I decided to email the NHC about two new models, NNIC and NNIB, which appear this year in their model data and I included the email I got back below. I also confirmed that AP21 through AP30 are the extra 10 members of the GEFS. I will add the additional information to HurricaneCity's model system over the next few days. I need to review the changes they made to their model names file and update my site with the new information. I had already added some information I had found about the NNIC and NNIB models, but will update it to reflect the new information as well. Here is some additional information on the NNIC/NNIB models: NNIC is the Neural Network Intensity Consensus model NNIB is the Neural Network Intensity Baseline model The inputs to NNIC include intensity forecasts from 4 deterministic intensity models (HWFI, AVNI, DSHP, LGEM) and 4 other predictors. The model input includes 5 predictors as follows: (1) the mean intensity from the 4 models, (2)-(5) The deviation of each model from the mean. The four other inputs include the following: (6) The previous 12 hr intensity change (t=0 minus t=-12h max wind) (7) The latitude along the OFCI track (8) The SST along the OFCI track, (9) The 850-200 hPa shear along the OFCI track Predictors (6)-(9) are obtained as part of the SHIPS model diagnostics. The NNIB model is just the simple mean of the same four models used as input to NNIC. It is used as a baseline to see if the neural network can improve on the simple mean. The NNIC forecast is from a fully connected neural network with 1 input layer (9 nodes), 1 hidden layer (9 nodes) and 1 output layer (1 output per forecast time). The network was trained on data from 2013-2020, with a separate network for each time from 12 to 168 hr. The Atlantic and combined eastern and central Pacific were trained separately. At least 2 of the 4 input models must be available for NNIC to make a forecast. Because HWFI only provides a forecast to 120 hr, the longer range forecasts are less reliable. NNIC is experimental and is undergoing its first end to end test in 2021 where the output is being put in the public A-decks. Caution should be exercised downloading this model. I've also just updated the nhc_techlist.dat file on the FTP server - so you can see that AP21-30 are indeed the additional GEFS members. I'll try and get the deprecated techlist updated in the next few weeks, but if you see something that hasn't been used for several years, it's a pretty safe bet to assume it is now deprecated. |