Rainfall Estimation from Satellite Data Satellite Applications Workshop, September 2003 Beth Ebert BMRC

Outline Rain estimation systems Satellite rain estimation methods – how they work and when to use or not use them –Geostationary (VIS/IR) –Passive and active microwave –Estimates using combined sensors Accuracy of satellite rainfall estimates Tropical Rain Potential (TRaP)

User requirements Short-term needs Nowcasting of severe storms Weather forecasting Initialisation of NWP River management Flood control Medium-term needs Intraseasonal variability Agriculture Long-term needs Climate change Hydrological planning

Rain measurement systems - rain gauges Advantages: “True” measurement of rain Disadvantages: No coverage over oceans or remote regions Point measurement not representative of area Wind  underestimates of rain Different gauge designs

Rain measurement systems - radar Advantages: Excellent space and time resolution Observations in real time Disadvantages: Little coverage over oceans or remote regions Signal calibration Corrections required for beam filling, bright band, anomalous propagation, attenuation, etc. Z-R relationship Expensive to operate

Rain estimation from model Advantages: Excellent space and time resolution Estimates in real time Includes meteorological context from other model fields Disadvantages: Forecast, not observation Model does not represent processes perfectly

When would you use satellite rainfall estimates in real time? When you don't have gauge or radar data When you don't entirely trust the model To get supporting evidence for model predictions Typical situations: –Tropical convection –Mid-latitude convection –Storms moving on shore from sea (especially tropical cyclones)

IPWG

Rain measurement systems - geostationary satellite (VIS/IR) Advantages: Good space and time resolution Observations in near real time Samples oceans and remote regions Consistent measurement system Disadvantages: Measures cloud-top properties instead of rain  May mistake cirrus for rain clouds  Does not capture rain from warm clouds

VIS/IR rainfall estimates Principle: Rainfall at the surface is related to cloud top properties observed from space: VIS reflectivity Brighter (thicker clouds)  heavier rainfall Dark  no rain IR brightness temperature Colder (deeper clouds)  heavier rainfall Warm  no rain NIR brightness temperature |T NIR -T IR |~0 (large drops or ice)  rain more likely |T NIR -T IR |>0 (small water drops)  no rain

GOES Precipitation Index (GPI) Simple threshold method: R = 3.0 mm/hr * (fraction of pixels with T B  235K) Arkin (1979) Works better over large areas and long times (i.e., monthly) Better suited to convective rainfall

Global annual rainfall from GPI Bob Joyce, NCEPhttp://tao.atmos.washington.edu/data_sets/gpi/

IR imageRain rate Power law technique R = a (T 0 -T B ) b - R 0 ( T B  253K)

Auto-Estimator Based on Scofield’s NESDIS Operational Convective Precipitation Estimation Technique R = R fit * RH correction factor * growth correction factor

GOES Multispectral Rainfall Algorithm (GMSRA) Rain indicator: VIS:   0.40 NIR: r e (eff. radius)  15  m OR T 11 -T 6.7 : Negative for deep convective cores (T 11 < 230K) Rain amount: R = probability of rain(T 11 ) * mean rain rate (T 11 ) * RH correction factor * growth correction factor

Rain measurement systems - passive microwave from polar orbiting satellite Advantages: Samples remote regions Consistent measurement system More physically based, more accurate than VIS/IR estimates Disadvantages: Poorer time and space resolution (~3 hr, ~5-50 km) Not a direct measurement of rain Beam filling Does not capture rain from warm clouds over land

Rain measurement systems - passive microwave from polar orbiting satellite Principle: Rainfall at the surface is related to microwave emission from rain drops (low frequency channels) and microwave scattering from ice (high frequency channels): Low frequency (emission) channels - ocean only Warm  many raindrops, heavy rain Cool  no rain High frequency (scattering) channels Cold  scattering from large ice particles, heavy rain Warm  no rain Excellent reference:

Special Sensor Microwave Imager (SSM/I) Spatial resolution 25 km 12.5 km

Special Sensor Microwave Imager (SSM/I) Ferriday and Avery, 1994 OCEAN LAND

Special Sensor Microwave Imager (SSM/I) PRODUCT: RAIN RATE (mm/hr) DATA FOR JULIAN DATE SATELLITE F15 IN ASCENDING NODE SI = a 0 + a 1 T 19V + a 2 T 22V + a 3 T 22V 2 - T 85V R = a SI b NOAA algorithm: {

Special Sensor Microwave Imager (SSM/I) Profiling algorithms: Iteratively match 7-channel T B observations to theoretical values computed from radiative transfer calculations and mesoscale cloud model (table look-up). Use cloud model to estimate rain rate Basis for TRMM algorithm Kummerow et al., 1994

Advanced Microwave Sounding Unit (AMSU) AMSU-A (~50 km spatial resolution): GHz GHz GHz 1589 GHz AMSU-B (~17 km spatial resolution): 189 GHz 2150 GHz 3-5~183.3 GHz (water vapour line)

Advanced Microwave Sounding Unit (AMSU) Rain rate is based on estimated ice water path (IWP) and rain rate relation derived from the MM5 cloud model data. RR = a 0 + a 1 IWP + a 2 IWP 2 (browse images-rain)

McIDAS – USPOL/AMSURR Updated hourly, written at 6, 18 UTC

Tropical Rain Measuring Mission (TRMM) TRMM Microwave Imager (TMI), 780 km swath: Band Frequency Polarization Horiz. Resol. (GHz)(km) V, H V, H H V, H V, H 4.4 Precipitation Radar, 220 km swath: Horizontal resolution of 4 km Profile of rain and snow from surface to ~20 km altitude » Use precipitation radar to tune TMI rain

Tropical Rain Measuring Mission (TRMM) “Instantaneous” rain rate

McIDAS – USPOL/TRMMRAIN Updated hourly, written at 6, 18 UTC

V D TRaP = (R avg X D) V -1 Hurricane Georges DMSP SSMI Rain Rates 1436 utc Sept Tropical Rainfall Potential (TRaP)

TRaP – TC Sam, December 2000 SSM/I rain rate 24 hr rain estimate

Satellites used to perform TRaP DMSP SSMI NOAA AMSU NASA TRMM NESDIS AE Resolution 15km 16km 5km 4km Frequency 1-2 per 12hrs 1 per 6-12hrs 1 per 24hrs 1 per 30min # Satellites Max RR 35mm/hr 20mm/hr 60mm/hr 50mm/hr Priority Slides courtesy of Sheldon Kusselson, NOAA/NESDIS Satellite Services Division

Future TRaP Initiatives Continued validation Automated operational TRaP products for: ALL STORMS... ALL THE TIME... WORLDWIDE AMSU-b Rain Rates Chantal AMSU-b TRaP SSM/I Rain Rates SSM/I TRaP Barry TRMM Rain Rates TRMM TRaP Humberto AE Rain Rates AE TRaP Helene TRaP Training:

Combined geostationary / passive microwave rainfall estimates Combines the best features of both approaches: Good space/time resolution of geostationary estimates Better accuracy of microwave estimates How to do the combination? 1. Blend rain estimates using weighted averages 2. Use matched VIS/IR and microwave image set to: (a) get a field of multiplicative correction factors (b) recalibrate VIS/IR algorithm coefficients (c) map IR T B onto microwave rainrates (d) adaptive neural network (e) morph microwave rainfall in space/time

Global Precipitation Climatology Project (GPCP) Weighted average of rain estimates from: IR (GPI), SSM/I, TOVS, rain gauges Products available from 2.5° monthly resolution: monthly average rain rate 4-, 8-hour lag correlations of rain rate standard deviation of instantaneous rain ratefrequency of rain sampling error for the monthly rain rate estimatefractional rainy area algorithm error for the monthly rain rate estimatenumber of available samples

Global Precipitation Climatology Project (GPCP)

Combined sensors Monthly mean rainfall Weighted average of TRMM, SSM/I, IR, rain gauges

Near real time SSM/I+TRMM 3-hourly rainfall Maps IR T B onto microwave rainrates, uses microwave in preference to IR where available

NRL hourly rainfall - "Blend" Maps IR T B onto microwave rainrates

NRL hourly rainfall - "Merge" Weighted average of several microwave rain estimates

PERSIANN Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks Uses an adaptive neural network to update the coefficients of an IR algorithm using surface or passive microwave rainfall observations.

Actual Microwave Observations t+0t+2 hrs t+ 1/2 hrt+1 hrt+1.5 hr IR t+1/2 hrt+1 hrt+1.5 hr CMORPH Interpolated “observations”

How accurate are satellite rain estimates?

Daily verification

Summer Winter 2003  wave IR  wave+IR

IR+microwave blending techniques Australian summer RMS (%)Correlation Eq. threat score Recalibrate IR coefficients (GPCP 1DD)73% Map IR T B to  wave RR (NRL "blend") 75% Adaptive neural network (U. Ariz. PERSIANN)69% Morphing (NOAA CPC CMORPH)--- NWP (mesoLAPS)63%

IR+microwave blending techniques Australian winter 2003 RMS (%)Correlation Eq. threat score Recalibrate IR coefficients (GPCP 1DD)55% Map IR T B to  wave RR (NRL "blend") 45% Adaptive neural network (U. Ariz. PERSIANN)38% Morphing (NOAA CPC CMORPH)34% NWP (mesoLAPS)29%

GPCP Algorithm Intercomparison Project results (1990's) Skill greater over tropics than over higher latitudes Passive microwave algorithms give most accurate instantaneous rain rate, esp. outside tropics Geostationary algorithms give best monthly estimates due to better sampling More recent IPWG results suggest that combining lots of microwave estimates overcomes microwave sampling limitations

Case study – 24 January TRMM mosaic

(go to training/case0.html)

24 hours totals 24 January 2003

Forecast situation: Convection in remote region with potential flash flooding Which satellite guidance would you choose? Want rapid time sampling  geostationary imagery Blended IR+microwave better than IR-only If not available then choose IR power-law algorithm, do a "reality check" against microwave estimates when possible

Forecast situation: Tropical storm moving onshore Which satellite guidance would you choose? Rapid time sampling perhaps less critical Over the ocean microwave-only may be better than blended IR+microwave TRMM most accurate but worst sampling; AMSU and SSM/I have better sampling If named tropical cyclone, then use TRaP

Forecast situation: Mid-latitude cold front moving onshore Which satellite guidance would you choose? Want good time sampling  geostationary imagery Blended IR+microwave better than IR-only The models generally handle this situation quite well – you may not need the satellite estimates

Forecast situation: Shallow rain showers moving onshore Which satellite guidance would you choose? NONE!! Reasons: IR algorithms only expect rain in deep systems Although microwave instruments can measure warm rain over the see, the microwave footprint is to big to "see" small-sized showers

Forecast situation: Orographic rainfall in the mountains Which satellite guidance would you choose? None, unless convection also developed Reasons: IR algorithms only expect rain in deep systems Over land microwave instruments cannot measure rain from warm-topped clouds.

Satellite rainfall estimates over Australia available within a few hours of real time McIDAS AMSU TRMM (Don't currently have an IR estimate…  ) Web-based NRL "blended" and "merged" IR+microwave TRMM-based hourly IR and 3-hourly microwave and IR+microwave

Global Precipitation Measurement Mission (GPM) In this configuration the "core" spacecraft serves as a high quality reference platform for training and calibrating the passive microwave rain retrieval algorithms used with the "constellation" radiometers. Radar + passive microwave radiometer Passive microwave radiometers