NMME

North American Multi-Model Ensemble

NMME (North-American Multi-Model Ensemble) is to improve intra-seasonal to interannual (ISI) operational predictions based on the leading US and Canada climate models.

Monthly Teleconferences               2018

November 1, 2018  Sea surface height (SSH) has been used to measure the ocean currents that move heat around the globe as a critical component of Earth's climate. Its relationships with SST, salinity, tides, waves, and the atmospheric pressure loading patterns are of great interest of climate forecasters. To promote R2O activities for service improvement, the NMME project is planning to make its SSH hindcast data available to the community. In the teleconference this month, Dr. Bill Merryfield of the Canadian Centre for Climate Modelling and Analysis (CCCma) was invited to give a talk on his study of the verification of CCCma Coupled Climate Model, versions 3 and 4 (CanCM3/4) SSH hindcasts using multiple ocean reanalyses, i.e. Ocean ReAnalysis System 4 (ORAS4, ECMWF), Global Ocean Data Assimilation System (GODAS, NCEP), Cimate Forecast System Reanalysis (CFSR, NCEP) and German Estimating the Circulation and Climate of the Ocean, version 2 (GECCO2, University of Hamburg). His results showed hindcast skill of CanCM3/4 SSH was generally high (comparable to that for SST), although large differences existed depending on region and verification reanalysis product used, particularly in the Atlantic & Southern Oceans. Among the four reanalyses, the spatial mean skill was the highest for GODAS; this relationship may be at least partially attributed to the use of GODAS ocean temperatures to initialize CanCM3/4 hindcasts, as well as the lack of Arctic data in GODAS. Dr. Merryfield's study revealed skill may be the highest when using the multi-reanalysis mean for verification. Inconsistencies were found between reanalyses in global SSH trends; most of the reanalyses lack any realistic trend. Further work to improve skill by replacing global trend with observed trends are in progress.

September 6, 2018  Discussions continued on the hot topic of the 2015-16 El Niņo event and the prediction of California rainfall. The Arctic Oscillation (AO) impact is critical to improving middle- to high-latitude climate outlooks, and most models did not catch the positive AO observed during that season. Michelle L’Heureux, Lead of the El Niņo-Southern Oscillation (ENSO) team, NOAA’s Climate Prediction Center did a more careful study, looking inside the complex problem. She demonstrated a strong correlation (r~0.9) between ensemble mean AO and ENSO values out to long lead times of 12 months in the NMME products, much stronger than the correlation in observations. Though her studies revealed the NMME has some skill in predicting the AO out to 5-7 seasons for winter and early spring, she noted that not all the AO skill (forecast vs. observations) can be attributed to ENSO, especially at short lead times. While ENSO can impact the winter AO/NAO, the predictable signal was smaller than the unpredictable noise, suggesting the NMME is overdoing the AO-ENSO relationship, especially during the spring and fall. This sensitivity may arise from the models underrepresenting or missing other sources of predictability. The details were explored in her paper with coauthors (L’Heureux et al. 2017).

Reference

L’Heureux, M. L., M. K. Tippett, A. Kumar, A. H. Butler, L. M. Ciasto, Q. Ding, , K. J. Harnos, and N. C. Johnson, 2017: Strong relations between ENSO and the Arctic Oscillation in the North American Multimodel Ensemble. Geophys. Res. Lett., 44, 11,654–11,662. https://doi.org/10.1002/2017GL074854

August 2, 2018  The ongoing discussion of the difficult prediction of anomalous dry conditions over Southern California during the 2015/16 El Niņo winter was continued by Dr. Young-Kwon Lim of the Global Modeling and Assimilation Office, NASA Goddard Space Flight Center. Earlier contributions to this discussion were provided by Arun Kumar (4/5/18) and Huug van den Dool (6/14/18). In addition to finding that the spatial distribution of the tropical Pacific warming in 2015/16 was not a main cause of the negative precipitation anomalies in Southern California (agreeing with previous researchers), Dr. Lim showed that the atmospheric response to the Northeast Pacific warm water blob (WWB) SST could decrease precipitation in the Southwestern U. S. region, and the model response to the WWB was not sufficient to fully overcome the relatively large El Niņo-driven positive precipitation anomalies. Regarding the role played by internal atmospheric variability, further regressing the intra-ensemble variance of the precipitation against the leading height modes produced precipitation anomalies that did much to close the gap between the observed and ensemble mean response, especially in the Southwest where the AO/NAO-like leading noise pattern was seen to play a key role. No NMME models predicted the negative precipitation anomalies over Southern California; rather, all of them displayed a strong and predictable response to the tropical Pacific SST forcing. “It may not represent a failure of the forecasts, but a failure to adequately provide the community with a quantifiable and understandable measure of the uncertainty in the prediction”, Dr. Lim said in conclusion.

July 12, 2018  “Since skillful subseasonal predictions of Northern Hemisphere cold season extratropical weather and extreme events are linked to the Arctic Oscillation/Northern Annular Mode (AO/NAM), it is desirable to predict the state of the stratospheric polar vortex, and thus potentially the surface temperature and storm track patterns out 20-40+ days into the future.” Prof. Jason Furtado of the School of Meteorology, University of Oklahoma gave a presentation on quantifying fundamental characteristics of AO/NAM and related predictability in the NMME Phase-2 models, and identifying models biases in the development and subsequent impact of major sudden stratospheric warmings (SSW). His research, using the three NMME Phase-II models with sufficient stratospheric data (CanCM3, CanCM4, and NCAR-CCSM4) showed that: 1) modeled surface AO signature includes a much stronger Pacific loading center than in observations, 2) models underestimate the frequency of positive AO after approximately 7 days, 3) an observed significantly higher frequency of negative AO enduring through days 10-16 is not represented by the models, and 4) the models underestimate Northern Hemisphere polar vortex variability. Flaws were also found in wave forcings and resulting patterns. The composite of post-SSW impacts on 500 hPa geopotential height showed the North Atlantic Oscillation signature was present, but slightly eastward biased, and models show little agreement with observations in the Pacific sector. It was discussed that stratosphere-troposphere coupling biases may be tied to incorrect wave-mean flow interactions in the troposphere following major SSWs.

June 14, 2018  Dr. Huug van den Dool of NWS Climate Prediction Center continued the discussion of the "unsuccessful" forecast of 2015/16 winter Southern California precipitation, following on the NMME teleconference led by Dr. Arun Kumar in April. Calibrated probabilistic forecasts (van den Dool et al. 2017), show improved reliability and resolution, but not predictability, which could be attributable to model deficiencies and shortcoming of the system such as signals captured by the models and the relatively short hindcast period. More in-depth thinking was explored, including (1) that the Southern California rainfall climate might be odd (intermittent, seasonal and non-Gaussian) and marginal, and the NMME models may not accurately represent the current Southern California climate; (2) interdecadal variations in ENSO/mid-latitude correlation could be poorly captured by the models; (3) eddy-mean flow interactions could impact the jet extension that reached the west coast of U.S. (e.g. 1982/83 vs. 2015/16); and (4) the noise could be changing, and may act "in our favor" on some occasions. An intense program of large-ensemble and shorter range forecasts was recommended for more insightful studies on the minority of model members that captured more accurate patterns for 2015/16. Lastly, Dr. van den Dool raised the question of how to make our probability forecasts a more effective and correct message in order to better serve our users. CPC forecasters are not predicting probabilities, but rather use probabilities to express uncertainty in the forecast (in lieu of error bars). Recalling the AMS statement from the 1980s on ethical standards for weather forecasting, “a forecast has to be unambiguous, reproducible and verifiable”, Dr. van den Dool made the point that a forecast should be correct or not correct, nothing in between.

April 5, 2018  Dr. Arun Kumar, Principal Scientist at NOAA Climate Prediction Center, gave a presentation on a research challenge on the predictability of 2015/16 winter precipitation anomaly over the west coast of the U. S., where the observation was opposite to the mean El Niņo signal. Key research questions were raised. 1) Were the differences due to unpredictable noise having an influence on individual seasonal mean? 2) Were the differences due to changes in atmospheric response to differences in ENSO SSTs, atmospheric response to other boundary forcing, or changes in ENSO teleconnections in a changing climate? Were those factors predictable? Using CFSv2 hindcasts (1982-2011) and real-time forecasts (2012-2015), Dr. Kumar demonstrated the model forecasts with DJF 2015/16 SST forcings were consistent with historical expectations; the contribution from noise could lead to subtle changes in circulation and could appreciably change seasonal mean precipitation outcomes from the “expected response”. For better understanding, there are further questions that need to be answered, e.g. 1) What is the PDF of seasonal mean atmospheric states during different El Niņo conditions? 2) How does the “response and noise” vary from one event to another? 3) How predictable are the variations in SSTs themselves? “Though we know the approach, i.e. ensemble of GCM simulations with multiple models to pursue attribution studies, we don’t know how to build confidence in answering some of the questions and bringing them to a closure. With such a view, the use of probabilistic forecasts in decision making on an individual forecast basis is a hard dilemma to come to grips with”, said Dr. Kumar.

Archives   2015, 2016. 2017