The First Deep-Learning Models of Basin-Level Water Behavior
A Future H2O Research Brief
Damavandi, H.G., Stampoulis, D., Shah, R., Wei, Y., Boscovic, D., Sabo, J.L. (2019). “Machine Learning: An Efficient Alternative to the Variable Infiltration Capacity Model for an Accurate Simulation of Runoff Rates,” International Journal of Environmental Science and Development (IJESD).
Damavandi, H.G., Shah, R., Stampoulis, D., Wei, Y., Boscovic, D., Sabo, J.L. (2019). “Accurate Prediction of Streamflow Using Long Short-term Memory Network: A Case Study in the Brazos River Basin in Texas,” International Journal of Environmental Science and Development (IJESD).
Big Takeaway?
These papers answer two questions of whether you can use a form of artificial intelligence known as deep learning to nimbly simulate how water will behave within a basin. How much water is there? And how will the water actually flow? We think these are the very first attempts at replacing or emulating hydrological models using a pure data-driven model.
What’s New?
Deep learning is the next generation of hydrological analysis. Conventional modeling to predict the movement of water is time-consuming and labor intensive. Scientists sometimes need up to two years to properly calibrate a model at a basin scale. Deep learning lets hydrologists do this analysis faster and more efficiently, in some cases replacing conventional modeling altogether.
Beneficiaries?
Hydrologists, for sure. But at the very top the real beneficiaries are companies that want to, say, restore a network of wetlands across a river basin, need to know the cost, and don’t want to wait years to find out. With machine learning they can do all this without the time commitment of conventional modeling.
Timeline/Obstacles to Implementation?
Scalability is the main issue. If you train this model on the Brazos River basin but tomorrow want to do another basin, you should be able to use the information from the machine learning for one basin on another. So for that software we will need to change the model in a way that it’s transferable. In two years we might have a final software product.
Bottom Line
Deep learning is advancing daily — but it’s already more capable than conventional machine learning at giving decision makers the insights they need on the water sustainability issues they face. Look for deep learning to soon become a critical component in water sustainability analysis.
The Future H2O Researcher
Hamidreza Ghasemi Damavandi is a postdoctoral research associate at ASU’s Future H2O, where his main research interest is to apply machine and deep learning strategies to mimic environmental and hydrological events. He holds a PhD in electrical and computer engineering from the University of Iowa.
For more information about ASU Future H2O’s work and research on creating opportunities for global water abundance, visit our website and subscribe to our newsletter.
