Your weather forecast

Most of us have some access to weather information — a weatherfax-type chart, text or voice forecast or a GRIB forecast. Rather than creating your own forecast from scratch, try starting with a reasonably good forecast (from a model or a forecaster), then tweak that forecast appropriately, based on your understanding of both the limitations of weather forecasts and models, and local influences on weather.

Forecasters base their predictions on one or more computer weather models, so you should understand how models work. Troves of environmental sensing data are continuously gathered by numerous satellites, balloons, aircraft, ships and other vessels at sea, buoys, airports and other land-based sensing stations. At scheduled times, the amassed data is fed into supercomputers that run complex calculations to arrive at numerical weather forecasts.

One often-used model, NOAA’s Global Forecasting System (GFS), specifies a numerical forecast for nearly 100 parameters (such as temperature, pressure, moisture and many more) at as many as 64 different layers (altitudes of the atmosphere) at 80 different time periods into the future (out to 16 days), for nearly 300,000 locations on Earth (every 0.5 degree of latitude and longitude). The GFS model completes this process four times each day.

The result is a huge file, several gigabytes in size. Many services slice and dice this file, and send us only what we need, often only 10 kb or 20 kb of data, which passes easily over most communications methods, and is displayed with special GRIB viewing software on a computer.

Any good forecast offers sufficient temporal and spatial resolution (frequent and closely spaced forecast points), and detail and specificity so you can make good tactical decisions. In my opinion, weatherfax-type charts and forecasts prepared for consumption by the general public lack many of these attributes.

Weatherfax-type charts offer three or four snapshot forecasts from now through 96 hours, with only a few forecast points predicting averaged conditions over a large region at a specific time. Most text or voice forecasts either cover such a large area that they lack specificity, are intended for a general audience with little understanding of weather and/or predict averaged conditions. For that 96-hour period, a GFS GRIB forecast offers 33 forecasts, at three-hour intervals, with a forecast point within a few miles of your position. With the right tools and some practice, you can learn when and how to adjust most forecasts — I’ll offer some tips below.

Regardless of whether you’re using a weather chart, text or voice forecast or a GRIB forecast, you should start your analysis, as most forecasters and weather models do, by accumulating as much recent observational data as you can. This can be as simple as looking outside, noting the present values (and the values over time!) of the parameters of interest to you — wind speed and direction, barometric pressure, temperature of air and sea, dew point (or relative humidity if you don’t have dew point), precipitation/squalls, cloud description, etc. Use all your communication resources to accumulate as much additional observational data as you can.

Next, compare observations with the forecast for that time. How do observations differ from the forecast? Can you identify any patterns in the observations? (If an observation doesn’t fit the observed pattern, and you’re not sure why, you should give it less weight.) How do patterns in observations compare with patterns (over time and/or location) in the forecast?

In absence of localized effects (which we’ll discuss later), wind speed and direction are driven by differences in surface pressure (barometric pressure) over distance. Weather charts and GRIB displays indicate these lines of equal pressure (isobars). Wind generally flows along isobars (parallel to isobar lines), with a slight tendency to blow toward areas of lower pressure. Wind velocity is dictated by how closely isobars are spaced — tighter spacing means higher wind speeds.

Based on global models
Remember, nearly all forecasts are based on models, often the GFS, and all models have limitations. Typical distances between high and low pressure systems may be 1,000 miles or more. The GFS, with 30-mile horizontal resolution, 64 slices of vertical resolution, and three-hour temporal resolution has an excellent chance of getting the big picture correct. During the first 24 or 48 hours, the GFS has probably placed each pressure system in about the right location, within 60 to 120 miles (two to four data points), and has correctly estimated the pressure of each system (within a couple of millibars), and the resulting gradient of pressures between the systems is probably about right. But small-scale or short-term events fall between forecast points, and are often missed. Forecast errors become larger the farther into the future you look.

Good global computer models use normalizing protocols that reject extreme solutions for any data point. Extreme solutions can be caused by faulty observational data ingested by the model, or by possible but unusual combinations of factors. Hurricanes are a good example of an extreme and unusual occurrence — the GFS rejects the absurdly tight gradient near a hurricane, and predicts less severe weather than is likely to occur. Hurricane forecasters use models optimized for tropical storms, which lack such normalizing protocols. When tropical systems threaten, never rely on a computer model forecast. Always obtain a forecast from NOAA, or another trusted source.

Making adjustments
Here’s a list of adjustments I make to the GFS model (and other models where specified), and why. Many forecasts need similar or additional tweaks. If you are interested in learning more about effects on weather due to adjacent land areas, convection, pressure systems, tropical weather and the surface effects of upper-air (500 mb) weather, you should pick up a copy of my book, Coastal and Offshore Weather, the Essential Handbook, from which much of this info was excerpted.

1) Increase wind in squalls, since most forecasts are only for wind due to large-scale pressure gradient. Enhanced wind in squalls is caused by vertical downbursts of air. When downbursts hit the surface (of land or water), they don’t just stop — downbursts radiate outward, and add to or subtract from or change direction of the underlying gradient wind. More often than not, the result is an overall increase in wind speed. The best way to predict squalls is to obtain GRIB data including precipitation and lifted index (instability) parameters. Heavy precipitation, especially when accompanied by a negative Lifted Index, is very likely to contain squalls. In stable conditions, precipitation along a cold front can contain squalls, especially in cooler climates.

2) In the tropics or sub-tropics, when wind forecast exceeds 20 knots, add 5 to 10 knots to the forecast, due to enhanced evaporation leading to low-level moisture pooling, with causal or incidental vertical movement of air resulting in downbursts — especially if lifted index is negative (unstable).

3) Increase wind as much as 25 percent to 33 percent if the atmosphere is unstable ahead of a cold front. Models probably have the horizontal wind component about right, but under-forecast the vertical component of lifting air — which then sucks additional air in from surrounding areas, increasing surface wind speeds throughout a region. This is most extreme when the cold front and its low are intensifying and surface features are aligned (phased) with upper-air trough or low — but generally not unless the atmosphere is also unstable along the cold front.

4) Discount most mesoscale Low pressure systems (primarily a GFS bias): the GFS model has a tendency to overdevelop a cluster of convective thunderstorms into a small low pressure system, to maintain the small low for several days and to move it poleward but along the axis supporting thunderstorm development. The problem is GFS’s relatively coarse forecast grid and lack of convective-feedback prevention in calculations. While you might think this should be fixed, I’d argue it’s valuable to have a model that rarely under-forecasts convection — I’d rather be surprised by not having squalls than the other way around. Of course, when the mesoscale low is supported by other models, it’s probably a valid feature.

5) Guard against extreme GFS solutions for high pressure systems ushering in a cold air mass. Unless GFS is supported by other models, it’s probably too strong with high winds due to an advancing polar high. However, in the northwest Caribbean, even the GFS is almost never strong enough with wind due to a polar high (understand local anomalies, too).

6) Decrease wind on the back side of a ridge of high pressure (GFS only): GFS may have a bias toward keeping wind a bit too high on the side opposite any areas of lower pressure (a low or a trough). The backside of a ridge is typically an area with very light wind. If the GFS predicts 5-10 knots, expect less than 5 knots.

7) In almost any forecast, increase wind on the surface-convergent side of a trough or low (the side where there’s an adjacent high or ridge). Often called the “squeeze zone” or “squash zone,” models and other forecasts often normalize the unusually tight pressure gradient (similar to what global models do when under-forecasting wind in a hurricane). If there are squalls in this area, add even more wind due to downbursts.

8) Adjust for poor recent performance: If you identified a pattern to forecast errors during the last cold front, and the approaching cold front looks similar, there is a good chance the forecast will display a similar pattern of errors.

9) If you’re concerned about an upcoming weather event, and you use forecast models, try to download models a couple times each day, and verify consistency. Also, try to compare several models. If models are in close agreement, and vary little from run-to-run, forecast confidence should be high. Where there are inconsistencies, lean toward the more consistent model, or the solution supported by more models. Also refer to the NOAA forecast — NOAA does an excellent job ensuring forecast continuity, sometimes at the risk of not identifying changes soon enough. Always be prepared for the entire range of possibilities. If there is a discussion product available for your location, refer to the text discussion.

10) Account for many influences from land in coastal areas, such as:

• Light winds at night due to low-level inversion near land.
• Enhanced wind during afternoons due to sea breeze blowing toward land.
• Enhanced wind at night due to katabatic (down-slope) wind off land, especially mountains, sometimes called by a local name. These even occur along relatively flat Florida, with cool air rushing down the peninsula and off the coasts late at night and into mornings, in addition to the typical land breeze.
• More or less wind due to adiabatic (up-slope) wind, especially blowing up mountainsides.
• Enhanced wind due to compression between or funneling along land areas.
• Lighter wind in the “wind shadow” on lee side of land areas.

Observe local land effects and learn the typical anomalies over time, then compare observational data from the previous 24 hours. Unless you have reason to believe conditions have changed, apply a similar correction to the forecast as you have observed was necessary during the previous 24 hours. Otherwise, apply the typical correction.

Some computer models, such as the Navy’s COAMPS, attempt to correct for local influences. In my experience, these corrections are inconsistent, at best. I believe it’s better to use a model, such as the GFS, that gives you a gradient wind forecast, and let you adjust for known local influences.

I advised earlier to monitor observed conditions over time. In addition to major tweaking after receiving a forecast, you should make additional adjustments as subsequent observations become available. The Japanese call this kaizen, or continuous improvement, and that’s exactly how making ongoing tweaks to any forecast will impact your weather knowledge, forecasting skill, safety and enjoyment.