UAH CI "Strength of Signal" Probability Forecasts Up and Running at 2012 GOES-R Proving Ground

With the EWP forecasters stationed in southwest Texas today, we were noticing some of the newer instances of convective initiation flaring up just across the Rio Grande River in Mexico.  Provided are a couple of the first views of the new UAH CI  SATCAST Strength of Signal (SS) output, as seen in the AWIPS-II display.

Figure 1.  AWIPS-II screenshot of the UAH CI Strength of Signal output from 1732 UTC (top-left pane) and corresponding satellite IR image (top-right pane), from the same satellite scan.

Figure 2.  AWIPS-II screenshot of the UAH CI Strength of Signal output from 1845 UTC (top-left pane) and corresponding satellite IR image (top-right pane), from the same satellite scan.

This new version of the SATCAST algorithm provides a continuous scale from 1-100 of the strength of the satellite-derived signal for near-future CI.  Per forecaster feedback from last year's 2011 SPC Proving Ground, this new product has taken a step away from the binary "Yes/No" output CI forecasts and now combines all of the available satellite spectral information into a single numerical value.  Higher values correspond to faster rates of cloud top cooling rates, colder/higher cloud tops, and wider updraft widths.

The Strength of Signal values are currently derived from the early beginnings of an ever-growing neural network in which the algorithm is trained from an empirical database with the correct answers of what immature forms of convective clouds go on to produce CI and which ones do not.  As this database continues to grow over the coming weeks, months, and years, the algorithm will "grow smarter".

Furthermore, this new paradigm is paving the way to soon begin training the algorithm with input from Rapid Refresh Numerical Weather Prediction model data. As this transition occurs in the near-future, the output will look the same, but it will become increasingly more meaningful as a true probabilistic product.

This work ties into ongoing research with NOAA's Earch System Research Laboratory (ESRL) in which the SATCAST data is being assimilated into developmental versions of the Rapid Refresh (RAP) and High Resolution Rapid Refresh (HRRR) models in order to "put the convection in the right place" for hot-starting the models, inevitably leading to more accurate short-term model output.  To that end, we envision a truly integrated and data-fused product in which a positive feedback effect occurs as short-term model data characterizing the broad-scale atmospheric environment is ingested by the CI algorithm in order to produce more accurate CI forecasts, and these more accurate CI forecasts are assimilated into the short-term prediction models in order to significantly improve how and where the models are simulating CI.  If the models can get the placement of early convection correct, then their longer term forecasts will take a giant leap forward in terms of accuracy in all aspects of their output.