If you were following the weather recently, you know about the Snowquester. What happened was that there was very little snow in Washington DC, and lots of snow in Boston and the Northeast. While this shouldn’t sound surprising, it really blindsided weather forecasters. Forecasters predicted lots of heavy wet snow for DC, which caused government services, municipal services, and schools to shut down before the flakes even began to fall. When the storm came, only a few inches appeared. The forecast was a bust and quite costly to the city. In the Northeast, Boston kept schools open based on a prediction of 6-10 inches of snow, but then received almost 30 inches of the white stuff. Another bust for forecasters. What exactly happened?
The finger pointing began almost immediately and almost everyone and everything that could be blamed was. The result, however, was a perfect storm for those of us that study HPS and STS.
Recently, there has been a lot of disciplinary controversy, both from within and outside of HPS and STS. HPS doesn’t seem like a unified discipline, and practitioners are constantly being asked to justify the applicability and usefulness of their work. STS tends to be more “policy oriented” but faces a criticism that it merely describes science. Can’t the scientists just do that themselves? Below I match a few HPS/STS topics with questions that arise from the recent Snowquester controversy. If nothing else, I hope it shows that HPS/STS deals with topics relevant and important to scientific methodology and policy.
HPS/STS Topic: Epistemic Responsibilities
Snowquester: What should users be aware of when using forecast information? What responsibilities do forecasters have?
This Washington Post blog from 2011 details some of the responsibilities of forecast users, including “knowing the source of the forecast” and “know the limitations of the forecast.” One could assume then, that forecasters would make that information available. For the DC Snowquester prediction, the National Weather Service didn’t convey any uncertainty information, nor, to the best of my knowledge, did they report what model(s)they were looking at to make such a forecast. Uncertainty is notoriously hard to communicate, but essential information when using a forecast. Taking uncertainties into account led the Washington Post to joke: “The best forecast for Snowquester was one we could not issue with a straight face, and one most Washingtonians would have ridiculed: Rain, sleet, and/or snow likely — heavy at times — with snow accumulations of 0-14 inches.”
What information should forecasters consider before making a prediction? As you’ll see below, there’s a lot of talk of how good models are. Well, what seemed clear to many after Snowquester is that forecasters were only consulting US-funded modeling products. The European Center for Medium-Range Weather (ECMRW) model, was not predicting anything near the amount of snow predicted by the American model. If they didn’t consult the ECMWF model, should they have been required to?
HPS/STS: Expertise and Authority
Snowquester: Models vs. Forecasts: How should model information be used? How justified are predictions?
Many people equate information from a model with weather forecasts. But forecasts and models are different. Meteorologists use models as tools, but are supposed to bring their expertise to bear on the data that models produce. This is why studies show that in many circumstances, meteorological expertise adds skill to a model prediction. It has been suggested that forecasters followed the model predictions blindly when making the DC and Boston Snowquester forecasts. Furthermore, model ensembles, which employ many different model runs and are used to help gauge uncertainty, showed that uncertainties were between 5-10 inches – which nearly covered the entire amount of the DC forecast. How do we determine when a model is justified? How can we use this information responsibly?
HPS/STS: Justification of Models and Simulations
Snowquester: What makes a good model? What should we strive for in building our models?
I’m giddy over the fact that weather models made the NBC Nightly News. The American weather models are setup slightly differently than the European’s. The USA uses two models, a global model that has a course resolution, and a North American model, which has a finer resolution, but only covers North America. The North American model gets its boundary conditions from the global model. The ECMWF model is a global model at high resolution. The Europeans have faster computers, and this allows them to use a more advanced strategy to assimilate actual weather data into their models. One response to the “failure” of the American models has been to ask for more money to do what the Europeans do. These are two different modeling approaches, and one may have certain advantages over others. The assumption at work is that “bigger and faster equals better” – but that isn’t always the case. Perhaps it would be best if the two groups chose modeling methods that complemented each other?
HPS/STS: Epistemic Cultures
Snowquester: Should science involve lots of collaboration? What role do scientific organizations play in the projection of knowledge.
The ECMRW is setup very differently from its American counterparts. The ECMW is known to be a smaller group of scientists, who emphasize the application of scientific research, and encourage other scientists to visit their center. American weather research is sprawling and emphasizes basic research. These are certainly very different organizations, and how these organizational styles relate to the quality of forecasts is an open question.
HPS/STS: Infrastructure Globalization
Snowquester: Do the American models need to beat the European ones? Should science be divided along national boundaries?
So the ECMWF model made a better prediction. So? Those in charge of American meteorology have used this to argue that America needs to catch up with Europe, that we’ve “fallen behind” in the race to capture the weather. Is there really a race between the Americans and Europeans? Why does America need to catch up? What is never mentioned in this debate is that the ECMWF information is available and the US can simply buy it (which is probably much more affordable than a few supercomputers and a new research center). It makes more sense to work collaboratively with the Europeans and use each model and research team to their strengths. Is this the best way to make progress?