GPU-accelerated Hydrology Algorithms for On-prem Computation: Flow accumulation,(…)
summary
GPU implementations rework flow direction, flow accumulation, stream ordering, watershed delineation and runoff into PI / SP / LP primitives and a GPU-parallel flow-accumulation trick. They report processing the Lower Ganga basin — 250,000 km2 at 30 m resolution — on a commodity GPU workstation and matching accuracy of CPU methods. Runoff becomes a jump-matrix lookup by precomputing pixel-to-pixel travel in time steps, turning long iterative propagation into fast table lookups.
AI-written summary · about
chapters (7)
- 0:00Hydrology algorithms needs
- 4:54Raster to modeling approaches
- 8:03Core Stack data management
- 10:54High-res simulation challenges
- 13:06Flow direction & accumulation on GPUs
- 16:40Runoff simulation with jump matrices
- 18:24Future directions & Q&A
related talks
01h 08mGPU Data Science with Rapids
29mRay Data on GPU - Breaking Preprocessing Bottleneck for Model Training
38mRAPIDS- Open Source GPU Data Science
49mGraph Based Processing of Big Images
34mGPU as a Service Over K8s: Drive Productivity and Increase Utilization
41mLuxuries, necessities, and the challenges that remain
28mDecoding the weather: How GRAF & New Computer Modeling can Change Everything
01h 12mGPU-Accelerated Data Analytics in Python