Case Study


The ARRIA NLG Engine can write a detailed 3-day weather forecast for 5,000 locations in one minute

Meteorological systems are highly sophisticated with a mass of weather data captured and available as weather changes happen moment by moment across the world. The biggest challenge is the highly skilled human effort of first interpreting this raw weather information, and then writing specific forecasts to inform the general public and industry.

See an example of the ARRIA NLG Engine’s output below

Since 2009, Arria has been working with a leading national weather service and others to embed the expertise of weather forecasters into the Arria NLG Engine. To date, the ARRIA NLG Engine™ has been used to create several types of weather reporting:

Site Specific:The Arria NLG Engine can be configured to write on-demand, detailed weather forecasts for any site specific purpose in the world such as docking supply boats and flaring excess gas on offshore oil rigs in the North Sea, harvesting on a farm, or even for the weather at the beach you want to go to. In most countries there are tens of thousands of locations of public interest where it would be useful to have a detailed local forecast but, instead, there are only generic forecasts available for big regions like “the south east”, because of the time required to write these forecasts.

Now, our NLG software takes our existing weather service client’s complex raw data inputs and can write a detailed 3-day weather forecast for 5,000 locations within a single minute—and can update them instantly. Right now, our NLG software tackles tasks that would be impossible for forecasters to complete manually. It would take a forecaster 1.5 months to create the equivalent of our system’s one-minute output.

We envision that the Arria NLG Engine will be used to generate instant, personalised, localised, on-demand mobile phone content such as our detailed site specific written weather forecasts, to any smart phone user in the world.

Area Specific: Area Specific: The Arria NLG Engine can write on-demand weather forecasts for specific geographic areas, and it can also write area specific hazard warnings such as in the complex process of forecasting for gritting and other winter road maintenance procedures.

The Core Arria NLG Weather Engine

Weather reports play a key role in marine, aviation, ground transportation, finance, retail, space and defence. The Arria NLG weather reporting technologies benefit these categories. We are currently working with our existing national weather service client to develop a user configurable Arria NLG Core Weather Engine. This Engine creates the potential for a completely new weather forecasting paradigm: an NLG Engine that will be able to output on demand, detailed, weather reports for any site, for any area, for any purpose, for any period of time; an NLG Engine with a user-configurable interface that could include:

  • Natural Language Output Options that Control:
    The grammatical formation of sentences, the variety of the language.
  • Format Options:
    Written text, annotated graphs and maps, speech.
  • Geographic Place Options:
    Site specific forecasts, area wide forecasts, point-to-point forecasts, road and surface temperature forecasts (used by the road industry and commercial aviation for runway information).
  • Window Options:
    24 hour forecasts, 2–5 day forecasts, etc.
  • Primary Weather Area Options:
    Wind, rain, temperature, etc.
  • Primary Market Area Options:
    Public, Military, Road, Rail, Retail, Construction, Financial, Commercial and Private Air, Marine.
  • Delivery Options:
    iPad app, iPhone app, SMS, RSS feeds, websites, fax, email.


Arria NLG reports were rated as more accurate than human written reports


A challenge with forecasting is that it is not written using grammatically standard English. There are fewer verbs, and the sentences are truncated.


Our NLG software is fluent in “weatherspeak”, writing the reports as an expert forecaster would write them.


An evaluation of our SumTime NLG weather outputs showed that experienced forecast users rated them more appropriate, more accurate and easier to read than human generated forecasts.1



To write the forecasts shown (to the left below) the NLG Engine processed these requests:


For selected date:

1) Choose length of output
2) Read raw model data
3) Choose output area: frame of reference (GIS shapefile)
4) Analyse precipitation data, intensity type, spatial variation and cloud cover
5) Recognise patterns, shapes and isolate
6) Discern movement of patterns, shapes
7) Language analytics processing
8) Set output length less than 100 words
9) Generate text output describing weather in the selected data subset





1 Source: Research Paper “Choosing Words in Computer Generated Weather Forecasts, June 2005 (SumTime was the name given to
an NLG system that generates weather forecasts texts from numerical weather prediction data).