110 likes | 380 Views
Groundwater Modeling, Inverse Characterization, and Parallel Computing. Kumar Mahinthakumar NC State University. My Background. Numerical modeling of groundwater flow and transport Developed PGREM3D – P arallel G roundwater REM ediation model – 3D finite element
E N D
Groundwater Modeling, Inverse Characterization, and Parallel Computing Kumar Mahinthakumar NC State University
My Background • Numerical modeling of groundwater flow and transport • Developed PGREM3D – Parallel Groundwater REMediation model – 3D finite element • GW2D – two dimensional educational models for groundwater flow and transport • High Performance Computing • Parallel algorithms, Solvers, Parallel performance analysis • Optimization and Inverse modeling • Groundwater source identification • Hydraulic conductivity inversion • Water distribution source identification and leak detection • Population based optimization algorithms (GA, PSO) • Markov Chain Monte Carlo Methods
Groundwater Remediation Modelingusing PGREM3D Savannah River Site Investigation 1997
Groundwater Source Identification: 3-Source release history reconstruction sampling points flow direction 11 10 12 1 2 3 14 13 15 6 5 4 (x2,y2,z2) 17 16 18 C3(t) C1(t) 8 9 7 (x1,y1,z1) 167 m C2(t) 333 m Sources C1(t), C2(t), C3(t) are the unknown release histories
Hydraulic Conductivity Inversion using the Pilot Point Method Prior True K-field Inversion with Regularization Inversion without Regularization
Scalability of PSO on ORNL’s Jaguar Supercomputer Jaguar PF: 299,008 AMD Cores Weak Scaling of our PSO Simulation-Optimization framework Showing Over 80% efficiency up to 200,000 cores
WSC Project Tasks • Hydrologic Modeling (4.3) • PIHM – Penn-State Integrated Hydrologic Model for groundwater surface water interaction • SWAT-MODFLOW simulations • Water Infrastructure Models (4.4) • Groundwater pumping effects (MODFLOW or PGREM3D) • Reservoir model • Parallel computing • Ensemble reservoir stream flow calculations