Numerical Modeling of Ecohydrological Processes and Contaminant Transport

Using Microsoft Azure Cloud

Chunmiao ZhengCenter for Water Research

Peking University(http://hydro.pku.edu.cn)

Three Grand Challenges in Hydrological Sciences

• Water Cycle: An Agent of Change

• Water and Life• Clean Water for People

and Ecosystems

− National Research Council (2012)

China Water Crisis:Water Scarcity & Water Pollution

用于污水灌溉的河流

Water Pollution Control Action Plan~2 Trillion RMB

Science 340May 17, 2013

C. Zheng & J. Liu

Case Study 1: Numerical Modeling of Ecohydrological Processes in Heihe River Basin

Heihe River Basin

Total Area: 130,000 km2

The second largest inland river basin In China

Gobi Desert

Qilian Mountains

Hexi Corridor Oases

Gobi DesertMountain: 34%Oasis: 9%Desert:57% Main River Channel

Landscape of Heihe River Basin

Qilian Country

Qilian Country

Ejina Basin

Terminal Lake

Zhangye BasinField Study

Heihe Research Program

• An on-going major research programme of National Natural Science Foundation of China (2011-2018)

• Conduct an integrative study of ecological and hydrologic processes in Heihe River Basin toward more sustainable water resources management

• Led by Prof. Cheng Guodong of Chinese Academy of Sciences and advised by an expert panel of multidisciplinary scientists

• 200 million RMB core funding (~32 million USD)

Overall Objectives

Integrate observation, experimentation, and modeling

Improve predictive capability

Increase water use efficiency

Toward more sustainable water resources in arid ecosystems

Decision maker

Decision support system

Future forecast

Hydrological Ecological Economic

Scientist

HEIFLOW (Hydrological-Ecological Integrated watershed-scale FLOW model)

• Physically-based distributed-parameter 3D numerical model

• Include all key components of the hydrological cycle

• 2D overland flow and river channel hydrodynamics

• Saturated/unsaturated zones• 3D solute and heat transport• Modules for agricultural

crops and desert vegetation

Heihe River Basin Geodatabase

Simulation

Raster series

Surface water :(2-D)Waterbody (Polygon feature class): lake, ponds, (swamps)Waterline (Line feature class): streams, riversWaterPoint (Point feature class): springs, water withdrawal/discharge locationsWatershed (Polygen feature class): drainage areas

Groundwater: (2-D , 3-D)Aquifer (Polygon feature class):confined and unconfinedWell (Point feature class): monitoring, water supply, and irrigation wellsBorehole ( Boreholelog Table): vertical dataBoreline and BorePoint: 3D (z-enbled) line and point feature classGeoArea, GeoPoint

Framework

Subsurface

Georaster

Hydrogeological map

AquiferLandcover

TM

Geology control points

“Digital Heihe” Datasets

Foundation data DEM，Topographic, Hydrological map

Earth observation dataLandsat, ASTER, QuickBird——Thematic dataGeology, Hydrology, Vegetation——

Observation dataMeteorology、Hydrology、Groundwater——

Experimental dataField Survey, test——

Model dataRadiation，Land assimilation，SWAT——

http://westdc.westgis.ac.cn/

DIGITAL RIVER BASIN

3D View of Subsurface with Cross Sections

Observation well

Meteorological Station

River

Glacier and snow

Quaternary aquifer

Tertiary aquifer

Bedrock

LegendsHydrogeological Units

Middle & Lower Heihe River Basin Model Domain:~100,000 km2

Grid spacing:1 km by 1 kmRows: 548Columns: 404Layers: 5

Surface Water Model for Middle/Lower Heihe

Discretization and Parameter Zonation for Subsurface Model

Layer 1 2 3 4 5

Unconfined aquifer

Aquitard 1 Shallow confined

Aquitard 2 Deep confined

HEIFLOW on the Cloud

Conceptual Model

Model Execution

Model Calibration

Uncertainty Analysis

Optimized Decision

Compute Nodes

(Worker Role)Front End

Node

(Web

Role)

Azure Storage

(Blob/Table/SQL)

Element Model Execution Calibration

Parameterization Design Execution

Integration Execution Calibration

Head Node

(Worker Role)

Client Applications

ArcGISMatlab GMS

Blob Upload & Download

Hpcpack

Cluster & jobmanagement

Jobscheduling

Microsoft Azure

HPC on Microsoft Azure for ecohydrological modeling

Deployment of HPC on Azure for ecohydrological modeling

Managing NodesInstalling HPC Uploading Model Verification Application

Deployment of HPC on Azure for ecohydrological modeling

Managing NodesInstalling HPC Uploading Model Verification Application

Deployment of HPC on Azure for ecohydrological modeling

Managing NodesInstalling HPC Uploading Model Verification Application

Deployment of HPC on Azure for ecohydrological modeling

Managing NodesInstalling HPC Uploading Model Verification Application

Deployment of HPC on Azure for ecohydrological modeling

Managing NodesInstalling HPC Uploading Model Verification Application

Compute NodesFront End Node

Azure Storage

(Blob/Table/SQL)

Head Node

Client Applications

ArcGISMatlab GMS

Blob Upload

Hpcpack

Clustermanagement

Jobscheduling

Microsoft Azure

Ecohydrological model execution in HPC on Azure

Data Preparing & Uploading Job Submitting

Blob Down

Status Monitoring

Result Getting

Compute NodesFront End Node

Azure Storage

(Blob/Table/SQL)

Head Node

Client Applications

ArcGISMatlab GMS

Blob Upload

Hpcpack

Clustermanagement

Jobscheduling

Microsoft Azure

Ecohydrological model execution in HPC on Azure

Blob Down

Data Preparing & Uploading Job SubmittingStatus

MonitoringResult Getting

Compute NodesFront End Node

Azure Storage

(Blob/Table/SQL)

Head Node

Client Applications

ArcGISMatlab GMS

Blob Upload

Hpcpack

Clustermanagement

Jobscheduling

Microsoft Azure

Ecohydrological model execution in HPC on Azure

Blob Down

Data Preparing & Uploading Job SubmittingStatus

MonitoringResult Getting

Compute NodesFront End Node

Azure Storage

(Blob/Table/SQL)

Head Node

Client Applications

ArcGISMatlab GMS

Blob Upload

Hpcpack

Clustermanagement

Jobscheduling

Microsoft Azure

Ecohydrological model execution in HPC on Azure

Blob Down

Data Preparing & Uploading Job SubmittingStatus

MonitoringResult Getting

(a)

800 1000 1200 1400 1600 1800800

1000

1200

1400

1600

1800

Co

mp

ute

d (

m)

Observed (m)

Computed vs. Observed Head

Middle HRB

Lower HRB

(b)

Model Calibration:

(a) Comparison between contour maps of computed

and observed groundwater levels;

(b) Comparison between computed and observed

heads at monitoring wells;

(c) Comparison of computed and observed

streamflows and evapotranspiration

0

20

40

60

80

100

120

140

160

180

Str

eam

flo

w (

bil

lio

n m

3/

s) 高崖 （NS=0.92） Simulated

Observed

0

20

40

60

80

100

120

140

160

180

200

Str

eam

flow

m

3/

s)

正义峡（NS=0.85）Simulated

Observed

Gaging Station 1

Gaging Station 2

(c)

Simulated Evapotranspiration Dynamic Patterns (2000)

Element Model - Calibration

Comparison of parameter estimation runtimesobtained from a dedicated local desktop array and virtual machines run on the cloud

PEST

Results of sensitivity analysis on Cloud

00.5

11.5

22.5

3

rch

_71

rch

_64

rch

_77

rch

_76

rch

_63

rch

_74

rch

_78

rch

_82

rch

_91

rch

_79

rch

_83

rch

_84

rch

_81

rch

_80

rch

_73

rch

_65

rch

_86

rch

_75

rch

_90

rch

_88

rch

_89

rch

_69

rch

_72

rch

_70

rch

_61

rch

_85

rch

_67

rch

_62

rch

_68

rch

_60

rch

_66

Sensitivity_recharge

0

0.05

0.1

0.15

0.2

0.25

hk_

27

hk_

54

hk_

29

hk_

18

hk_

42

hk_

11

hk_

3

hk_

25

hk_

13

hk_

14

hk_

1

hk_

9

hk_

22

hk_

5

hk_

17

hk_

49

hk_

41

hk_

30

hk_

19

hk_

33

hk_

35

hk_

38

hk_

48

hk_

39

hk_

51

hk_

31

hk_

44

hk_

52

Sensitivity_conductivity

629

43

0

100

200

300

400

500

600

700

Local Execution with 1 core HPC Execution with 24 cores on Azure

Computational performance of PEST for sensitivity analysis to

HPC on Azure

Tim

e in

Ho

urs

Case Study 2:Thorium Reactive Transport Modeling

Baotou tailings pond, one of the largest

tailings in China piled above ground

surfaces.

Of greatest concern is the potential for

radioactive pollution of the Yellow River

nearby which is the primary water

source for 150 million people.

Most productive “secondary mines”,

approximately 11 floors high!

Modeling Contaminant Transport and Remediation

Zheng 1990Zheng and Wang 1999Zheng 2010Zheng et al. 2013

MT3DMS

RT3D

MGOSOMOS

SEAM3DSEAWAT

PHT3D

MT3DMS-Based Transport Modeling Tools

Seawater

Intrusion

Biodegradation

kinetics +

transport

Geochemical

reactions +

transport

Bio-chemical

reactions +

transport

Management

Optimization

Management

optimization

and monitoring

network design

Tailings pond leakage

(Henning Prommer, 2013)

Numerical model Transported chemical undergoes

surface complexation, with mineral

dissolution/precipitation.

R

Governing equation::

Single species transport

model in MT3DMS with

advection, multiple

species reactive modeling

in PHREEQC.

Architecture of Modeling Application on Azure

Simulation Results

Conservative transport plumes

Thorium reactive transport plumes

Sensitivities of thorium mass-fluxes into collection-trench

to 24 hydrogeological parameters

FD, TVD, split-operator algorithm

72 layers, 745 columns

Execution Time and Costs

Acknowledgements

National Natural Science Foundation of China (grant no. 91225301)

Microsoft Research Asia(grant no. CNIC-MSR2013)

Our cloud modeling research is supported by

PKU: Guangjun Zhang; Yingying Yao; Xiang Huang; Yong TianMicrosoft Resxearch Asia: Lily Sun; Xin Ma

Microsoft Privacy Policy statement applies to all information collected. Read at research.microsoft.com

Save the planet and return your name badge before you

leave (on Tuesday)