Monthly Teleconferences 2016 December 8, 2016 Kirstin Harnos of the Climate Prediction Center presented "NMME Sea Ice Reforecasts: An Update", work that she has done with Michelle L'Heureux and Qin Zhang, both of CPC. Kirstin's team has been assessing the general question "how well does the NMME handle sea ice?" As a first check, they confirmed that the five participating models (NCEP-CFSv2, NCAR-CCSM4, CMC-CanCM3 & 4, and GFDL-FLORb01) produce more ice in March than in September. With this basic quality assessed, they examined the representation of both the total sea ice extent (SIE) and the year-to-year variability in the SIE, during the common hindcast period of 1982-2010. Since the trend in SIE is very strong over the hindcast period, but may be non-linear and is difficult to identify, examining these two quantities allows for an understanding of how the models and the NMME represent the trend (total SIE) and how they represent interannual processes that may be independent of the trend. The individual models generally either overestimate or underestimate the total SIE, with the exception of the CanCM4, which has a fairly realistic simulation; the NMME multi-model ensemble mean is an improvement over any of the individual models. The NMME has the greatest improvement over the individual models in total bias and in the anomaly correlation of the year-to-year variability. The models, including the NMME, do not capture the recent trend well. However, the observed year-to-year variability is within the envelope of model simulations. (By Emily Becker) November 3, 2016 Michelle L'Heureux of Climate Prediction Center gave a review of NMME prediction of the 2015-2016 El Niņo in the November NMME teleconference. Overall, the official CPC/IRI forecasts captured both the strength and the timing of the event. The development of El Niņo was declared in March 2015, followed by the official forecast of a potentially strong event broadcast in June 2015 and further alert of the seasonal Niņo-3.4 SST exceeding 2.0°C, a top 3 event in history, issued in early August 2015. A final El Niņo Advisory with the return of ENSO-neutral condition was sent out in early June 2016. Apparent CFSR cold biases in the tropical Atlantic Ocean impacted CSFv2 forecasts and consequent skill scores during the 2015-16 event. The "target period slippage" problem, i.e. the model forecasts were slow to transition into and out of ENSO event, a long-standing problem, was prominent in forecasts by both NMME/dynamical models and statistical models from the IRI/CPC plumes. It appeared to affect longest lead forecasts the most and be more severe during lower skill/lower variability ENSO periods. Despite the difficulties, North American Multi-Model ensemble (NMME) and other multi-model plumes were valuable, providing a backstop for ENSO forecasts. September 8, 2016 The teleconference had two presentations on recent progresses in research and operation. Dr. Kathy Pegion of GMU began her talk with a question, “Which predictability estimates are most realistic?”, then gave a fidelity assessment of predictability estimates using metrics of spread/error and autocorrelation. The results revealed a common problem of overestimation of signals of Niņo 3.4 SST and US temperature and precipitation, and showed the skill limit of Niņo 3.4 prediction at 3-months lead had almost been reached. Dr. Huug van den Dool of NCEP/CPC, taking NAO prediction as an example, gave an update on the unequal weights project of NMME, in which Community Earth System Model (CESM) had been added recently. It showed using distributional information could help weighting relative to using ensemble means, but it was tough to calculate trustworthy weights, or to beat equal weights or skill-based weights. Some hopes were seen in special solutions and sub-sampling by deleting certain models ‘upfront’ with negative weights. May 5, 2016 A progress report on skill and bias evaluations of the NCAR Community Earth System Model (CESM 1.0) was presented by Dr. Huug van den Dool with co-investigator Dr. Li-Chuan Chen of NCEP/CPC in the May NMME teleconference. The simple metrics, anomaly correlation, was used and the assessments were made from various angles, e.g. as functions of prediction lead-time, start/target month, relationship between skill and fidelity etc. The comparison of NMME performances with and without CESM, and using CESM compared to CCSM3 were also conducted. The bottom line showed CESM 1.0 with no glaring errors was slightly better than CCSM3. Though a marginal improvement obtained, it is a good sign for CESM to get on board, since CESM 1.0 is more advanced than its predecessor CCSMs with new physical and chemical climate system components. Continuing collaboration to push NMME to a new height was earnestly discussed. Following the telecom agenda, Dr. Qin Zhang of CPC gave a brief talk about the effort of collecting sea ice model output from the NMME models, which is part of the CPO/CVP funded project of understanding Arctic sea ice mechanisms and predictability. March 10, 2016 Dr. Wanqiu Wang of Climate Prediction Center (CPC) gave a talk on the CPC experimental sea ice forecast. The forecast skill assessment showed CFSv2 performance was comparable to other dynamical forecast systems but much less skillful than Lamont Markov statistical model, which could be taken as a reference for the dynamical model improvement. Much room for improvement was demonstrated by comparison of prediction skill (3-5 months) versus potential predictability (4-12 months). Efforts were made focusing on 1) the sea ice concentration errors in initialization, and 2) the model bias, especially on excessive surface downward solar radiation flux related to the negative bias in cloud amount. The results were promising, setting a high bar for NMME. The telecon continued with a progress report by Drs. Joe Tribbia and Huug van den Dool on i) NCAR Community Earth System Model version 1.0 (CESM1.0, a potential new member to the NMME family) hindcast skill assessment and ii) timeliness of NMME real-time forecasts. It followed discussions on possibility to make the NMME daily forecasts accessible from NOAA's National Centers for Environmental Information (NOAA/NCEI) for user needs. February 4, 2016 Dr. Reinel Sospedra-Alfonso of Canadian Centre for Climate Modeling and Analysis, Environment Canada gave a talk on potential and actual predictability of snow in the Canadian Seasonal to Interannual Prediction System (CanSIPS) in the 2016 February NMME teleconference, showing skillful forecast of Snow Water Equivalent (SWE) with appreciable actual skill compared to potential skill, and the dependence of values and duration of Potential Predictability (PP) on the timing of forecast initialization. It was demonstrated that current CanSIPS SWE initialization was reasonably good, and improved SWE initializations (e.g., initialized directly from an accurate analysis) would improve actual skill. This month NMME teleconference also discussed possibility of expanding the NMME Phase-I real-time forecast variables. The forecast skill was a concern. Thus, skill assessment would be important for adding new variables to the real-time production suite. |
