IPWG training course: Global Satellite Mapping of Precipitation (GSMaP) project Kazumasa Aonashi (MRI/JMA) [email protected]
IPWG training course:
Global Satellite Mapping of Precipitation (GSMaP) project
Kazumasa Aonashi (MRI/JMA) [email protected]
• Artists like Andoh
Hirosige appreciated rain. • But…
Motivation:
Motivation: Needs for Precip. Remote-sensing
• Global precipitation monitoring is very important for disaster prevention, water resource management, …..
• Direct precipitation observation data are sparse, especially over oceans.
Outline of Lecture
• Overview of GSMaP • MWR algorithms • Microwave-IR merged algorithms • Validation • Next version of GSMaP • Summary
Inhomo-geneity of the data quality
Overview of GSMaP
Algorithms Products
GCOM-W AMSR2
DMSP SSM/I, SSMIS
Global Rainfall Map + Gauge-calibrated
Rainfall Map (0.1 degree grid, Hourly)
Rainfall Data from each Microwave Radiometer
Merged Microwave Rainfall Data
NOAA/MetOp AMSU
Overview of GSMaP Algorithm
(Okamoto et al. 2005, Kubota et al, 2007, Aonashi et al. 2009, Ushio et al. 2009, Shige et al. 2009, Kachi et al. 2011)
IR Imagers
Microwave-IR Merged Algorithm (CMV, K/F)
GSMaP Microwave Radiometer Retrieval Algorithm
Good: high-frequent (wide swath, multi-satellites) Bad: cannot measure vertical structure (need info. from radar)
Microwave Imagers & Sounders
GPM-Core GMI
TRMM PR
Precipitation Radars
GPM-Core DPR
Data Base
Geostationary Satellites
http://sharaku.eorc.jaxa.jp/GSMaP/
Nov. 12, 2015 MRI JICA lecture
Correlation of satellite-estimated and Radar-estimated instantaneous rain rates vs normalized RMS differences
(Ebert & Manton, 1998)
Observed TBs
Precip.
Ts, Temp Precip Profiles DSD Mixed phase inhomogeneity
Look-up Table
Forward calculation Retrieval Calculation
Basic Idea of the MWR Algorithm
RTM Screening Inhomogeneity estimation Scattering part Emission part
Find the optimal precipitation that gives RTM-calculated TBs fitting best with the observed TBs: PCT37, PCT85 (land) TB10v,TB19v, TB37v, PCT37, PCT85 (sea)
GANAL
Statistical Precip-related Variable Models (PR)
Nov. 12, 2015 MRI JICA lecture
Physical Basis of Microwave Precip. Retrival
• Over Land: Scattering by frozen
particles (Higher Freq.) • Over Ocean: Scattering (Higher Freq.) + Emission from Rain (Lower Freq.)
MWR retrievals over ocean are more accurate than those over land and coast.
GCOM-W AMSR2
DMSP SSM/I, SSMIS
Global Rainfall Map + Gauge-calibrated
Rainfall Map (0.1 degree grid, Hourly)
Rainfall Data from each Microwave Radiometer
Merged Microwave Rainfall Data
NOAA/MetOp AMSU
Overview of GSMaP Algorithm
(Okamoto et al. 2005, Kubota et al, 2007, Aonashi et al. 2009, Ushio et al. 2009, Shige et al. 2009, Kachi et al. 2011)
IR Imagers
Microwave-IR Merged Algorithm (CMV, K/F)
GSMaP Microwave Radiometer Retrieval Algorithm
Good: high-frequent (wide swath, multi-satellites) Bad: cannot measure vertical structure (need info. from radar)
Microwave Imagers & Sounders
GPM-Core GMI
TRMM PR
Precipitation Radars
GPM-Core DPR
Data Base
Geostationary Satellites
http://sharaku.eorc.jaxa.jp/GSMaP/
Present IR
GEO IR data
Cloud motion vectors
Past GSMaP data
MWR data observed during present 1 hour
Meridional
Zonal
Present MWR data
GSMaP interpolated by the motion vectors
Kalman filter
1-hour-before IR
1-hour-before GSMaP
Present GSMaP
(MWR overpasses)
(Outsides MWR overpasses)
Flowchart of MWR-IR merged algorithm (GSMaP_MVK algorithm)
Ushio et al. (2009)
MWR-IR merged algorithm To fill the temporal gaps, we developed morphing techniques using Geo IR data. But, the precipitation retrievals loose their quality by +- 90 minutes from the MWI observation time.
Note: High-resolution conical-scanning MWI (TMI) gives retrievals that have the highest correlation with precipitation observation.
GPM-GSMaP Product list
Product name Variables Horizontal resolution
Temporal resolution
Latency Correction
L3 GSMaP Hourly
Hourly Precip Rate (GSMaP_MVK)
0.1×0.1 deg.lat/lon
1 hour 3 days None
Gauge-adjusted Hourly Precip Rate (GSMaP_Gauge)
adjusted by daily rain gauges (NOAA CPC Gauge-Based Analysis, Chen et al. 2008)
Product name Variables Horizontal resolution
Temporal resolution
Latency Correction
L3R GSMaP Hourly
Hourly Precip Rate (GSMaP_NRT)
0.1×0.1 deg.lat/lon
1 hour 4 hours None
Gauge-adjusted Hourly Precip Rate (GSMaP_Gauge_NRT)
Correction by empirical coefficients
Standard product (Latency: 3 days)
Near-real-time product (Latency: 4 hours)
GPM-GSMaP data during Mar. 2000-Feb. 2014 period was reprocessed as reanalysis version (GSMaP_RNL), and was open to the public on Apr. 2016.
You can access the products via: http://sharaku.eorc.jaxa.jp/GSMaP/index_e.htm
GSMaP real-time version (GSMaP_NOW) • To reduce latency from 4-hr to “quasi-realtime”
– Using data that is available within 0.5-hour (GMI, AMSR2 direct receiving data, AMSU direct receiving data and Himawari-IR) to produce GSMaP at 0.5-hr before (observation).
– Applying 0.5-hour forward extrapolation (future direction) by cloud motion vector to produce GSMaP at current hour (just now).
• Since Nov. 2015, web site and data (GEO-Himawari region) are open to the public from http://sharaku.eorc.jaxa.jp/GSMaP_NOW/
“GSMaP_NOW” product
Dr. Yamamoto Orographic rainfall correction method
Prof. Shige Orographic rainfall correction method
Dr. Aonashi Current reader, Imager algorithm
Ms. Kachi NRT system
Prof. Takahashi Precipitation physical modeling
Dr. Akimoto Surface emissivity
Dr.Arai NRT system
Drs.Hamada and Yokoyama Precipitation type classification
Dr. Mega MWR-IR merged algorithm, Gauge-adjustment method
Prof. Takayabu Precipitation type classification
Prof.Hirose Precipitation type classification
Group Photo 2016/6/3
Ms. Yamaji DSD DB
Prof. Seto Rain/No-rain classification method
Prof. Higuchi IR algorithm
Prof. Ushio MWR-IR merged algorithm, Gauge-adjustment method
Kubota(me) Imager/sounder algorithm
MWR algorithm
Basic Idea Precipitation cloud models Orographic rain algorithm
• Precipitation structures (Precipitation Profile, Melting layer, DSD) • Atmospheric variables (temperatures,…)
Look-up Table
Retrieval Algorithm
The relationship between rain rate and brightness temperature is tabulated by assumption of precipitation physical model and calculation of the radiative transfer model (RTM).
Precipitation physical model
(Aonashi and Liu 2000, Kubota et al. 2007, Aonashi et al .2009)
RTM calculation
Basic Idea of GSMaP_MWR algorithm
Observed Brightness
Temp.
Rainfall rate
Microwave Radiometers
TRMM/PR (GPM/DPR) database in the algorithm
LUT for convective rain
Hydrometeor profile for
Stratiform rain
Hydrometeor profile for
Convective rain
Convetive/stratiform ratio (Currently, statistical weights are adopted.)
・No brightband model ・Convective DSD model
・Brightband model ・Stratiform DSD model
LUT for stratiform rain
LUT for retrievals • Each frequency • 5.0 × 5.0 deg. lat-lon boxes • 6-hourly
Effects of vertical profiles
Effects of conv/strat ratio
Effects of drop size distribution (DSD)
Look-up table (LUT)
Rain rate(R)
Brig
htne
ss T
emp(
Tb)
• Rain/no-rain classification method • Correction of LUT using estimated
inhomogeneity
Rain estimates by the retrievals
Nov. 23 2012 IHP training course@Nagoya Univ.
Precip type classification
Precip Profile
Data base
Precip profile data base
Precipitation Profile Model
Rainfall rate [mm/h]
Hei
ght
from
1 d
eg
level
[k
m]
1℃ level Example:
TRMM PR averaged preciptation profiles
for each type, surface precip, conv/stra
(land) 0: thunderstorm, 1: shower, 2: shallow, 3: frontal rain, 4: organized rain 5: highland (sea) 6: shallow 7:frontal rain, 8:transit, 9:organized rain
10 types (land 6, sea 4) are classified from TRMM PR data (2.5 deg, 3 monthly)
Microwave-IR merged algorithms
Basic idea Use of rain guages
Present IR
GEO IR data
Cloud motion vectors
Past GSMaP data
MWR data observed during present 1 hour
Meridional
Zonal
Present MWR data
GSMaP interpolated by the motion vectors
Kalman filter
1-hour-before IR
1-hour-before GSMaP
Present GSMaP
(MWR overpasses)
(Outsides MWR overpasses)
Flowchart of MWR-IR merged algorithm (GSMaP_MVK algorithm)
Ushio et al. (2009)
+ + +
= = =
t t+1 t+2
Validation of the current products
Global validation using TRMM
Regional Validation using ground-based observation
Monthly Sensor map (Jul. 2013)
AMSR2 TMI
SSMIS F16
SSMIS F18
SSMIS F17
MHS Metop-A
MHS NOAA19
Zonal mean analysis (Jul. 2013)
Monthly merged map (Jan. 2013)
MVK_FW
MVK_BW
MVK(CMB)
GSMaP_ Gauge
MWR
PR
CPC Gauge
Zonal mean analysis (Jan. 2013)
GSMaP_MVK (V6) Radar-AMeDAS (5 Jul. 2013)
Comparisons in daily averaged rainfall estimates around Japan with 0.25 x 0.25 deg. resolution with reference to the gauge-calibrated
ground radar dataset (JMA Radar-AMeDAS precipitation analysis).
Validation using JMA Radar-AMeDAS analyssis
(White: missing values)
The evaluation system at the Kyoto Univ. was imported to the EORC.
• List of Statistics – Correlation coefficient
– Root Mean Square Error (RMSE)
– Probability of Detection(POD) • The hit rate, an index of rain detection
ability.
• POD=a/(a+c)
– False Alarm Ratio(FAR) • An index of false rainfall.
• FAR=b/(a+b)
– Equitable threat score, frequency bias (not shown here)
radar observing
rain
radar observing no-
rain satellite estimate
giving rain a b
satellite estimate giving no-rain
c d
2 × 2 contingency table
Statistics were computed for daily estimates. 15-day-running mean was adopted.
Method : Statistics
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Corr
ela
tion
Date
MVK_Correlation
Gauge_Correlation
MFW_Correlation
NRT_Correlation
MVK_15日移動平均
Gauge_15日移動平均
MFW_15日移動平均
NRT_15日移動平均
Results: Correlation coefficients
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
RMSE
(mm
/hou
r)
Date
MVK_RMSE
Gauge_RMSE
MFW_RMSE
NRT_RMSE
MVK_15日移動平均
Gauge_15日移動平均
MFW_15日移動平均
NRT_15日移動平均
Results: RMSE
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
PO
D
Date
MVK_POD
Gauge_POD
MFW_POD
NRT_POD
MVK_15日移動平均
Gauge_15日移動平均
MFW_15日移動平均
NRT_15日移動平均
Results:Probability of Detection(POD)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
FAR
Date
MVK_FAR
Gauge_FAR
MFW_FAR
NRT_FAR
MVK_15日移動平均
Gauge_15日移動平均
MFW_15日移動平均
NRT_15日移動平均
Results: False Alarm Ratio(FAR)
Next version
GSMaP V04 [Algorithm version V7] is scheduled to be released
in Dec. 2016.
Utilization of NOAA/NESDIS Snow/Ice Cover Maps
• False rainfall signals were sometimes found over surface snow or sea ice. – related to difficulty to distinguish falling snow and surface snow in MW
frequencies such as 89GHz. • MW 165 and 183GHz channels are preferable for falling snow measurement,
but several microwave imagers are lack of them.
• Kubota is now trying to use NOAA/NESDIS Snow/Ice Cover Maps (autosnow) in order to mitigate of the misidentification.
Example of autosnow data (white: snow, sea ice: yellow)
Snowfall estimation with Prof. Liu’s method (Kubota)
GSMaP
GSMaP +
Liu
grey: missing values
• In the current GSMaP, no snowfall estimates (missing values) Kubota is now trying to integrate the method by Prof. Liu (Florida State University) into the GSMaP algorithm.
• Snow-Rain Separation method (Sims and Liu 2015)
• Snowfall estimation (Liu and Seo 2013)
Development of classification in precipitation types using GPM/KuPR
• Takayabu, Hamada, Yamaji, and Kubota are now developing precipitation type/profile/DSD database using GPM/DPR data for the GSMaP_MWR algorithm
OLD NEW Precip type for Jan
Land
Sea
Statistical error analyses of the forward-calculated MWI TBs
• The forward calculation using the conventional first guess of CLWC overestimated TBs with very weak precipitation.
GMI TB37v for Pr (0-0.2 mm/h) vs. PWC over Sea (prtype=8) TBo: Blue, TBc (0.5 kg/m2) : Green, TBc (CLWC=0) : brown
• The forward calculation from the PR surface precipitation tended to underestimate subtropical TBs with large PWC..
PR Rainsurf vs. TMI (TB19v-TBc0mm) for SST (290-295 K) over Sea (prtype=11) TBo: Red, TBc (CLWC=0.5 kg.m2):Green
Modification of the first guess of physical variables
• We set the CLWC first guess as a function of PWC and SST.
• This improved the precipitation retrieval by enlarging weak precipitation areas and reducing positive biases for subtropical precipitation.
Monthly mean Precip. (mm/dy) for Jul. 13 (up) TRMM PR 2A25 (mid) TMI conventional (dwn) TMI improved
NRT basis rain-gauge (empirical) adjustment
• Ushio and Mega will substantially improve NRT basis rain gauge (empirical) adjustment method in the next version.
Flow of GSMaP Gauge NRT
30 days
c, σv, (σw, μw)
GSMaP Gauge NRT
1 2 3 ..………….. 29 30 31 32 33 day
GSMaP NRTGSMaP MVKGSMaP Gauge
GSMaP Gauge NRT model
24 hour
By Dr. Mega and Prof. Ushio (Osaka Univ.)
Summary
• Overview of GSMaP • MWR algorithms • Microwave-IR merged algorithms • Validation • Next version of GSMaP
Thank you for your attention!