Flood forecasts are not predictions

Hey there, time traveller!
This article was published 25/04/2013 (4580 days ago), so information in it may no longer be current.

As the flood risk continues to rise, I have been reflecting on how the public views flood forecasting, particularly after the implied criticism of it in the recent report of the province’s task force on the 2011 flooding. I suspect flood forecasting is not well-understood by most Manitobans and thought I might shed some light on the matter.

The Manitoba government regularly updates us on what may happen through the release of flood forecasts. In order for this information to be most useful, it helps to understand what these forecasts are and are not.

Forecasts are not predictions. That the Jets will win the Stanley Cup this year (or not) is a prediction; that the Jets have a 10 per cent chance of winning the Cup is a forecast.

Forecasts are estimates of the probability of future contingent events actually occurring. We are familiar with weather forecasts and have adjusted to the fact they are fairly accurate in describing tomorrow’s weather but less so when looking ahead one week.

The science of flood forecasting (and it is a science when properly applied) can now utilize the combination of almost unlimited computing power and models of various levels of sophistication to construct scenarios of spring runoff and its impact on water levels.

The models process two types of data; determinate data such as soil moisture, snow depth, topography, channel capacities — in other words data that are (theoretically) either invariant or can be accurately measured — and contingent data such as future daily weather and ice conditions that cannot be predicted, only forecast.

The results therefore, are probabilities of specific water levels occurring at specific times in specific locations.

Every aspect of flood forecasting has a degree of uncertainty. For example, soil moisture is estimated based on field measurements at relatively few locations; snowpack and precipitation data are measured again at relatively few locations and then interpolated to represent all contributing watersheds.

Even converting accurately measured water levels into flow rates can be out 10 per cent.

If the forecast is for tomorrow, a single scenario can have a very high probability of occurrence, a virtual prediction. If we are looking a week or more into the future, the probability of even the most likely scenario will recede from 100 per cent.

To declare that a forecast is “wrong” when judged against the actual event means we have underestimated the likelihood of its occurrence. In that sense, of course, virtually every forecast is “wrong” because it cannot assign absolute certainty to any future scenario while the actual event will, after the fact, have had a 100 per cent probability.

A realistic expectation of forecast accuracy has to account for the uncertainty of the contingent data — principally weather forecasts — as well as the estimated accuracy of the “known” measurable data that have under present circumstances their own degree of uncertainty. Early forecasts will have wide confidence boundaries that narrow as the flood peak approaches and the forecast period shortens.

Manitoba’s flood forecasting has produced over the years remarkably useful forecasts that have allowed us to be well-prepared for most flood events. That is not to say the forecasts predicted the exact flood peaks and their timing, but that the most likely scenario identified the need for protective measures and the lead time to put them in place; neither has it been the case, however, that all past forecasts assigned a high probability to the events that actually occurred. That is the nature of forecasting.

We should expect of our forecasting system to have in place as the four necessary conditions for success the most up-to-date models calibrated to Manitoba conditions; a data-gathering program that ensures a data-rich environment for these models; sufficient staff and monetary resources; and a well-documented methodology guide that will ensure continuity in the face of changing personnel.

As the 2011 Flood Review Task Force Report pointed out, experience can only enhance our forecasting capability. But we should not forget flood forecasting is a science and not necromancy.

The forecasters can only work with the tools they have been given. In 2011, for example, as a result of several years of reductions to the national hydrometric monitoring network, the data environment suffered from substantial gaps. The forecasters did the best they could with what they had.

Government is well aware of the four necessary conditions noted above. Some steps have already been taken to address them and no doubt action will accelerate as a result of the task force. This should be of the highest priority.

Weather forecasters are not only blamed for not predicting the weather accurately (although of course they don’t predict but only forecast) but also on occasion for the weather itself if not to our liking.

Our flood forecasters don’t cause the floods and they work long hours to identify the likeliest course of events; but nature will inevitably surprise us.

All the more reason to continue to expand and improve our permanent flood-mitigation measures, as we did with the floodway after the 1997 Flood of the Century and err on the side of caution until the entire province is flood-proofed to an acceptable standard.

We’re in a new century now — let’s not grow complacent.

 

Norman Brandson was deputy minister of the former Manitoba departments of environment, water stewardship and conservation from 1990 to 2006.