Thursday, August 7, 2014

Long Range Models Likely Incorrect in Winter Outlooks

After extensive analysis of multiple long range climate models, it looks like the consensus is for a moderate to strong El Nino on all guidance members, something that is not expected to occur, and thus rendering their forecasts incorrect.

NMME/CPC
The image above shows sea surface temperature anomaly forecasts from eight different long range models, all valid for the month of August 2014. These forecasts were made in July 2014, meaning this is only the first month forecast. Even though it's only one month out, we still see incredible inconsistencies. Starting along the top row, the CFSv2 model has warmer than normal SST anomalies persisting to the north and south of the Equator, with neutral anomalies along the Equator itself. This is a reasonable scenario, and we will examine this model further later on in this post. The CMC1 model allows a moderate to strong El Nino to develop, projecting SST anomalies of 1º to 2º C above normal (dark red) to form. Closer analysis indicates anomalies of over 2º C are actually present, seen in the dark brown. Checking out the CMC2 model, a relative of the CMC1, we still see moderate to strong El Nino conditions present per the SST anomalies, something highly unlikely to happen in the next ~25 days. Lastly, analyzing the GFDL_FLOR model, we once again find well above normal anomalies present along nearly the entire Equatorial Pacific region, from Ecuador to Australia. Aside from the well above normal anomalies, full-basin coverage of the El Nino is not anticipated, further degrading this model's credibility. Out of the four top-row models, only the CFSv2 model seems somewhat reasonable right now, and even that assertion is debatable.

Moving on to the bottom row, we come upon the GFDL model, a close relative of the GFDL_FLOR model. As such, the GFDL model continues the above normal sea surface temperature anomaly trend, though not as intense as many of the top-row guidance. This makes me think the GFDL may be on to something worth watching, and this will be discussed later on in the post. The NCAR model initiates a likely-Strong El Nino in the next couple weeks, something that will not be happening. The NCAR_CCSM4 model follows up with a nearly-identical scenario, so we can toss that solution as well. Lastly, the NASA model shows a moderate El Nino in place for the month of August. Because it's not such an extreme solution, we'll go over it further down the road.

After analyzing these models, we have found three of the eight models - CFSv2, GFDL, and NASA - to have somewhat reasonable scenarios. That means only ~38% of these forecasts have a chance to verify. Think about that- over half of these models can't even get the forecast right in the first month. It's pitiful, to some degree, but that's why we have scientists working to improve them.

Let's now analyze a sea surface temperature anomaly forecast for 3 months out, rather than 1 month out, to see how well our three aforementioned models are doing.

NMME/CPC
Let's now analyze the CFSv2 model in the top-left corner. In the forecast month of October, the CFSv2 has a Central Pacific-based moderate El Nino, possibly nearing Strong El Nino-strength. Unfortunately, considering conditions seem favorable more for a weak El Nino (MAYBE up to a moderate El Nino), we'll have to toss this forecast model.

Analyzing our second chosen climate model, the GFDL, brings about mixed results. On one hand, we see the forecast for a likely-moderate strength El Nino, something not too far from the realm of possibility. However, the El Nino is based in the west-central Pacific, something not showing up on other forecast models. For that reason, it may be wise to toss this forecast model as well.

Our last model of hope, the NASA model, shows unfortunate circumstances for October 2014. The model induces a moderate to strong El Nino across nearly the entire Pacific basin, something highly unlikely to happen. We can discount this forecast, too. Even lightly auditing the first-month forecast models can't save our chosen three models from the poor forecast ability just three months out.

Why do I think this is important? Many long range guidance models are calling for a warm United States winter, possibly accompanied by a cool South Plains at times. However, when you look at the SST forecast, it becomes apparent that the forecast cannot be trusted, due to a very slim chance of the SST anomaly outlook actually verifying. Thus, don't take the predominantly warm forecasts as they are unless verifying the SST outlook; the same goes for predominantly cold winter outlooks, as their SST forecasts must be verified as well.

Andrew