble adequately covers all likely outcomes, then you can take the probability fields as a predicted probability of the event. In our experience it is slightly more likely that the verification will fall outside all the perturbations than that it will fall between any particular pair of consecutive perturbation values, so this assumption is almost true and the ensemble probabilities are slightly (less than 10%) closer to 50% than they should be. Since all the spaghetti diagrams are for 12z, you can look at the 00hr AVN fields as verification, and the 00hr AVN perturbations show the initial anomaly amplitudes. REFERENCES: On the generation of the NCEP ensemble, see Toth and Kalnay,1993, BAMS, Vol. 74, p. 2317-2330.) and NCEP Office Note 407. On the use of ensemble forecasts at NCEP, see Tracton and Kalnay, 1993 (Weather and Forecasting, Vol. 8, 379-398).