Human Data Interaction in Meteorology: The Decision-Making Behind Our Forecasts

Design for information differs from most of its sibling disciplines. For one, it is more reliant on quantitative user research. As humans, our ability to comprehend (and gain value from) mass quantities of complex data can only be measured through rigorous research and user testing, and unlike other areas of design, there is little room for personal opinion.

Still, with a lack of available research and the potential for small and domain specific variables to create significant flux in results, we find ourselves continuing to employ guess work when designing many methods of communicating information. When it comes to designing for information display in meteorology, the lack of research is, of course, not a singular problem — there are others that make the field a challenging one. Our current era of information is different from any that have gone before it, especially so for atmospheric sciences. We have more data to contend with, a vaster array of tooling and technologies serving a medium for this data, and of course, we are under more wide scale scrutiny of the impacts of our decisions. It’s easy to see why the communication of big data is a difficult problem. However, what is even more difficult, is the challenge of making fast paced and high stakes decisions off the back of this information.

All of these factors combined mean that we are presented with an enormous research opportunity. Forecasting has long said to be an art form, but this assertion can actually be broken down. Our meteorological experts are asked to develop and employ some of the most complex decision mechanisms known in short time cycles, all leading to decision making that — in some cases — will mean the difference between property damage or protection, or at the most extreme, life or death.

As researchers who are lucky enough to be based just across the road from some of the world leaders in meteorology, we are beginning to develop research programs that will help us to capture these mechanisms, and discover human methods of comprehending extremely complex data, in the hopes of reimagining the way in which we design for human data interaction, particularly within the field of atmospheric sciences.

What makes our problem unique?

It’s true that many areas of industry are facing similar challenges. Communicating big data is a challenge in any situation. However, capturing decision making and comprehension in meteorology presents specific challenges.

For one, weather model outputs are highly uncertain. Meteorologists work regularly with multiple model outputs of what the future might look like, and are therefore required to skim these different realisations of data in order to make decisions.

Similarly, our data isn’t just big — it’s complicated. Meteorologists must ingest vast amounts of data, but must also work within a multi-dimensional space in order to build an accurate picture of the Earth’s atmosphere. Weather happens in a geospatial construct, so how do we capture the cognitive process of conversion from binary digits on a screen, to a functional model of a multi-dimensional space?

Lastly, many of the decision making factors are hidden. A meteorologist might glance at several sources of information. They’ll take note of several features, attention drawn by the colours or shapes they are trained to look out for, but there’s a lot more happening that we, as outsiders, don’t see. As decisions are made, meteorologists are employing embedded domain knowledge, such as geography, topography, and previous experience, perhaps without even being aware of doing so. Capturing these steps of the decision model can be difficult, and relies heavily on well designed methods of cognitive assessment.

Research areas and benefits

Forming accurate representation of meteorological decision models is a difficult task, however successful research in this area could present many various opportunities for future application, for example;

• Improving design and communication of complex information:
  As someone from an information design background, the driving force behind the research is to reinvent the way we interact with complex information, be it through more effective design of digital displays, or exploration of more intuitive data displays, such as 3D visualisation. Learning from decision mechanisms can allow us to highlight where we are inflicting unnecessary cognitive load on our data interpreters, and how we can design systems that instead work in cohesion with our cognitive models.
• Capturing decision mechanisms for the advancement of machine learning models:
  Machine learning often strives to carry out tasks that have thus far been completed by humans. We would therefore like to explore whether capturing the hidden human factors of decision making in forecasting could be iterated back into machine learning models, in order to advance their functionality.
• Understanding the psychology of meteorology: Projecting psychological theory onto our current data interpretation could allow us to recognise areas of cognitive bias, or the difficulties in recognising particular information representation methods. For example, are we psychologically programmed to miss particular colours? Do we deploy different attention mechanisms when asked to carry out different tasks? Our hope is that a better understanding of how data representation impacts the way we extract meaning could offer insight into how we can increase the value gained from it.

Next steps

Over the next few months, we will be completing our first pilot experiment with our meteorologists. It’s expected that this process will teach us vast amounts about how we capture the human factors of forecasting, and how we can design experiments that will allow us to gain insight into cognitive processes to develop the way we design systems for the communication of complex information.