Section 8.1.15.1 New Product Format From 18 May 2021

At inception around 2010, the extra-tropical cyclone diagrams (= "cyclone database" products) were generated using bitmap manipulation. As a result the products had a somewhat noisy/jagged visual appearance relative to many NWP products available nowadays (2021). In addition the software was proving costly to maintain. So in May 2021, and in close collaboration with the UK Met Office, ECMWF upgraded to a Python-based code suite, which applies the same graphical processing principles, but which saves the frontal objects in vector form and recreates fronts from those vectors wherever they are needed on a product. For these products this provides an aesthetically pleasing visual appearance, clearer rendering of details and a capacity to use transparency settings. In future this software change will also facilitate greater platform portability (e.g. for ecCharts, commensurate with user requests) and easier development of other product variants. During the upgrade a few other minor changes were applied, including correction of a pre-existing bug that affected the strike probability products. The main differences in output that resulted from these various changes are highlighted below, for user reference:

• a) Cleaner front charts, with no variations in front thickness, but more "front breaks"

Fig8.1.15.1-1 Comparison of a Control run T+24 synoptic chart forecast frame, between the old and new systems (same input data).

Focussing on the front that crosses the south of Baffin island (e.g. near 64N, 68W) on Fig8.1.15.1-1 one sees in the old product a fairly continuous red-blue line, of varying thickness, that in the new product is broken up into 3 segments. This is the result of the new software deciding whether to accept a front or not, based on the thermal-gradient-related masking criteria, then converting any front segments thus accepted into a vector format, and then replotting those vectors with constant front width. In the old version the mask erases, in places, part of the potential front that had been drawn as a graphical entity more than 1 pixel wide, and then that partially erased coloured line is retained as is on the final product.

• b) Minor changes to cyclonic features identified

Comparing the spots (=cyclonic features) between the two renderings on Fig8.1.15.1-1 we see some differences. In general the substantial features (large dots) are the same, whilst the other (small dot) features, that are innately weaker, are more liable to differ slightly. Differences are essentially random and relate to identification algorithm sensitivities. These algorithms can respond one way or another when one of the identifying thresholds required for feature acceptance is on the borderline of being met. Most thresholds relate to field values (e.g. computed thermal gradient), although one defines a minimum separation distance between features. One cannot say that one plot is better than the other. When looking at an ensemble (e.g. on Dalmatian charts) such random differences will tend to cancel, and so are unlikely to change user interpretation (e.g. see Fig8.1.15.1-3)

• c) Precipitation plotted as interpolated colour fill rather than non-interpolated grid-fill, giving a smoother appearance

On Fig8.1.15.1-1 the cyclone over NW Russia is mainly associated with snow (blue) and a rain-snow mix (pink). The newer rendering has a smoother appearance, but is more "approximate" because of the interpolation (e.g. one pink pixel has disappeared). Note also that the representation with colour-fill of two dimensions - type and amount - means that it is not possible to generate a completely smooth transition across all colour boundaries - this means that spurious narrow filaments of lighter shading (falsely implying less precipitation) will sometimes appear between different colours (e.g. crossing 63N 60E on Fig8.1.15.1-1).

• d) Spaghetti Fronts uses transparency

Dalmatian charts of all types, including the "synoptic feature type" option shown in Fig8.1.15.1-3, have all benefited from improved rendering. Output from the newer version has also been enlarged, to further improve legibility, recognising also that workstation displays have generally evolved since 2010 to have higher resolution. If this enlargement causes issues for a user note that browser windows/tabs can always be rescaled (typically using Ctrl+ or Ctrl-). Minor differences between the spots again relate to threshold handling in borderline cases, as discussed in (b) above. Plotting mechanisms, such as scaling, colours, symbol styles and overlaying order should all be the same in the new system as they used to be, for all types of Dalmatian chart.

• f) Bug removed for Strike Probability charts
  

Fig8.1.15.1-4  Strike probability chart examples from the old (left) and new (right) systems from the same input data

Whilst some of the differences between the two plots in Fig8.1.15.1-4 arise for reasons discussed in (b) above, most relate to removal of a bug which had previously caused a few cyclone tracks to be inherited from the wrong valid time. Most notably the low probabilities of cyclone activity (at this 34kt threshold) that one can see, from the old system, in regions peripheral to the more brightly coloured "storm track zone", have correctly been removed in the new version. Plot size has also been increased slightly.

• g) Miscellaneous

The look of the cyclone plume diagrams, and the track animations, accessible by clicking on a cyclone at T+0, have also changed in the new version, but the content and functionality are the same. The entry point for the cyclone database web products has also been upgraded, whilst retaining the same functionalities. Output images are now all in "png" format; previously most were "gif".

See the new Cyclone Database Products.