Where did all the moisture go?

The models stole it all!

One of the additions this year is an observational program (brought to you by NSSL) where we have an Microwave Radiometer (MWR) offering vertical profiles of moisture and temperature over the lowest 4km AND a new radiosonde intercomparison to go along with it (a Vasaila RS92 sonde and a new IMET sonde). The goal is to use the soundings to compare the MWR and thus offer a calibration data set, but also to see how well the moisture retrievals compare to actual, in clear-sky conditions.

Since we have had two sonde launches this week so far we can see how well the IMET and Vasiala measurements compare and also how the MWR compares to both of them. So far the results are that the two sondes are very close. The comparison with the MWR was at first horrible, until the MWR was re-calibrated. Sensor drift from one of the eight channels was to blame at least for the low level structure of the moisture (most noticeable in the RH field). for the next point we have to understand that the majority of the information content in the moisture channels is contained below 4km and only amounts to about (on average) 1.6 pieces of information. This means within the lowest 4 km we have at most 2 effective measurement points for moisture. So the vertical profiles tend to be smooth and more like an average which is why agreement aloft between moist layers and the MWR is essentially low.

OK has been pretty low on moisture the last couple weeks. We spent some time forecasting in the Iowa area the other day where model precipitable water was in excess of 1", but in reality was only around 0.5". This posed no problems for the models to generate storms along a cold front associated with an upper low. Some warnings resulted from MI arcing through IA that evening with a few wind and hail reports, but generally good CI and SVR forecasts.

At issue was this anomalous model moisture and if/how it would play a role in the forecast. So many models had storms though that it was hard to discount storms even in this lower moisture environment. Even looking back at the verifying soundings from DVN, ILX, DTX it was evident that the CAPS ensemble control member was 2-3 g/kg too moist through the depth of the boundary layer. How can we have errors that large and yet have some skill in the convective forecasts? The models simulated storms were a bit on the high side in terms of reflectivity, lasted a bit longer, were a bit larger, but still had similar enough evolution. I guess you could consider this a good thing, but this should really drive home the point that better, more dense moisture observations are needed.

We really need to see WHY these errors occur and diagnose what is contributing to them. In this case, the control member was an outlier in terms of the overall ensemble. Why? Was it the initial conditions, lateral boundary conditions, the perturbations applied to the ensemble members, or some combination thereof that set the stage for these differences in convection? Or was it the interplay between the various model physics and all previous factors? We will need to dig deep on this case to get any kind of reasonable, well constrained answer.

In order to address, at least partially, these issues we need observations of moisture within the PBL. In fact we could even benefit from knowing the boundary layer. It is at least plausible to retrieve that field and then derive the PBL moisture. Such is the goal of the MWR type of profiler: to derive the lowest layer moisture structure, addressing at least some of our issues. Regardless, these high resolution models REQUIRE observations to verify both the processes and statistics of these models in order to make improvements.