Keywords
watershed; parallel processing; scaling
Start Date
6-7-2022 12:00 PM
End Date
6-7-2022 12:20 PM
Abstract
This talk describes changes to the VELMA model that allow the model to scale up to allow simulation of much larger watersheds than the original model version allowed. The U.S. EPA’s Visualizing Ecosystems for Land Management Assessment (VELMA) is a spatially-distributed (grid-based) ecohydrological model. Users specify the grid size based on land cover complexity and computational considerations, e.g., a 30m is often used for mixed use watersheds under 3,000 km2. Simulating moderately large watersheds at this scale was challenging with the original formulation of VELMA due to the amount of computing resources needed and processing time. There was a need to simulate even large watersheds, which prompted the work described here. A number of ways of applying parallel processing were considered. The final solution involved breaking the watershed into any number of catchments and processing them in parallel. The current implementation has a number of benefits: processing individual watersheds reduces the amount of memory needed; catchments can be processed in parallel, subject to upstream dependencies; information is computed for interior pour points, rather than just the final pour point; and individual catchments can have tailored simulation parameters, including different soil and land cover calibrations and scales. The model has been applied to watersheds with more than 5 million cells. This is not the upper limit. Any watershed can be broken up into a manageable number of catchments with the only real limit being the computational time to process all the catchments
Scaling Up a Watershed Model for Large Basins
This talk describes changes to the VELMA model that allow the model to scale up to allow simulation of much larger watersheds than the original model version allowed. The U.S. EPA’s Visualizing Ecosystems for Land Management Assessment (VELMA) is a spatially-distributed (grid-based) ecohydrological model. Users specify the grid size based on land cover complexity and computational considerations, e.g., a 30m is often used for mixed use watersheds under 3,000 km2. Simulating moderately large watersheds at this scale was challenging with the original formulation of VELMA due to the amount of computing resources needed and processing time. There was a need to simulate even large watersheds, which prompted the work described here. A number of ways of applying parallel processing were considered. The final solution involved breaking the watershed into any number of catchments and processing them in parallel. The current implementation has a number of benefits: processing individual watersheds reduces the amount of memory needed; catchments can be processed in parallel, subject to upstream dependencies; information is computed for interior pour points, rather than just the final pour point; and individual catchments can have tailored simulation parameters, including different soil and land cover calibrations and scales. The model has been applied to watersheds with more than 5 million cells. This is not the upper limit. Any watershed can be broken up into a manageable number of catchments with the only real limit being the computational time to process all the catchments
Stream and Session
false