RASM-ESRL is a sophisticated physics-based computation at rather high spatial and temporal resolution; it is not a simulation.
To extend past the initial state, RASM-ESRL uses a full suite of satellite inputs such as AMSR2 sea ice concentration, separate forecasts for ice and air temperatures and movement; water temperatures, salinity and currents, and so forth.
Like ECMWF, RASM-ESRL makes testable predictions about near-term future conditions but is farther-reaching in its suite of meteorological, surface ice and oceanographic outputs.
While its predictions won't be a perfect match to subsequent observational data, it is pure fantasy to think one can adequately intuit future developments in Arctic Ocean ice by navel-gazing at weather charts, not with ~49 variables in play. Better mousetraps were built decades ago. And this one is free and just a click away.
My interest is whether RASM-ESRL can be improved. That won't happen to any extent through better meteorological forecasts because those largely hit the wall long ago. Nor will it come about from a finer grid as computing resources are already strained, nor from code tweaks as the latest are already in use.
The opportunities more likely lie in advanced processing of the satellite input data that resets the daily initial state. Two that come to mind are UH AMSR2 3.125 km concentration and UH SMOS-Cryosat2 thin ice. The appeal of these is that the algorithm doesn't need to change, only pointers to better input.
From our perspective, since neither makes forecasts but both are compatibly formatted as netCDFs with merger of the three datasets seamlessly supported within Panoply, we may be able to go forward on our own with hybrid products rather than wait on NOAA's timetable.
In twelve days time, we could revisit this particular ESRL prediction by comparing it to what actually transpired around the ice pack periphery according to SMOS and AMSR2. The advance with Panoply is that we can walk back the comparisons from colored map projections to the data grids themselves with no need for matlab, command line, or number viewing.
Another option is to utilize narrow but better products in reanalysis. That is, ESRL is concerned with forecasts, not hindcasts. Here we could take for example SMOS, mask out places where its signal saturates (becomes unreliable), and use the rest of it (where it is more accurate) to replace less accurate peripheral thickness in ESRL initial states. This can be done seamlessly within a shared palette in Panoply. The hybrid is then like a reanalysis product. It does not make forecasts but could be lead in to such products as the earlier frames.
Technical note: UH SMOS is seasonal, limited to ice <1.3 m thick and suffers from a larger pole hole; its posted date lags two days behind calendar and one day behind UH AMSR2. The latter is shown below with the fixed open water boundary of Oct 1st in dark gold superimposed on the other dates.