The two lows that crossed central New Zealand in the second week of October brought much-needed rain to many areas suffering from drought. Blenheim Airport received 94 mm of rain over six days, when the average for October is only 66 mm. In September, Blenheim Airport had just 11 mm of rain—the third-lowest September rainfall on record.
In fact, September was the driest on record in many places in the south-west of the North Island, including Wanganui, Palmerston North Airport, Levin, Paraparaumu Airport and Wellington Airport. All these places received good rain from the early October storms, most getting more than the monthly average.
Among the unusual side effects of the rain was the discovery of some fossils on a farm near Otaki. The farmer, Barry Mansell, had found a large fossilised shell in the Otaki Gorge area around 40 years ago, so when he came across a fresh slip on the road he tried his luck again. After five minutes searching he found half a dozen rocks containing fossil shells the size of scallops.
Of course, the storms were not entirely beneficial. Some roofs were lifted in Wellington and a few trees came down, cutting power lines here and there. The Cook Strait ferries stopped sailing, and wet roads contributed to a spate of accidents. But on the whole the storms were a blessing for the rain they brought.
How far ahead can forecasters see such rain-bearing lows coming our way? The global computer models that are one of the mainstays of modern forecasting are now very good out to two or three days into the future and useful out to four or five days ahead. But by five days ahead the computer predictions of the locations of low centres can be in error by 1000 km, and the intensity of a low can be out by 30 hPa.
These errors come from inaccuracies in the initial measurement of weather conditions for the start of a computer model’s run, as well as approximations in the equations used to describe atmospheric processes.
Initially, errors grow in a linear fashion, so that after two days the errors are twice as big as after one day. But after more time has elapsed the errors tend to grow much faster, in a non-linear fashion.
For example, a small error in the measurement of wind speed and direction in the jet-stream at 10 km above the Tasman Sea can affect whether a computer model deepens a low or not. If the low is deepened, it will change the structure of the jet-stream in a way that further enhances the deepening of the low for a period of 12 hours or so. If the computer model triggers a feedback process like this, it will quickly send its predictions down the wrong path. Equally, if the computer model misses such a development it will also end up way off track.
Lack of observations to correctly describe the initial state of the atmosphere at the start of a computer run has always been a major problem in the southern hemisphere. Much of New Zealand’s weather comes from over the southern Indian Ocean and the seas south of Australia, where there are almost no conventional observations—just a ship report now and again, plus a handful of drifting buoys and the weather station on Kerguelen Island, more than 3000 km southwest of Perth.
This observation gap has diminished significantly in recent years as more sophisticated observing instruments have been put on the newer weather satellites. Some of these measure upper-level winds by tracking the movement of high clouds. Others measure surface wind speeds by measuring capillary waves on the ocean surface. Air temperature and humidity are measured more accurately and over a finer scale both vertically and horizontally than they used to be.
The benefits of this flood of new measurements have been seen in improvements to the five-day forecasts over the past decade, but the problem of rapid error growth in the models remains beyond about five days. Consequently, researchers have been working on a new way to handle the problem of error growth, known as ensemble forecasting.
Making use of the tremendous power of newer computers to handle millions of calculations in the blink of an eye, researchers now run a computer model many times for the same situation, using slightly different initial conditions for each run. The variations in initial conditions are all in agreement with the observations but represent different possibilities in the initial atmospheric conditions, given the uncertainties between the points for which there are observations.
There are two tactics involved in choosing the range of variations in the initial conditions for the different model runs. One is to have the most variations over data-sparse areas, such as the Indian Ocean, while the other is to concentrate the variations around an important weather system such as a developing low-pressure centre or jet-stream, wherever they happen to be on the day.
Once the computer has completed all the runs, the results can be compared. If over one area, such as New Zealand, they produce a similar pattern of isobars, meteorologists can have high confidence in the result. If, instead, the weather patterns are widely different, we have low confidence in the forecast.
One way to present the ensemble results is known as a spaghetti diagram, where the same one or two isobars are plotted from each of the model runs. An example of a spaghetti diagram, of the 500 hPa map for midday on October 10, appears on page 8. Twenty-three different model runs are shown, each based on an analysis time of midday October 4 and run for six days ahead.
The point to note from this example is how much spread there is in the lines from about South America around to western Australia, in contrast to lines in the Tasman Sea—New Zealand area, which are quite close together. Closely spaced lines indicate that there is considerable agreement in all model runs that there will be a large trough of low pressure over New Zealand. The widely spaced lines in the area from South America to Australia indicate much more uncertainty as to what will happen there.
The amount of variation between individual runs in an ensemble can be used to give a measure of confidence in the forecast and to derive a percentage probability of say 1 mm of rain or 10 mm of rain for a particular region.
Another way to present the ensemble runs is to average the results. This has been done with the surface isobars for midday October 10, shown above left, using 23 model runs out to 15 days based on a start time of midday September 25. A trough of low pressure is indicated over New Zealand and the Tasman Sea, but seems not to be intense, as it is delineated by only a single isobar.
By contrast, the analysis of what actually happened on the day, shown above right, has five isobars around the trough over New Zealand. This is typical of the problems the forecaster faces when considering the ensemble products, as deep lows 10 days or more ahead will always be “blurred” to look like shallow features.
Ensemble forecasts can also be run at very high resolution over much shorter time periods. In the United States, research is going on to try to use ensembles to forecast the individual thunderstorms that give rise to tornadoes.
Ensembles do not necessarily involve computers. The forecast for the Allies’ D-Day landing in France in June 1944 (see New Zealand Geographic, Issue 22) was the work of three independent forecasting teams. One team was from the British Meteorological Office, another from the Royal Navy, and the third from the United States Army Air Force. Their results were then coordinated via a conference call, and a single forecast agreed on.
The best ideas have often been around for a long time!