Q&A: The University of Michigan’s Jeff Cwagenberg on going from intuition to (computer) cognition



Jeff Cwagenberg (photo credit: University of Michigan)
The 2015 World Solar Challenge from Darwin to Adelaide, Australia had its tightest finish in history. Less than an hour separated first and fourth place. Top US finisher, and fourth overall, University of Michigan trimmed nearly 11 hours from their top 10 finish in 2013. To do this, they (and IBM’s small camera with access to big weather data) turned to the sky.

Jeff Cwagenberg, the team’s meteorologist, explains how cloud predictions, burn off, and air shifts at Flinder’s Range all came together for the team’s best finish down under.

What did IBM's technology give UM that they otherwise would not have known, or been able to do during the World Solar Challenge?

Jeff Cwagenberg: From competing in this competition since the early 1990s, our team has developed a strong understanding of the weather that can be expected over the route [in Australia]. One of the tools that forecasters often use is simply intuition: seeing patterns, and recognizing from past inferences to draw future conclusions.

During the middle of the event, the trailing tail off of a low pressure system draped about a 100 km wide band of clouds across the route. We were able to determine that the clouds would burn off into the late morning, and we were able to ‘squeeze the lemon’ and drive to avoid the cloud cover.
– Jeff Cwagenberg, UM Solar Team Meteorologist
Even with our almost three decades of experience forecasting, the UM team from an intuition perspective, is at a disadvantage not being local. With the IBM cognitive forecasts and cloud camera technology, we were able to use machine learning to essentially build that intuition into a computer model. By synthesizing historical model, satellite, and ground-based sources, the IBM forecasting tools improved our native human intuition by utilizing existing data resources [that no human could have analyzed alone]. 

What data was most relevant, and crucial for the UM team's race plan? 

JC: The most relevant, and crucial data for developing race strategy are solar irradiance and wind data. Solar irradiance forecasts help predict energy generation so we can develop strategy around how much energy we can expect to have. And wind data helps develop strategy around how much energy we can expect to burn.
Generally, the more sun, the faster the race. There are some strategies that in broken cloud cover, you might speed up a bit while under clouds, and slow down under the sun to spend the most time in the sun without meaningfully impacting energy usage. 

How did UM use the data provided by the solar cameras? 

JC: The solar cameras provided a great platform for two purposes. The first of which was to validate the weather forecasts. Through both the camera itself, and sensors attached, we were able to confirm the forecasts as we drove down the rout and across time. The other manner we were able to use the camera was as a short-term estimate of when clouds could be expected to block the sun (and therefore charging).

UM Solar Team Lead Strategist Leda Daehler
reviewing race data.
(photo credit: University of Michigan
Solar Car Team and Epik Studios)
One good example is how we used the camera to follow the tail off of a low pressure system draped across the route. Using the cloud camera as a tool to measure motion and density of the clouds, we were able to confirm what the model had predicted within an incredibly small error range. 

The IBM forecast gave use a place and time where we could expect the clouds to begin to dissipate; we went to that point, waited for that time, and exactly as forecasted, the clouds began to dissipate. 

What challenges did Australia, and this route, present for the team, the solar camera, and the data analysis? 

JC: One of the most difficult parts of forecasting in the Australian Outback is the lack of data and access to it. In the past, forecasts were generated by meteorologists analyzing various data sources, and developing forecasts from there. As the IBM model utilizes all of those sources to make a composite forecast, it makes the process significantly faster. With satellite internet extremely limited, the composite forecast allowed us to use significantly fewer resources to essentially generate a similar result.

In general, how did cloud cover, or other environmental factors, impact the race? 

JC: In general, weather in the Outback is impacted in two ways: localized ocean-influenced systems near the coasts, and large scale pressure-driven weather systems, throughout. During this event, we saw both types of weather. The most significant impact to the race, though had to be clouds that formed as moist air lifted over the Flinders Range in the last 300 km of the route. With multiple weather forecasts giving differing results, we were able to use all the resources available, and generate a forecast that was better than any one model.

Read more about how solar forecasting was used at the World Solar Challenge, and follow the UM Solar Team.
Clouds over the World Solar Challenge route
(photo credit: University of Michigan Solar Car Team and Epik Studios)

Labels: , , ,