Thompson microphysics: what we would want to test before switching
We ship WSM3 today. Thompson is the obvious next-step microphysics for soaring forecasting. Here is what we would want to validate before paying the extra runtime cost.
Thompson is the microphysics scheme most often recommended for forecasting applications where cloud detail matters. It is a six-class double-moment scheme: it tracks both the mass and the number concentration of cloud and ice particles, where the simpler schemes only track mass and assume a fixed droplet number. The double-moment treatment is what lets the scheme respond to airmass cleanliness - clean maritime air carries fewer cloud droplets, which changes how the model partitions condensation between cloud water and precipitation, which changes cloudbase and precipitation timing.
That is the kind of behaviour that should matter for a UK soaring forecast, and it is the obvious next-step upgrade from the WSM3 scheme we ship today (covered separately in the WSM3 post). The question is what we would want to see before paying the extra runtime cost to switch.
Why this matters specifically for soaring. UK air is often clean maritime, and a fixed-droplet scheme like WSM3 is known in the literature to bias modelled cloud properties in those conditions. The bias direction varies by paper but the most common observation is that a single-moment scheme over-predicts saturation in clean air and therefore puts cloudbase a touch lower than it should be. Whether that matters in practice for a 4 km soaring forecast is exactly the kind of question we cannot answer without comparing schemes side-by-side.
What we would want to test. Three things, all of them downstream of having the verification pipeline in place. First, side-by-side runs of WSM3 and Thompson on the same set of cycles, with cloudbase from both compared against pilot-reported observed cloudbase at known launch sites. Second, a check of the convective onset timing - some published comparisons suggest Thompson's more responsive condensation slightly delays first-cumulus appearance relative to single-moment schemes, which would matter for trigger-time forecasts. Third, the runtime cost on our specific 4 km configuration, which is reportedly 15-20% extra per cycle but depends heavily on resolution, time step, and how much active convection is actually present.
What would push us to switch. A clear validation signal that Thompson cleanly improves cloudbase RMSE on UK days, with a runtime hit we can afford within the four-cycle-a-day budget. Failing that, a use case that genuinely needs better mixed-phase or icing detail (we do not have one yet, but a wave-flying or alpine-coverage product might).
What would push us not to switch. If validation shows the cloudbase improvement is small enough to be lost in the post-processing noise. Or if the runtime cost squeezes the cycle window enough that the freshness of the morning forecast suffers. Either case is a reasonable outcome.
Microphysics is on the roadmap behind the validation pipeline. Once that pipeline can score a Thompson run against the live WSM3 baseline cleanly, this is the next physics upgrade in the queue. When the comparison happens this post will be updated with the numbers that come out of it.