Parallel Computation of River Basin Parallel Computation of River Basin
Hydrologic Response Using DHMHydrologic Response Using DHM Reports Environmental Hydrology Team: NCSA
Alliance All-Hands meeting
May 23-25, 2001 Urbana, Illinois
Baxter E.Vieux
Daniel Weber
Fekadu G. Moreda
Henry Neeman
Zhengtao Cui
Contact: [email protected]
www.coe.ou.edu/emgis
University of Oklahoma, Norman, Oklahoma
OverviewOverview
Objectives The Distributed Hydrologic Model Preparing the existing model for parallel
computing Parallelization Time of computation Coupling the model with ARPS
ObjectivesObjectives
Near term– Couple atmospheric model and surface runoff
model for flood forecasting– Improve computational efficiency of surface
runoff model
Long term– Integrate model into EH system
Rainfall
Infiltration
Runon Runon
Runoff
Stream
Overland
Direction
Flow Characteristics Channel Characteristics
- Cross-Section Geometry- Slope- Hydraulic Roughness
* Rainfall excess at each cell
- Soil infiltration rate - Rainfall rate - Runon from upslope
Grid Cell Resolution Finite ElementsConnectivity
Watershed Runoff Simulation
Runoff SimulationRunoff Simulation
Digital WatershedDigital Watershed
Arc.water.feaArc.water.feaImport DEM
Project Setup
Extraction
Simulation
Draw Hydrograph
Import DEM
Project Setup
Extraction
Simulation
Draw Hydrograph
Forecast Location
Model componentsModel components
Time static (Preprocessing)– Importing DEM– Watershed delineation– Setup specific experiment
Time Dynamic– Extraction– Simulation– Routing
Preparing The Model for Preparing The Model for Parallel ComputingParallel Computing
Optimizing the existing code (rewrite in C++)
Isolate the I/O operations
ParallelizationParallelization
MPILoad balancing algorithm
Load Balancing AlgorithmLoad Balancing Algorithm
1 2 3 4 5 6 7 8
0 1 2 3 3 2 1 0
Processes (Descending loads)
Processor assigned in an alternate fashion
Basin
Proc
Illinois River Basin, In Illinois River Basin, In Okllahoma and ArkansasOkllahoma and Arkansas
Illionois river at Tahek
Computation based on Computation based on Subbasins Subbasins
Total # Subbasins =57
Max# Grid cells 12231
Min # Grid cells 54
Load BalancingLoad Balancing
Processor # processes Processes Assigned
0 3 3 49 281 3 23 48 502 3 5 22 513 3 27 29 244 3 30 47 525 3 31 46 536 3 32 45 547 3 33 44 558 3 4 43 129 4 17 42 19 14
10 4 34 41 25 2011 4 35 8 6 1012 4 36 40 11 2113 4 37 9 18 1514 4 2 39 16 115 4 7 38 26 13
Distribution of loadsDistribution of loads (16 Processors) (16 Processors)
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
0 2 4 6 8 10 12 14 16
Processor #
Nu
mb
er o
f ce
lls
Ass
ign
ed
Distribution of loadDistribution of load (4 Processors) (4 Processors)
0
5000
10000
15000
20000
25000
30000
35000
40000
45000
50000
0 1 2 3 4
Processor #
Nu
mb
er o
f ce
lls
Ass
ign
ed
Time of computationTime of computation
Computation Time vs. Number of ProcessorsMoniter Time = 10000 min. (7 days)Time Step = 2 secondsNumber of Cells = 172335
Number of Processors
0 2 4 6 8 10 12 14 16 18
Co
mp
uta
tio
n T
ime
(H
H:M
M:S
S)
03:00:00
05:00:00
07:00:00
09:00:00 9 h
5 h
3 h 1h:30min
01:00:00
Prototype Operational DomainPrototype Operational Domain Illinois River Basin
Area: 2400km2
Resolution 30m x 30m
#Subbasins 370
7 days of monitoring
Timestep = 2sec
Prediction: better load balancing
Coupling DHM with ARPSCoupling DHM with ARPS
Output of ARPS (Rainfall) -> Input to the surface runoff model
Flows at subbasin and river streams are predicted
Interface to run both models from webVisualization of results (VisAD)
ARPS: Rainfall prediction:01hARPS: Rainfall prediction:01h
01z0 m m1- 5 mm5- 1 0 10-1 515-2 020-2 525-3 030-3 535-4 0No D ata
W atrsh edStre am s
ARPS: Rainfall prediction:02hARPS: Rainfall prediction:02h
02z0 m m1- 5 mm5- 1 0 10-1 515-2 020-2 525-3 030-3 535-4 0No D ata
W atrsh edStre am s
ARPS: Rainfall prediction:03hARPS: Rainfall prediction:03h
02z0 m m1- 5 mm5- 1 0 10-1 515-2 020-2 525-3 030-3 535-4 0No D ata
W atrsh edStre am s
ARPS: Rainfall prediction:04hARPS: Rainfall prediction:04h
04z0 m m1- 5 mm5- 1 0 10-1 515-2 020-2 525-3 030-3 535-4 0No D ata
W atrsh edStre am s
Flow PredictionFlow Prediction
0
100
200
300
400
500
600
0 120 240 360 480 600 720 840 960 1080 1200 1320 1440
Time (min)
Flo
w m
3/s
Rainfallfor for hrs
DiscussionDiscussion