Organizations possess a wealth of information, but are stymied by how to use it. Managers are unsure what questions they should be asking and often lack fresh perspective, sometimes relying on computational methods that fail to deliver new insights.
LMI solves this dilemma by pairing a data scientist with a domain expert to understand the data you have and the questions your data can answer. This collaboration involves applying out-of-the box thinking—identifying the right computational methods that enable managers to answer critical questions.
Predicting Climate Extremes
We recently applied this approach to large amounts of climate data. The National Climatic Data Center (NCDC) maintains a vast trove—petabytes in size—of environmental data on everything from ocean and land temperatures, to precipitation totals, to weather events such as hurricanes, tornadoes, and floods. Unlocking this data helps us anticipate changes in our environment.
Asking the Right Question
Historically, organizations have used NCDC’s data to estimate climate “normals,” or averages for specific locations and dates. “Unfortunately, it’s climate extremes that cause catastrophic impacts to physical infrastructure and crops,” said Dr. Adam Korobow, Director of Data Science at LMI. When temperatures exceed certain levels, cooling systems fail, prolonged droughts devastate crops, and large floods break pumps and storm drains.
“The question managers really need answered is, ‘What is the likelihood the temperature will be far above or below the norm on a particular date?’” Korobow said.
To answer this question, subject matter experts from LMI’s Energy and Environment group collaborated with two LMI data scientists to develop a computational approach—based on fitting sequences of probability distributions—to better estimate the probability of weather extremes on given days. “We tapped into vast amounts of NCDC data and squeezed that data harder to derive the probabilities of rare, extreme values,” said James Hebden, data scientist at LMI.
Despite NCDC’s robust datasets, conventional methods require still even more data to accurately quantify the daily probabilities of rare events. But the new computational approach, applied to the same datasets, yields daily probability distributions for temperature, precipitation, and wind levels, including the probabilities of extremes in the tails of the distributions.
Under this method, a policymaker or insurer, for example, can easily determine the tail risk posed by extreme temperatures on a given day, rather than simply relying on historical averages or spreads that don’t speak to the tails of the distribution. With Hebden’s approach, an insurer or policymaker has a better idea of the risk of a “black swan” type event.
Results Inform Critical Decision Making
Greater accuracy in predicting climate extremes has many applications: farmers may protect against crop loss due to drought; urban planners may mitigate the risk of heat stroke; building managers may keep air conditioning systems running; and engineers may design, build, and maintain infrastructure and systems more tolerant of weather extremes.
By asking the right questions and applying novel computational analytic methods, LMI helps clients derive value from their data to improve decision making.