Microsoft Corp. has awarded NOAA Fisheries researchers 2 years worth of high-powered computing time and expertise to develop an artificial intelligence model. It will forecast how changing flows in the Columbia River Basin affect salmon habitat.
The AI model will give water and salmon managers more accurate and timely information for less cost. This will help them make the best decisions for the economy, environment, and the public.
“The model trained on high-resolution satellite images should more quickly and easily show salmon and water managers how changes in flows can expand or reduce available salmon habitat along rivers and streams,” said Morgan Bond, a research scientist who leads the project at the Northwest Fisheries Science Center. Such information previously required extensive field studies and analysis to evaluate even small areas of the basin.
“This approach should speed that up significantly,” he said.
The award provides support from Microsoft’s AI For Good Lab. It’s part of a special invitation for projects that advance sustainability, public health, and human rights in Washington state, marking Microsoft’s 50th anniversary. The lab will collaborate with project leaders to make use of $5 million worth of credits for access to Microsoft’s cloud computing capacity.
"Microsoft's AI for Good Lab was inspired to join this project because protecting wild salmon runs is vital to the economy and culture of the Pacific Northwest,” said Juan Lavista Ferres, chief data scientist at Microsoft. “By providing technical assistance and Azure computing power, we're excited to help NOAA gain a deeper understanding of real-time waterflows throughout the year and how these varying flows impact salmon science and habitat recovery."
Insight at the Watershed Scale
The partnership will develop the model and analysis through cloud computing that will save scientists the time and cost of downloading and processing high-resolution satellite images. The images comprising many gigabytes each would otherwise consume so much bandwidth that they slow down computer systems.
The new model would show managers how changes in flows would alter the amount and location of salmon habitat on the landscape. “This would put us far ahead of where we are now in terms of seeing how different water flow choices would affect the landscape in very short order,” Bond said. The AI model will simultaneously assess river habitats in fine detail over a large area, which has eluded researchers in the past, but can best support recovery of threatened and endangered salmon.
“That is where this really pays off,” Bond said.
One of the most common recovery tools in the Columbia Basin involves purchasing water rights to maintain more reliable flows in rivers and streams, for example. However, it’s often difficult to tell how much the extra water actually benefits salmon, because the relationships between flows and habitat are not clear, Bond said.
“Are these additional flows providing a meaningful benefit for fish?” Bond said. “We assume that is the case, but don’t really have the tools to quantify it.” The new model could quickly forecast exactly how much more habitat the additional flow could provide. That will help salmon recovery leaders understand which projects provide the most additional fish habitat for the money.
“In some cases a small change in flow could make a big difference in habitat,” he said. “We could see exactly where we get the biggest bang for the buck.”
