RECENT ACTIVITIES - OSTI R2O

November 7, 2022  The November WPO–OSTI Webinar for Weeks 3-4/S2S was held with two speakers. Dr. Judah Cohen from Atmospheric and Environmental Research (AER) presented his work with collaborators of applying machine learning (ML) to improve S2S forecasts. By applying the adaptive bias correction (ABC) method that combines state-of-the-art dynamical forecasts with observations using ML, he demonstrated that ABC clearly improved temperature and precipitation forecasting skill in CONUS. To explain ABC skill gains these improvements were coupled with a practical work-flow based on cohort Shapley, which further identified higher-skill windows of opportunity at specific climate conditions. The second presentation was given by Dr. Dan Collins of CPC who gave an overview of the success of the Bayesian Joint Probability (BJP) calibration in improving skill and reliability over other calibration methods. The BJP methodology is currently used in an experimental setting for CPC subseasonal temperature forecast, aiming to be transition to operation and extended to subseasonal precipitation prediction. The video record of the webinar and the presentation slides are available online at https://vlab.noaa.gov/web/weeks-3-4-s2s-webinar-series.

September 12, 2022  The September OAR/CPO-NWS/OSTI Weeks 3-4/S2S Webinar was held on 9/12. Two presentations revolved around the research topic of stratosphere-troposphere coupling, an important source of sub-seasonal predictive skill for tropospheric temperature and precipitation. Dillon Elsbury of CIRES/NOAA Chemical Sciences Laboratory reported the research progress of his team in developing a diagnostic toolbox to access the capability of NOAA’s GSFSv12 to represent the dynamical coupling between the stratosphere and troposphere. Using the toolbox he demonstrated GEFSv12 had stratospheric biases similar to other forecast systems. It underestimated the amplitude of the QBO, particularly its easterlies, and showed that MJO skill was higher during easterly QBO. The predictive skills of Sudden Stratospheric Warmings (SSWs) were demonstrated out to two weeks, and that of NAO were higher when reforecasts were initialized with strong/weak polar vortex. This toolbox is being implemented at CPC for real-time verification (Fig. 1). The second presentation by Kai Huang of George Mason University applied a stratospheric zonal-mean nudging in a subseasonal prediction system and revealed that the freely-evolving zonal wind anomalies as a result of the thermal wind balance to the nudged QBO temperature were likely to be the key for capturing the observed QBO-MJO connection. The video record of the webinar and the presentation slides are available online at https://vlab.noaa.gov/web/weeks-3-4-s2s-webinar-series.

August 8, 2022  The August WPO-OSTI Weeks 3-4/S2S Webinar was given by Yuejian Zhu of NCEP/EMC. It had two topics: 1) Development of Unified Forecast System (UFS) Coupled Global Ensemble Forecast System (GEFS) for Weather and Subseasonal Forecasts, and 2) ENSO Predictions from the Preliminary UFS Seasonal Forecast System. Based on the statistics and diagnostics of the results, the first part of the presentation concluded that by testing the different horizontal resolutions, the coupled GEFS overall extended prediction skill from current operational GEFS with a caveat that insufficient members may not represent full uncertainties. The second part showed some promise that the preliminary UFS Seasonal Forecast System could perform better than CFSv2 in ENSO prediction. Problems were also found, such as the ensemble spread of Niño 3.4 SST is much larger than the root mean square (RMS) error. The video record of the webinar and the presentation slides are available online at https://vlab.noaa.gov/web/weeks-3-4-s2s-webinar-series.

July 30, 2022  The third quarterly review of the FY22 CPC-STI projects was held virtually. The CPC project managers reported what had been accomplished and the issues/challenges that needed to be overcome. Among the promising results reported, the Arctic sea ice prediction by UFS improves over that by CFSm5 and CFSv2 for all seasons, and is even better than or comparable to that by MMEs. The cause of inaccuracy in GEFSv12 reanalysis soil moisture is found, bringing much hope for remedy.

June 22, 2022  Dr. Robert West of Northern Gulf Institute and NOAA/AOML gave a talk entitled “Interbasin SST as a Predictor of Seasonal Atlantic Hurricane Activity” in an NMME teleconference. By exploring the seasonality of Pacific and Atlantic sea surface temperature (SST) contributions to Atlantic hurricane activity, he demonstrated that early-season (JJA) hurricane activity is largely influenced by the sea surface temperature anomalies (SSTAs) in the tropical Atlantic main development region (MDR), but the tropical Pacific (Niño 3) SSTAs are equally important in the late-season (SON). His analysis using an MDR-Niño 3 interbasin index derived from hindcasts of the North American Multi-Model Ensemble (NMME) with May initial conditions further revealed increased predictability of high-impact tropical cyclone (TC) activity in JJA, SON and JJASON. Because NMME prediction skill of MDR SSTAs is lower than that of Niño 3 SSTAs, increasing prediction skill for MDR SSTAs would be the key to improving seasonal hurricane outlooks. The follow-up discussions focused on an interesting result showing the relative operating characteristic (ROC) score of accumulated cyclone energy (ACE) index forecasts by the interbasin SST index using NMME super ensemble SST predictors is not superior to that using NMME model member SST predictions (Fig. 1). It was suggested that the method of calculating NMME probabilistic forecasts required some further investigation, and that sampling may be an issue for the early season statistics.

May 3, 2022  The OSTI-WPO monthly webinar for Weeks 3-4 and S2S was held on 5/3. First, Dr. Kathleen Pegion of George Mason University shared her recent research experience in understanding predictability of daily Southeast US (SEUS) precipitation using explainable machine learning. Her results showed that using Indices of large-scale climate phenomena (e.g., NAO, AMO, PDO, ENSO, MJO, etc.) is insufficient to skillfully or reliably predict the sign of daily precipitation anomalies in the SEUS. Using a convolutional neural network (CNN) and gridded fields as predictors gives more reliable and accurate results. Further applying explainable machine learning helps to identify which variables and grid points of the input fields are most relevant for confident and correct predictions. Her study revealed that the local circulation is most important as represented by maximum relevance of 850hPa geopotential heights and zonal winds to making skillful, high probability predictions; and the corresponding composite anomalies identify connections with the El-Niño Southern Oscillation during winter and the Atlantic Multidecadal Oscillation (AMO) and North Atlantic Subtropical High (NASH) during summer. Next, Dr. Tara Jensen from NCAR gave a presentation on recent enhancements to METplus, a comprehensive verification and diagnostic software package, which is available now to the Unified Forecast System (UFS) developers and users via the Developmental Testbed Center (DTC). Recent emphasis has been on the including Subseasonal to Seasonal (S2S) diagnostics to look at processes associated with Madden-Julien Oscillation, Sudden Stratospheric Warming (SSW), and El Nino-Southern Oscillation (ENSO), along with adding in ensemble verification capability deemed necessary by the Environmental Modeling Center, Climate Prediction Center, and Ocean Prediction Center. Some of the challenges were discussed to ensure the S2S additions being able to move into operations. The video record of the webinar and the presentation slides is available online at https://vlab.noaa.gov/web/weeks-3-4-s2s-webinar-series.

March 23, 2022  Dr. Kai-Chih Tseng from NOAA GFDL was invited to present his research progress on the possibity of multiseasonal forecasts of atmospheric rivers (AR) in an NMME teleconference. His research explored the dominant predictability sources and challenges, and showed the existing potential of multiseasonal AR frequency forecasts with predictive skills 9 months in advance.

February 23, 2022  In this month's NMME teleconference, Michelle L. L'Heureux of CPC was invites to talk about the prediction challenges associated with errors in linear trends of tropical Pacific SST. She showed NMME continues to have positive SST trend errors in the eastern Pacific Ocean, which is strongly related to errors in the amplitude of precipitation anomalies in the Central Pacific. SST trend errors may partially explain an increase in El Niño false alarms in recent years.

February 14, 2022  Professor Saravanan of Texas A&M University gave a seminar on Predicting the unprecedented event at UMD ESSIC. In his talk, as increasing of the model complexity the model “reducibility limit” is discussed, which tries to explain the recent shift by IPCC towards relying more on observational assessments of climate sensitivity.

January 31, 2022  Professor Elizabeth Barnes of Colorado State University gave a seminar entitled “Viewing Anthropogenic Change Through an AI Lens” at UMD ESSIC. She demonstrated 1) how the explainable AI (XAI) techniques can help to visualize and quantify the footprint of human activity across the global land surface in near-real time, 2) the utility of XAI for extracting forced climate patterns through time amidst a sea of climate noise and model disagreement, and 3) how XAI can help to detect mechanisms of temporal variations in decadal warming trends. In closing, Prof. Barnes envisioned that the explainable machine learning is a game changer for scientific research. AI can incorporate science, or the science can be what the AI learns.

Past Activities:  2021