Forecasting the weather when you don't know - precisely - what's happening now

Wednesday, 19 July 2017

It snowed - even in places it rarely gets to, such as Tokoroa - the rain poured down, and freezing gales whipped through many areas. Winter threw a mean combination last week, but at least we knew the hits were coming.

Winter arrived in force last week with freezing winds, heavy snow to low levels and torrential rain. Forecasters were predicting the storm well ahead of time, and now they're predicting heavy rain and severe gales for the end of this week. But even in these days of giant supercomputers, the forecasters don't always get it right. Why, with all the technology available, is there still plenty of uncertainty?

Part of the challenge is that weather forecasters aren't just grappling with the future. They also have to deal with uncertainty about precisely what is happening in the present.

The forecasters at MetService have huge amounts of computer power, and data from a host of sources, but they're still grappling with unknowns and uncertainty as they go about their work.

So, even after the high-powered computers have run their highly sophisticated models, humans still make the final call.

Estimates of the current state of the atmosphere are always slightly imperfect. There are unknowns in the initial conditions fed into a model, or assumptions in the model calculations. One way meteorologists try to overcome that problem is with something called ensemble forecasting.

Big southerlies and high tide - chances are the waves will be coming over the road in parts of Wellington.

'Weather models essentially work by taking what the model thinks the atmosphere 'looks like' right now after considering all physical and satellite derived observations … and essentially hits the 'run' button, letting the model run its course,' MetService meteorologist April Clark explained.

**READ MORE:

Flooding in Matamata, Waikato, ahead of Cyclone Cook in mid-April. In the end, Cook made a mess in some places such as Bay of Plenty but the impact on Auckland and Wellington was nowhere as severe as it might have been. Which shows how hard it still is to predict precisely where severe weather is going to hit.

* In photos: New Zealand shivers through a classic winter storm

* Cyclone Cook was supposed to be a record-breaker, so what happened?

Also in April, a week before Cyclone Cook, the country had a shocking reminder of the power of the weather when the Bay of Plenty town of Edgecumbe was inundated after a stopbank gave way.

* Upgraded state of emergency and worry over Edgecumbe stopbank as new deluge nears**

'Ensemble forecasting accepts that we don't exactly know the state of the atmosphere right now - there are gaps in observational data around the globe and also errors in the observations themselves. Ensemble forecasting also take into account that we don't know exactly how the atmosphere behaves - we know pretty well, but there is still a small amount of uncertainty, especially in very dynamic situations, such (as) with tropical cyclones,' Clark said in a Facebook post.

'Ensemble forecasts work by changing ever so slightly how the model 'sees' the atmosphere right now, then hits the 'play' button again and again, and under slightly different forcings. Each of these is called an ensemble member.'

Along with all the super fast number-crunching, and all the data that goes into the models, MetService emphasises the human forecaster remains one of the three essential components that go into forecasting the weather.

Mostly MetService, a state owned enterprise, starts with three models of the global weather - produced in the UK, Europe and US. Those global observations are 'ingested' as starting initial conditions for a forecast. They are run through a New Zealand local area model with improved terrain features.

'That increases the resolution because we've got more observations over New Zealand and we can make it more fine-tuned,' MetService corporate affairs general manager Jacqui Bridges said.

MetService had a standard set of New Zealand observations that were made available internationally, as did all national weather services. But it also had many more observations used just in New Zealand that were fed into the limited area model.

MetService expert meteorologist Leigh Matheson said the global models were 'mathematics on steroids', with some of the world's most powerful supercomputers needed to make the computations involved.

'It is physically impossible for a person to calculate the dynamics of the atmosphere in a reasonable length of time - so this is left to the computers.'

Given that assumptions were made in both the starting conditions and the calculations used to mimic the behaviour of the atmosphere, an error was created that became larger over time. So, the error in a day seven forecast would be greater than in day two.

'In order to gain some knowledge of this error, the ensemble was created,' Matheson said.

Among facilities sprinkled around the country to collect data on the weather is this weather radar station at Kaeo in Northland

The multiple solutions from the ensemble gave a forecaster a statistical guide to the error in a single model, as well as a guide to the variety of solutions across different models.

Numerous solutions created a lot of 'noise', and that's where meteorologists come in - 'to critically analyse this data to make an educated decision on the final forecast'.

The weather station on Brothers Island in Cook Strait. It can only be serviced by helicopter.

'The forecaster has knowledge both of how the models operate, and how the atmosphere behaves, and is hence the essential element that binds model solutions and real time observations,' Matheson said.

A forecast was chosen by a meteorologist after weighing up the options. The process was constantly monitored alongside real-time observations.

An automatic weather station at Mid Dome in Southland.

'The forecaster will step in at any time if a model solution or the current forecast is starting to deviate from the real world,' Matheson said.

A reassessment of the forecast policy was prompted every 12 hours when data arrived from a new model run - allowing the forecast to 'follow the solution of highest probability'.

Collecting weather data beside the Desert Road.

Over at crown research institute Niwa, principal scientist for forecasting Chris Brandolino explains the data problem facing forecasters: 'Getting now right is a challenge,' he said.

Observations were used to estimate the state of the atmosphere. 'Those estimates have errors … With the chaotic nature of the atmosphere those errors can lead to large differences.'

Like MetService, Niwa starts with global models produced by major overseas agencies. Then it concentrates on New Zealand and the surrounding area to produce its forecasts.

MetService employs 66 meteorologists working as operational forecasters, with at least 20 working during the busiest periods of the day, and always some on through the night.

Much of the time, one forecaster is focusing just on thunderstorms. Models struggled with conditions very close to the ground, which were key for thunderstorms, MetService meteorologist Clark said. 'You get the temperature slightly wrong and that makes a huge difference.'

A warning for thunderstorms could only be issued when a cell was identified and tracked on radar, corporate affairs' Bridges said. 'It's very much in the now.'

MetService gets its data for New Zealand conditions from:

* 195 automatic weather stations

* 95 Niwa-owned environmental weather stations

* 31 voluntary observer ships that report wind, temperature, humidity, air pressure and sea state

* 24 drifting buoys in the seas around New Zealand that report atmospheric pressure and sea temperature

* weather balloons released twice a day from four upper air observatories (Raoul Island, Whenuapai, Paraparaumu, Invercargill) that report wind, temperature and humidity through the atmosphere to about 80,000 feet

* automated observations from aircraft flying on domestic and trans-Tasman routes, reporting temperature and wind along their flight paths

Under a national weather services contract with the Ministry of Transport, MetService provides a range of public forecasts, including severe weather warnings, watches and outlooks for rain, wind, snow and thunderstorms. There are also marine forecasts and warnings for recreational and coastal areas, and brief mountain and general forecasts. It also has a contract with the Department of Conservation to develop and provide national park forecasts. MetService also produces forecasts for the aviation and electricity industries.

In MetService's 2016 annual report, chair Anthony Howard and chief executive Peter Lennox mentioned the challenge posed by rapid technological change. 'Adding to this is the advent of a new breed of competitor, for whom meteorological expertise ostensibly takes a back seat to big data analytics and computing power.'

Bridges said that comment referred to IBM buying US-based The Weather Company. The Weather Company's brands include The Weather Channel and Weather Underground. In a press release in June, The Weather Company said it was 'able to generate hyperlocal forecasts for over 2.2 billion locations around the globe every 15 minutes'.

Bridges said The Weather Company was taking a big data analytics approach to weather forecasting, which was new for the weather industry 'but not a trend we will follow'.

'Where big data relies on computational power to produce data-derived insights, MetService relies on meteorologists,' Bridges said.

While there might be debates about just what role big data and computational power have in weather forecasting, there's no doubt modern forecasting is heavily dependent on computers.

A major part of Niwa's work is analysing weather that has already happened, but it does also produce forecasts, and is active on the internet, notably during major weather events.

It also produces seasonal climate outlooks that try to give an indication of what the weather will be like in the coming three months. The outlooks show probabilities for temperature, rainfall, soil moisture and river flows in the country's regions.

The days of Niwa's current supercomputer, nicknamed FitzRoy, are numbered. In June the Government announced $31 million was being spent on three new supercomputers to meet the needs of the scientific research community. Niwa is putting up $18 million of the cost.

'That will give us the computing capacity to push the boundaries of modelling,' Niwa's Brandolino said. Models could be run at higher resolution.

For now, one of the models used by Niwa for the New Zealand area could run at a resolution of 1.5km, but to be able to accurately forecast heavy rain, or other types of precipitation, a resolution of below 1km was needed.

The new supercomputers would enable that to be done in coming years, although it might possibly be restricted to urban areas, Brandolino said.

The meteorologists at Niwa and MetService might be wishing they could forecast more than just the weather. In 2016 the Ministry for Business, Innovation and Employment launched a review looking at whether the two government weather agencies were making enough of their data freely available. That came after a long campaign by privately-owned forecaster WeatherWatch.

Relevant ministers have the report but getting them to give any indications about what's in it has so far been one of those getting-blood-from-a-stone efforts. The silence from the Beehive has extended to whether the review could have any implications for the way government weather activities are organised, and whether there's any problem with duplication by the two agencies.

WeatherWatch chief executive Philip Duncan said his company had a contract with Weather.com (The Weather Channel) and Wunderground (Weather Underground) that provided forecasts for 1500 locations in New Zealand.

'The power of IBM is that they have dozens of supercomputers. We're looking at multiple supercomputers churning out specific forecasts for New Zealand.' Forecasters were also constantly tweaking the data.

Wunderground automatically learnt from its mistakes and that process was leading to constantly improving forecasts, Duncan said.

WeatherWatch ran raw forecasts for most smaller places from Wunderground but overrode forecasts at times for the larger centres. In New Zealand it was important to have an understanding of the country's weather history, and at least 20-30 years experience of how weather systems operated, Duncan said.

The data that went into the WeatherWatch forecasts came from many countries, including the US, Australia, the European Union and Canada. There was some data from New Zealand, including some collected by satellite.

WeatherWatch used data from a variety of sources. 'Most of the data we rely on comes from the US… The US government invests so much money in open weather data …Our Government gives nothing to us.'

The main issue with the lack of open data from MetService and Niwa was reporting on major weather events such as thunderstorms. There was nothing more powerful than being able to report observations showing what was happening 'right now', Duncan said. Open data would give WeatherWatch 'a much much more realistic way of protecting lives and property during storms'.

'In the US, when a severe weather event happens, you will see on TV live rain radar on all the channels. The whole point of that is so people can see.'