RESEARCH: SOUTHWEST LAKE MICHIGAN REGIONAL FLASH FLOOD MONITORING

Phase 2:  About the MRCC Flash Flood Potential Index

The Midwestern Regional Climate Center’s Flash Flood Potential Index tool provides a general assessment of current conditions for numerous locations across the Milwaukee, WI and Chicago, IL National Weather Service offices’ county warning areas.  Based on multiple environmental conditions such as recent precipitation amounts, local topography, and land cover usage, the Flash Flood Potential Index’s algorithm forecasts the potential that a given location has of experiencing a flash flood within the next 12 hours given specific conditions.  The forecast is updated every 6 hours with the most recent precipitation data available during the months of April through October.

Data and Methodology

The Flash Flood Potential Index tool uses data for many factors to predict the probability of flash flood occurrence.  These include:

These data were collected for all locations where a National Weather Service flash flood local storm report had been issued at some point between 2002 and 2017.  The data was then provided to a random forest machine learning algorithm in order to train it on patterns that led to flash flood conditions in the past.  Once the algorithm had processed the data, it is then provided a variety of possible 3-hour accumulations that could occur over the next 12 hours.  For each of these possibilities, the algorithm predicts the probability that any given location will experience a flash flood based on those conditions over the next 12 hours.  This probability is categorized into risk levels in order to better convey the risk that each location is under.

It should be noted that while a flash flood report by the National Weather Service largely guarantees that a flash flood did, in fact, occur, the lack of a flash flood report does not guarantee that a flash flood did not occur, but merely that it was not reported.  From here on, a precipitation event that did not have a flash flood report associated with it will be referred to as a non-flash flood event.

Output and Interpretation

The algorithm can only forecast for locations that experienced a flash flood at some point in the study record (2002-2017).  As such, the forecast provided by the output is specific to the discrete points shown on the interactive map.  The user does have the ability to select a polygon map output to shade areas nearest to a point with the same risk that the point was forecasted to have.

The risk levels of the algorithm’s forecast were meticulously categorized to accurately describe the risk that any location is under to receive a flash flood.  The risk levels are:

While these risk levels are designed to convey the susceptibility of an environment to produce a flash flood, they are, by no means, a guarantee.  The lowest risk level, Marginal, does not guarantee that a flash flood will not occur.  Likewise, the highest risk level, Very High, does not mean that a flash flood is imminent.  Users should exercise caution when interpreting the output of the algorithm.  Never use this tool in isolation for decision-making purposes.

Capabilities and Limitations

The Flash Flood Potential Index provides a unique approach to discerning the susceptibility of a location to flash flooding (given specific conditions).  It can provide an additional tool to National Weather Service forecasters and emergency management to aid in decision-making and predicting.  However, like any forecast, it is not perfect.  When testing this tool on prior events, the algorithm was able to forecast locations that received a flash flood as being at a “Slight” risk or higher 84% of the time.  Similarly, more than 84% of non-flash flood events were forecasted as only a “Marginal” risk.  But by definition, this also means that 16% of flash flood events in the past were only categorized as “Marginal”, and that approximately 16% of non-flash flood events were categorized as being a “Slight” risk or higher.  Because flash floods are rare occurrences already, the 84% of flash flood events categorized as being at a slight risk or higher is far outnumbered by the 16% of non-flash flood events that were categorized similarly.  This further illustrates that this tool should not be used alone and should rather be used in tandem with other tools, weather models, and professional forecasts that are available.

Other Notes and Disclaimers

 

^TOP