Weather forecasting at regional and national scales has improved vastly over the past half century with accuracy three or four days ahead now about equal to one day out in the 1960s. This improvement has largely been driven by the phenomenal increase in supercomputing power combined with modelling developments harnessing that, but this has exposed a growing deficit in local ultra-short range weather forecasting.
Improvements at that scale, in areas down to just one two square kilometers and timescales of a few hours, have not kept pace largely because the grids used for the models are not small enough and sufficiently reliable data cannot be collected fast enough.
There are signs of that changing and mobile communications is playing a significant role overcoming that data bottleneck, in concert with efforts to increase the resolution of the 3D grids used by the major agencies for global forecasts shared around the world and now widely available on the web.
The global numerical prediction models run by national weather services apply Newton’s laws of motion modified for rotating fluids to run forward in set time steps over a 3D grid around 12km in length. The grid points are populated with data interpolated from a variety of sources including satellites, ground-based weather observation stations and balloons.
There are two problems with this approach for local mesoscale forecasting. Firstly, the grids are too large to represent local detail given that by simple geometry with grids 12km long, the smallest object that can be represented with any accuracy is about 50km across.
Agencies such as the UK Meteorological Office and Météo-France have been developing nested mini grids within these grids, with the first level coming down to around 4km grid sizes and then the second level to 1.4km. These nested models can then be run within the global models and output localized results in detail.
But apart from any computational capacity constraints these run up against the second problem, lack of timely and accurate data at the greater densities needed to populate these higher resolution grids adequately. Crowdsourcing from available sources such as smartphones and vehicles has been applied in combination with machine learning algorithms that correct systematic biases or inaccuracies in this data through collaboration against the smaller number of official stations in the area using professional grade instruments.
But this is work in progress, and in any case useless in sparsely populated areas where smartphones and vehicles may be almost as rare as official weather stations. In such situations mobile networks are starting to figure in combination with low cost sensors that can be deployed in the field.
A good early example is in the Australian outback of Queensland State, where the country’s telco Telstra has deployed weather stations linked to NB-IoT low power WAN (LPWAN) network infrastructure. The area is ideally suited to such deployment, being too sparse for crowdsourced data but with a distributed population comprising a sufficient number of farmers to justify investment in mesoscale forecasting.
The national forecasts are of course helpful for the farmers, but only up to a point, failing to pinpoint local detail such as a developing hailstorm, approaching torrential rain, or dust storms, that can cause catastrophic damage to crops or sometimes livestock in a very short time. The ability to forecast such events with sufficient accuracy and timeliness depends on obtaining data locally and quickly.
Telstra is involved in this deployment as a pilot project organized with the Queensland Department of Agriculture and Fisheries (DAF) and the Australian the Bureau of Meteorology. Austrian company Pessl Instruments is providing the IoT-enabled weather stations, which will be installed initially at 55 NB-IoT sites, including private farms, and DAF research facilities around Toowoomba, in the Darling Downs region of southern Queensland.
Data integrity and accuracy is a challenge even when obtained from professional equipment, which could easily impose unacceptable delay. Telstra has provided a data hub to perform the necessary data checking, cleaning and organization for presentation to hyper-local weather forecasts for the region.
As Telstra pointed out, the data has to be processed more efficiently than before to make the system economically viable and it seems it is too early to adjudge whether it is yet delivering value. But the telco is cautiously optimistic and hopes to extend its hyper-local weather network across the country as its 5G NB-IoT network already has national coverage. It is seeking to enlist “existing networks and local organizations already servicing regional industry” to the cause.
“We think the most effective and sustainable way to deliver a national outcome is to partner with those organizations to develop commercial opportunities that can support regional industry,” said Telstra.
The operator claims that interest has already been expressed by other industries such as mining, but admits that agriculture is likely to the main beneficiary. “Day-to-day decision making and planning on farms across Australia often revolves around weather conditions – and yet in 2021, local weather observations, ‘nowcasts’ and seven-day forecasts are not specific or local enough to use that information productively,” said Channa Seneviratne, executive for technology development and solutions at Telstra.
If the system does gain momentum, Telstra aims to roll out thousands of IoT weather stations around the country in a model that may also be exported. In other countries there could be interest from different sectors including sports and events generally, given that many are susceptible to short term local weather.