role of observations in model testing, parameterization &...

Center for Hydrometeorology and Remote Sensing, University of California, Irvine

Role of Observations in Model testing, Parameterization & Modification: Case

Studies from Hydroclimatology Soroosh Sorooshian

Center for Hydrometeorology and Remote Sensing University of California Irvine

1st Workshop on: Understanding Climate Change from Data August 15-16 2011, Univ. of Minnesota


University of California Irvine (UCI) and Arizona (UA)

and many more …

CHRS & Affiliates: A truly International Team


PERSIANN System “Estimation” Global IR

MW-RR (TRMM, NOAA, DMSP Satellites)

Merged Products - Hourly rainfall - 6 hourly rainfall - Daily rainfall - Monthly rainfall

ANN

Error Detection

Quality Control

Merging

Sate

llite

Dat

a G

roun

d O

bser

vatio

ns

Products

High Temporal-Spatial Res. Cloud Infrared Images

Feed

back

Hourly Rain Estimate Sampling

MW-PR Hourly Rain Rates (GSFC, NASA; NESDIS, NOAA)

Hourly Global Precipitation Estimates

Gauges Coverage

GPCC & CPC Gauge Analysis

Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks (PERSIANN)


(CPC, NOAA)

Center for Hydrometeorology & Remote Sensing, University of California, Irvine

http://chrs.web.uci.edu/

Real Time Global Data: Cooperation With UNESCO


Short Range Long Range

hours days weeks months seasons years decades

Required Hydrometeorologic Predictions

Forecast Requirements

Short-range Mid-range Long-range


Recent Assessment of Climate Models

Regional trends in extreme events are not always captured by current models

It is difficult to assess the significance of these discrepancies and to distinguish between model deficiencies and natural variability

How Accurate Are Global Climate Models?


Recent Assessment of Seasonal Climate Forecasts •“ of the dozens of forecast techniques proffered by government, academic, private-sector climatologists, all but two are virtually useless, according to a new study” Livezey &Timofeyeva - BAMS,

June 2008. • “About the only time forecasts had any success predicting precipitation was for winters with an El Nino or a La Nina”

Quoting from Science, Vol. 321, 15th August 2008


Status of Forecast Skill in Hydrologic Models


MODEL

PARAMETER ESTIMATION

DATA

If the “World” of Watershed Hydrology Was Perfect!

Hydrologic Modeling: 3 Elements!


Hydrologic Modeling


UCt

UM

LM

Streamflow

Rainfall

percolation

LCt

UK

LK

PD = f(Z, X)

A look into the “heart” of R-R Models

Percolation Process is the Core element in Partitioning the rain between the various stores

PD = f(Z, X)


Model Calibration


The Concept of Model Calibration Measured Outputs

Yt

t

Real World

Measured Inputs

MODEL (θ) Computed Outputs

Prior Info θ

Computed Outputs

+ -

Optimization Procedure θ

“Calibration: constraining the model to be consistent with observations”


Calibration components

Objective Function

Search Algorithm

Sensitivity Analysis

Problems with identifiability


Data information content

“Bucket Model” Simple two parameter Model

Cm

ax

K S

P

Q


Data information content C

max

K S

P

Q

Multiple spills

Cmax

K

No spills

Cmax

K

Cm

ax

K S

P

Q


Parameter X

Obj

ectiv

e Fu

nctio

n

True Parameter Set

The Ideal case: Convex Optimization

Created By G-H Park


Obj

ectiv

e Fu

nctio

n

Parameter X

True Parameter Set

Difficulties in Global Optimization

Created By G-H Park


Parameter X

Obj

ectiv

e Fu

nctio

n

Global Optimum

Created By G-H Park

Parameter Estimation (non-convex, multi-optima)


SAC-SMA model 13 parameters to be estimated

The U.S. National Weather Service Flood Forecast Model ( SAC-SMA)


Optimization Strategy – Local Direct Search

Calibration of the Sacramento Model Downhill Simplex Method, Nelder & Mead, 1965

Duan, Gupta, and Sorooshian, 1992, WRR


More than one main convergence region

1.- Regions of Attraction

2.- Local Optima

Many small "pits" in each region

Difficulties in Optimization



2.- Local Optima




3.- Roughness Rough surface with discontinuous derivatives




4.- Flatness Flat near optimum with significantly different parameter sensitivities

3.- Roughness Rough surface with discontinuous derivatives

2.- Local Optima





5.- Shape Long and curved ridges



The SCE-UA Algorithm …

Duan, Sorooshian, and Gupta 1992, WRR

The Shuffled Complex Evolution Algorithm


Global Optimization – The SCE-UA Algorithm

Simplex Method

Shuffled Complex Evolution (SCE-UA)

Duan, Gupta & Sorooshian, 1992, WRR


Evolving Directions

Advances in Parameter Estimation


Land-Surface Model

RHODE ISLANDCONNECTICUT

FLORIDA

MISSISSIPPI

WESTVIRGINIA

WASHINGTON

OREGONIDAHO

MONTANA

WYOMING

NORTHDAKOTA

SOUTHDAKOTA

NEBRASKAIOWA

MINNESOTA

WISCONSIN

ILLINOIS

INDIANA OHIO

MISSOURIKANSASCOLORADO

UTAHNEVADA

CALIFORNIA

ARIZONA NEW MEXICO

OKLAHOMAARKANSAS

KENTUCKY

VIRGINIA

TEXAS

GEORGIA

ALABAMA

SOUTHCAROLINA

NORTH CAROLINA

TENNESSEE

PENNSYLVANIA

MARYLAND

NEW JERSEY

NEW YORK

VERMONT

MAINE

DELAWARE

LOUISIANA

MICHIGAN

MASSACHUSETTSNEW HAMPSHIRE


Multi-Objective Approaches

M(θ)

Model

Radiation

Inputs Outputs


Multi-Objective Optimization Problem

This image cannot currently be displayed.

MOCOM Algorithm:

Does NOT require conversion to a sequence of single optimization problems

Simultaneously finds several Pareto Solutions in a Single Optimization


ARM-CART SGP Site

100 km

Grid: ~100,000 km2

Luis A. Bastidas Z. ([email protected])


{H, Tg, Sw} {λE} {Tg} {Sw}

Single- & Multi-Flux Calibrations {H}

λE

Tg

H

Sw

Computed

Obs

erve

d


Large-Scale Irrigation and Incorporation in Models

Impact of Irrigation


Irrigation over central California Irrigation areas

122W 120W

39N

36N

CIMIS stations

• Meteorological conditions are the key factors to decide when and how much water to apply, • Californian Irrigation Management Information System (CIMIS) with more than 200 stations (nearly 150 active) provides the information to farmers. • So far, not many researchers have use this dataset to evaluate the climate variability and climate model output in this region


1- “Control” run: without modifying anything (MM5-C). Model makes no special consideration for irrigation ( no water added) 2- “Field Capacity” run: setting up root-zone soil moisture at Field Capacity at each time step (MM5-F) 3- “Recommended” run: setting up root-zone soil moisture based Hanson et al (2004) method (MM5-R).

• Irrigation starts when the root zone’s relative available Soil-

Water (SW) content is less than maximum allowable water depletion of soil (i.e. adding water into model soil);

• Irrigation ends when soil moisture reaches field capacity (i.e.

stop adding water).

Modeling Study Setup Using MM5


• MODIS ET: 0.05-degree, monthly (Tang et al. 2009) http://ftp.hydro.washington.edu/pub/qiuhong/usa/ • MODIS skin temperature https://lpdaac.usgs.gov/lpdaac/get_data/ • GLDAS-1 ET: 0.25-degree, monthly http://disc.sci.gsfc.nasa.gov/hydrology/data-holdings • Ameriflux at Blodgett Forest site http://public.ornl.gov/ameriflux/ • California Irrigation Management Information system (CIMIS) data http://www.cimis.water.ca.gov/cimis/data.jsp

Data Sources:

http://ftp.hydro.washington.edu/pub/qiuhong/usa/

https://lpdaac.usgs.gov/lpdaac/get_data/

http://disc.sci.gsfc.nasa.gov/hydrology/data-holdings

http://public.ornl.gov/ameriflux/

http://www.cimis.water.ca.gov/cimis/data.jsp


MM5-R

122W 120W

Mean skin surface temp. at daytime in June, July and August, 2007.

MODIS

39N

36N

122W 120W

Skin Temp. (oC)

MM5-C NARR

122W 120W 122W 120W

Adding irrigation into RCM (MM5), Improves the model’s ability to simulate the temperature patterns observed by MODIS

Sorooshian et al, (JGR 2011)


GLDAS MM5-C MM5-R MODIS

Monthly ET (mm/month)

Actual ET comparison-spatial distribution – July 2007

- with more realistic irrigation scheme, significant improvement in capturing ET over irrigated Central Valley in California:


Thank You For the Opportunity

Somewhere in New Mexico, USA - Photo: J. Sorooshian

role of observations in model testing, parameterization &...

Documents