Remote Sensing Precipitation Using GEO Satellite …ipwg/meetings/tsukuba-2014/training/T1...Center...

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

Remote Sensing Precipitation Using

GEO Satellite Information

Kuolin Hsu Center for Hydrometeorology and Remote Sensing,

University of California, Irvine

The IPWG7 Training Course Program 17-20 November 2014

Tsukuba International Congress Center, Tsukuba, Japan

New and emerging remote-sensing technologies for precipitation data sets and their applications and validation

Session 1: Precipitation Remote Sensing and retrieval algorithms I: Infrared Algorithm

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

Outline

• Precipitation Measurement

• GEO Satellite Information for Precipitation Retrieval

• Precipitation Estimation from Remote Sensing Information using Artificial Neural Networks (PERSIANN)

• PERSIANN-Climate Data Record (PERSIANN-CDR)

• Summary

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

Extreme Precipitation & Flash Flooding

Floods caused by extreme precipitation are the most widespread nature disasters

High spatial and temporal resolution of precipitation measurement is needed for operational hydrology

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

Precipitation Observation

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

Coverage of the WSR-88D and gauge networks

3 km AGL 2 km AGL 1 km AGL

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

Satellite Precipitation Monitoring

Geostationary IR Cloud top heights only 15-30 minute data

Meteosat 7 (EUMETSAT)

TRMM precipitation RADAR 3D imaging of rainfall 1-2 days between overpasses (35°N-35°S only)

TRMM PR (NSA/JAXA)

Passive Microwave (SSM/I) Some characterisation of rainfall ~2 overpasses per day per spacecraft, moving to 3-hour return time (GPM)

SSMI 85GHz (DMSP)

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

TRMM Rain Rate

“Instantaneous” rain rate from TRMM

http://trmm.gsfc.nasa.gov/

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

Typical Microwave Coverage in 3 Hr

TMI – white AMSR-E – medium grey SSM/I – light grey AMSU-B – dark grey

http://trmm.gsfc.nasa.gov/

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

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

+

-

- + +

Interpolation of 3-hour Precipitation

T T+3hr T+6hr t-hr t+3hr

Rain started between 3-hr period

Missed the peak

Rain ended between 3-hr period

Short-life event

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

Rain Estimation from Geostationary satellite (VIS/IR)

Advantages: • Good space and time resolution (half-hour, 4 km) • Observations in near real time • Near global coverage

Disadvantages: • Measures cloud-top properties instead of rain • May mistake cirrus for rain clouds • May not capture rain from warm clouds

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

VIS/IR Rainfall Estimates

• IR brightness temperature Deeper cloudsà colder à heavier rainfall Low clouds à Warm à no rain

• VIS reflectivity Thicker cloudsà brighter: heavier rainfall light clouds à Dark: no rain

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

VIS-IR Image vs. Rainfall

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

VIS-IR Image vs. Rainfall

Cold-Thin clouds

Warm-Thick clouds

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

Infrared image

IR (K)

Visible image

Albedo (%)

Hit Under Estimation Over Estimation

10

20

3

0

40

5

0

Latit

ude

Longitude-130 -120 -110 -100 -90 -80 -70

0 10 20 30 40 50 60 70 80

Before albedo adjustment After albedo adjustment

10

20

3

0

40

5

0

Latit

ude

Longitude-130 -120 -110 -100 -90 -80 -70

0 10 20 30 40 50 60 70 80

Before albedo adjustment After albedo adjustment

Albedo (%)

Multi-spectral Image for Rain Classification

Ali et al., J. Hydrometeorology, 2009

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

Florida : Hurricane Ernesto August 30 2006

Ch 1 : 0.6 µm Ch2 : 3.9 µm Ch3 : 6.5 µm Ch4:10.7µm Ch5 : 13.3µm

f) Ch3+Ch5

Hit Under Estimation Over Estimation

ETS=25 POD=74 FAR=45

ETS=29 POD=77 FAR=42

ETS=27 POD=78 FAR=44

ETS=36 POD=76 FAR=35

ETS=30 POD=80 FAR=42

ETS=30 POD=72 FAR=39

ETS=35 POD=79 FAR=37

ETS=37 POD=78 FAR=35

ETS=37 POD=80 FAR=36

ETS=48 POD=75 FAR=22

ETS=49 POD=79 FAR=24

g) Ch4+Ch5

i) Ch3+Ch4+Ch5

d) Ch5

f) Ch3+Ch5

Ali et al., J. Hydrometeorology, 2009

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

f(.) Nonlinear

multivariate function

GO

ES

IR VIS …

Location Topography Wind flow

Water Vapor . . .

Rainfall Intensity

High quality Rainfall sampling data

Rain gauge, radar, SSM/I, and TRMM

error

A Strategy for Satellite Precipitation Estimation

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

Some IR based Algorithms

• Global Precipitation Index (GPI): Arkin and Meisner, 1987

• Negri-Adler-Wetzel (NAW) technique: Negri et al., 1984

• Convective Stratiform Technique (CST): Adler & Negri, 1988

• AutoEstimator: Vicente et al., 1998

• Hydro-Estimator: Scofield and Kuligowski, 2003

• Tropical Applications of Meteorology using SATellite data (TAMSAT): Grimes et al., 1999

• PERSIANN/PERSIANN-CCS.PERSIANN-MSA: Hsu et al., 1997; Sorooshian et al., 2000; Hong et al., 2004; Behrangi et al., 2009

• and many more …

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

PERSIANN System Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks

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

Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks

PERSIANN System “Estimation” Global IR

MW-RR (TRMM, NOAA, DMSP Satellites)

HyDIS WEB

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

Radar Coverage

Feed

back

Hourly Rain Estimate Sampling

MW-PR Hourly Rain Rates

Hourly Global Precipitation Estimates

Gauges Coverage

Sorooshian et al., BAM, 2000

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

Designing the PERSIANN System

Input feature Classification

Rain Rate Estimation

Error Detection-Correction

Switch

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

Input Variables

Surface Type: Land, Coast Ocean

Tb at the central pixel

Tb and Tb -SD in the 3 x 3 window

Tb and Tb-SD in the 5 x 5 window

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

SOFM Classification Map (After Training) Comparing the rain rate distribution on the output layer with the weight distributions of input variables on the SOFM layer

VTbj VTbk

SOFM Layer

j k

Output Layer

Weight Vectors

Each Neuron for An Input Class

Weight Map of Surface Type

Weight Map of Tb

Weight Map of Tb -SD

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

IPGW Validation of Precipitation Measurement (Australia)

http://www.bom.gov.au/bmrc/SatRainVal Daily Rainfall: January 23, 2005

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

IPWG Validation of Precipitation (US) http://cics.umd.edu/ipwg/us_web.html

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

Rainfall Estimation Using Satellite-Based Cloud Classification Maps

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

· Tb: IR temperature of calculation pixel

· m3x3: Mean temperature of 3x3 pixels

· s3x3: Standard deviation temperature of 3x3 pixels

· m5x5: Mean temperature of 5x5 pixels

· s5x5: Standard deviation temperature of 5x5 pixels

Satellite Image Feature Extraction

200 225 250 275 300 325 oK

Cloud information

Pixel information

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

200 225 250 275 300 325 oK

Cloud Top Temperature Tb (ok)

Rai

nfal

l Rat

e (m

m/h

r)

Cirrus Cloud

Convective Cloud

Cloud Type Classification

Tb–R relationship

Cloud Types and Rainfall Distribution

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

Tb=220K

Tb=235K

Tb=253K

t=t0 t=t1 t=t2 t=tk

c1

c2

ck

Tb (K)

R (mm/h)

200 300 0

80

],,[)( texturepatchgeometrypatchcoldnesspatchVvectorFeaature Îv

T220K

T235K

T253K

KV220

v

KV253

v

KV235

vT253K, t=tk

Patch Classification Patch Feature Extraction

Image Segmentation

Rainfall Estimation

Patch-based Approach (PERSIANN-CCS)

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

400 Tb-R curves from PERSIANN-CCS

model

(1) simple threshold (2) Linear: single line (3) Nonlinear: single

curve

Multiple vs. Single Curve Fitting Models

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

Tb=220K

Tb=235K Tb=253K

Hei

ght

],,[ texturepatchgeometrypatchcoldnesspatchV Îv

T220K

T235K

T253K

KV220

v

KV253

v

KV235

v

T253K, t=tk

Patch Feature Extraction

Cloud IR Image

Image Segmentation T220K

T235K

T253K

KV220

v

KV253

v

KV235

v

Patch index j Patch index k

Patch index j

Patch index k

Features Extraction

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

1 10 20 1

10

20

G0 G1

G2

G3

G4 G5

G6

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

Near Real Time Global Precipitation Data http://hydis.eng.uci.edu/gwadi/

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

Global PERSAINN-CCS Hourly Estimates

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

Six-Hour PERSIANN-CCS Rainfall

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

Continue Development

• Adjust PERSIANN-CCS precipitation estimates using passive microwave rainfall

• Improve rain estimation from warm clouds

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

PERSIANN-CCS Hourly Estimates: Jan 4, 2012

CCS Estimation

CCS (PMW Adjusted) Estimation

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

Change Threshold from 253K to 280K

K

mm/hr

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

PERSIANN Precipitation Climate Data Record Reconstruction of 30-year+ Daily Precipitation Data

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

PERSIANN Precipitation Climate Data Record

• Daily Precipitation Data • Data Period: 1983~2014 • Coverage: 60oS ~ 60oN • Spatial Resolution: 0.25ox0.25o

http://www.ncdc.noaa.gov/cdr/operationalcdrs.html

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

LEO Satellites for Precipitation Estimation Limited PMW Samples

Before year 2000

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

Source: NOAA NCDC

• International Satellite Cloud Climatology Project (ISCCP) 1979 to present

10-km and 3-hour intervals

GOES-11 (135°West) GOES-12 (75°West)

MET-9 (0°East) MET-7(57.5°East)

FY2-C(105°East) MTSAT-1R(140°East)

1. U.S. Geostationary Operational Environmental Satellite (GOES)

2. European Meteorological satellite (Meteosat) series

3. Japanese Geostationary Meteorological Satellite (GMS)

4. The Chinese Fen-yung 2C (FY2) series.

Historical GEO Satellite Data

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

PERSIANN structure in a simple scheme

Global IR

TRMM, DMSP, NOAA Satellites

Parameter Adjustment

Sate

llite

Dat

a

High Temporal-Spatial Res. Cloud Infrared Images

Feed

back

Sampling

Instantaneous PMW Rain Estimates

PERSIANN Hourly Rainfall (0.25ox0.25o)

Tempo-Spatial Accumulation

PERSIANN Monthly Rainfall (2.5ox2.5o)

GPCP Monthly Precipitation (2.5ox2.5o) Bias

Adjustment

Adjusted PERSIANN Hourly Rainfall (0.25ox0.25o)

Products

PERSIANN Adjusted (Monthly Scale)

Bias Adjustment of PERSIANN Estimates

Center for Hydrometeorology & Remote Sensing, University of California, Irvine Center for Hydrometeorology and Remote Sensing (CHRS)

PERSIANN-CDR daily rainfall During Katrina 2005

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

Rain rate (mm/day)

Center for Hydrometeorology and Remote Sensing (CHRS)

Daily Precipitation: Hurricane Katrina, 2005

PERSIANN w/o GPCP adjustment PERSIANN w/o GPCP adjustment

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

App

licat

ions

Drought Management Flood Forecasting Water Resources

Satellite Precipitation Data for Hydrologic Applications A

lgor

ithm

Web

Ser

vice

s

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

Summary Advantages of GEO-based precipitation retrieval: • Good space and time resolution • Observations in near real time • Near global coverage

Improve IR-based estimation by: • Extending from pixel to texture based classification • Extending from single IR band to multi-spectral bands • Integrating information with LEO satellite PMW

measurements • Merging estimation with ground measurements • Applying advanced machine learning methods to learn

cloud-rain system

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

Thanks !!