By Mike Godsey

by Mike Godsey and Matt Sounders

Have you ever noticed that meteorologists are usually pretty accurate about the temperature and the amount of sunshine but are so often way off when it comes to wind?

Look at this graphic of the weather projections for today Sunday, May 28 by 10 major computer models for Treasure Island. Since Treasure Island is in the middle of San Francisco Bay nothing around it for thousands of feet you would think it would be a simple forecast.

And you would be partially right.

Look how all 10 of the models are in total agreement for the amount of sunshine to Treasure Island today. And even 3 days out they are almost all saying the same thing. See how easy forecasting is!

Now, look at the temperature forecasts for all 10 models. Unbelievably all 10 models are within 6 degrees of each other not only for today but also for Monday. Only for Tuesday do we start to see one model depart from the consensus (and that one, the NAM3 is probably better at forecasting the likely fog that will drop Tuesday’s max temperature.

Now let’s see the WIND forecast for all 10 models at the bottom of the graphic. Wow! Lots of bickering amongst the models. AND almost surely they are all reading low compared to the likely winds today and for the next several days. This is the sort of mess we at are faced with for every forecast for every region of the coastal USA.

So first why are sunshine and temperature so “easy” to forecast while winds are so very difficult.

Time to get a bit technical:

Compared to wind forecasting, making the right call for high and low temperatures is relatively simple. While there are many variables that influence the temperatures we experience here on the ground, an overly
simplistic first guess depends on two main factors: the temperature and pressure about 5000 feet above the ground, and clouds and moisture.

The temperature and pressure at 5000 feet are the largest pieces because air generally nearly instantly mixes to an equilibrium state and we know how pressure and temperature are linked in the lower atmosphere. On short time scales, our models don’t have as much trouble forecasting temperature and pressure 5000 feet up because friction and surface conditions play far less of a role and the atmosphere becomes simpler to model.

Second, we need to know something about moisture and cloud cover, since clouds and humidity tend to suppress the effects of solar heating and slow the release of heat at the surface at night. Models do have more trouble predicting cloud cover, but meteorologists develop pattern recognition and can make a very solid educated guess most of the time.

In most parts of the country forecast temperatures are very accurate but if you live near a coast where the marine layer clouds ebb and flows each day you know that temperatures can change very fast.

Now, how about wind forecasting.

Over the open ocean wind forecasting is relatively easy. There are no trees, buildings, mountains or valleys to create friction or to funnel or redirect the wind. Moreover, there is usually less mixing of the winds aloft with the surface winds since ocean water is so slow to heat.

But on upon hitting land, the wind is dealing with solid 3D topography that can locally accelerate or decelerate or block or redirect the wind on a very local basis. In this animation, you can see WNW ocean wind streaming toward the coast from a constant direction and velocity.

The “O” marks show some of the areas where this relatively constant direction/velocity ocean winds suddenly change in velocity and/or direction as it hits different types of topography. At the top of the image, you can see a one-sided venturi west of the Golden Gate and a sudden curve of the winds towards a distant low-pressure. There is a similar venturi at the bottom of the image.

Worse the inbound wind does not even have to reach the topography to change. In the middle of the image notice the famed Montara “Wind hole” where the relatively strong ocean winds suddenly weaken a 1/2 mile before land. This occurs as all the inbound air stacks up against the Coast Range in the higher Santa Cruz Mountains. This creates a high-pressure bubble that builds out to sea forming an invisible barrier to the inbound wind which is then redirected to the north and south.

Now all this looks complex but there is a bigger problem. This type of depiction of the wind by the models is usually not very accurate. Why?

1. None of the models are high enough resolution to pick up details of the topography.

2. Slight and rapid changes in the distribution and thickness of fog and clouds can rapidly change the local wind strength.

3. The direction of the curving you see in the image can change fast as the location of inland low-pressure areas change.

4. Let’s not even talk about the unpredictability of eddies that may form along the coast.

And, to keep things simple we are leaving out the interaction between the modeled surface winds you see in the animation and the winds just aloft that often have a different direction and velocity. And these winds aloft can impact the gust factor, velocity and wind quality of the surface winds.

All of this is why a human wind forecaster is still needed. Despite advances in models and Artificial Intelligence the pattern recognition skill of an experienced forecaster is still needed who using that knowledge, cam and satellite images can produce a more useful forecast. is busy working on a machine learning system that will take in some of that information and our sensor data to tweak the model output so forecasts can improve but we still anticipate the need for a human forecaster for a while.