Research activities in support of precipitation measurement,

analysis and predictionG. Tripoli1, T. Hashino1, W-Y Leung1,

E.A. Smith2 , A. Mugnai3, J. . Hoch1 , M. Kulie1

A.V. Mehta2

1 University of Wisconsin, Madison, Wisconsin2 Goddard Space Flight Center – Greenbelt, Maryland

3 Institute of Atmospheric Sciences and Climate – Rome, Italy

Relevant Activities of MASL (Mesoscale Atmospheric Simulation

Laboratory) • CDRD (Cloud Dynamics and Radiation

Database)• AMPS (Advanced Microphysics Prediction

System)– SHIPS (Spectral Habit Ice Prediction System)– SLIPS (Spectral Liquid Prediction System)– SAPS (Spectral Aerosol Prediction System)– BSSS (Blowing Snow Simulation System)

• CRSDAS (Cloud Resolving Satellite Data Assimilation System)

CDRD Activity• Creation of CDRD

– 4 CRM simulations made daily• 3 at randomly (precipitating) locations around the

globe• 1 at a random location within the Mediterranean

basin

– CRM simulation setup• 2 km grid resolution on an inner grid mesh• Outer grid nested within GFS model output• 24 hour simulation, profiles taken in last 12 hours

CDRD Activity• Creation of CDRD

– Information saved on CDRD for precipitating grid cells within the inner cloud resolving mesh:

• Description of the grid cell:– Simulation number, location and time– Grid dimensions, surface topography, l percent land, albedo,

slope– Surface variables, sst, skin T, soil moisture, etc

• Vertical Profiles– Atmospheric state

» P, T, rv, u,v,w– Microphysics description

» Specific humidity , concentration, density, particle skin temperature, fall velocity

– Derived Radiatiive transfer » Microwqve Brightness temperatures» Reflectivity factors» IR radiance to space

– Grid scale environmental tags– Synoptic scale environmental tags

CDRD Access

• CDRD made available on the web for users to mine and download profiles from the database.

http://mocha.aos.wisc.edu/CDRD/

Uses of the CDRD

• Precipitation or cloud retrieval Schemes– CDRD contains hundreds of thousands and

eventually tens-hundreds of –dertived millions of Global CRM profiles representing-

• Light rain• Snow • All types of thunderstorms• Frontal cyclones• Hurricanes and tropical storms

– Data mining schemes implemented– FTP files

Uses of the CDRD

• Basic Research– The CDRD is a unique tool containing CRM-derived

information relating microphysics, atmospheric state parameters and precipitation rates to:

• geographical, seasonal and synoptic settings • satellite observable radiances • reflectivity

for all varieties of precipitating weather everywhere on the globe over all seasons…Wow!

– Study regional, seasonal or diurnal differences and relationships

– Can look for global relationships among these quantities

Potential ProblemPotential Problem

The formulation of algorithms such as GPROFS6 assume that “the profiles in the model database occur with nearly the same frequency as those found in the region where the inversion method is to be applied.”

THUS………THUS………

Error in retrieval resulting from the use of database Error in retrieval resulting from the use of database profiles inappropriate for the application can be profiles inappropriate for the application can be reduced by only applying locally “relevant” database reduced by only applying locally “relevant” database profiles to the retrieval process.profiles to the retrieval process.

BAYES Theorem – As more constraints are added the more the probability of a “correct” profiles increases

CDRD APPROACHSynoptic Setting

(short-term model

forecast)

CDRD Season

Geographical Location

Database mining

Using CDRD Tags

Satellite Observed Radiance (TBs)

CRD

Retrieved Profiles

Surface Rain Rate

Two Types of Tags saved from CRM simulation and placed in

CDRD• 50 km Tags

– Describe regional synoptic setting– Can be obtained with good accuracy from

most recent global model forecast, eg GFS, ECMWF, etc

• 2 km High Resolution Tags– Accurately calculated in real time (topography

) or available from satellite (eg stratiform cloud fraction)

Mean Sea Level Pressure (hPa)

Freezing Level (m) Surface Theta Gradient

LFC Height (m)

Surface Temperature (F)

Lifted Index 700mb Theta Gradient LCL Height (m)

**U-Wind (m/s) Froude Number Surface Theta-EGradient

Topography Height (m)

**V-Wind (m/s) Surface Theta-E (K) 700mb Theta-EGradient

PBL Height (m)

U Momentum Flux Surface Brunt Vaisala Frequency

** Q Vector Convergence

Richardson Number in the PBL

V Momentum Flux **Temperature Surface Divergence Potential Vorticity Advection at 700

and 250 mb

CIN (J/kg) Potential Vorticity at 700 and 200 mb

Divergence at 700 and 200 mb

Height of Maximum Cape (m)

Maximum Cape (J/kg) Surface Vertical Vorticity

**Vertical Velocity (m/s)

Diabatic Moisture Term

Surface Cape (J/kg) Vertical Vorticity at 700 and 200 mb

Theta-E minimum (K) Latent Heat Term

Kinetic Energy 0-6km Wind Shear 500 and 850 mb thickness (m)

**Specific Humidity

50km Tags50km Tags

Cloud Ceiling (m) ** Temperature (K)

Topography Height (m) ** Specific Humidity Vector

Largest Topography Neighbor Difference (m)

** Vertical Velocity (m/s)

Topography Slope Cloud Fraction

PBL Height Convective Cloud Fraction

Mean Sea Level Pressure (hPA) Stratiform Cloud Fraction

Surface Pressure (hPA)

High Resolution Tags (2km)High Resolution Tags (2km)

Cloud Radiation Tags• Low-Altitude  13.6 GHz Radar Reflectivity Factor (Z)• Mid-Altitude  13.6 GHz Radar Reflectivity Factor (Z)• High-Altitude 13.6 GHz Radar Reflectivity Factor (Z)• Low-Altitude  35.5 GHz Radar Reflectivity Factor (Z)• Mid-Altitude  35.5 GHz Radar Reflectivity Factor (Z)• High-Altitude 35.5 GHz Radar Reflectivity Factor (Z)• 10.65v GHz Brightness Temperature• 10.65h GHz Brightness Temperature• 18.7v GHz Brightness Temperature• 18.7h GHz Brightness Temperature• 23.8v GHz Brightness Temperature• 23.8h GHz Brightness Temperature• 36.5v GHz Brightness Temperature• 36.5h GHz Brightness Temperature• 89.0v GHz Brightness Temperature• 89.0h GHz Brightness Temperature• 150.0v GHz Brightness Temperature• 150.0h GHz Brightness Temperature• 183.3v GHz Brightness Temperature• 183.3h GHz Brightness Temperature

Global Simulation DomainsGlobal Simulation Domains

Inner Grid Locations – 2km resolutionMiddle Grid – 10km

Outer Grid – 50km

10.65 GHz 19.35 GHz 89.0 GHz 150.0 GHz

Surface Temp. 0.120 0.259 0.559 0.645

Freezing Level 0.278 0.369 0.663 0.662

Surface Theta 0.237 0.365 0.608 0.662

500mb Temp. 0.299 0.414 0.648 0.677

850mb Thickness 0.278 0.399 0.636 0.673

Latent Heating 0.333 0.438 0.620 0.658

Sp Hum - 1000mb 0.135 0.274 0.558 0.623

Theta-E Min. - 0.204 - 0.302 - 0.416 - 0.547

Surface Pres. - 0.156 - 0.246 - 0.329 - 0.408

Froude Num. - 0.562 - 0.526 - 0.353 - 0.245

Selected CDRD Tag CorrelationsSelected CDRD Tag Correlations

Correlation coefficients computed between every CDRD variable

Correlation Coefficients for TBsCorrelation Coefficients for TBs

We are Investigating:

• How well do the simulated brightness channels alone sort the simulated precipitation rates when we look at a global cross section of storms?

• Does the addition of dynamics tags help sort simulated precipitation rates over and above what the brightness channels are capable of implying?

Use CDRD to depict how particular precipitation types of precipitation are captured globally

Surface snowfall rate (ordinate) Brightness Temperature (abcissa)

89 GHz183 GHz

23.8 GHz 6.6 GHz

Correlation between brightness channels colored by potential

vorticity advection tag

Correlation between brightness temperatures colored by freezing

level tag

Rain rate vs 89 Ghz brightness colored by 200 mb divergence

(18.7)BT(36.5)BT

200 mb divergence

10log rain rate (mm/hr)

High rain rates

571,967 profiles plotted

Colorado Snowstorm Colorado Snowstorm October 10October 10thth, 2005, 2005

Selected Range: 215 – 230K

CDRD Tags can

INCREASE the probability of

detecting snowfall

Two CDRD Dynamical Tags

Selected Range: 1 - 8Selected Range: 273 – 279K

P(A) = 0.3893

P(B) = 0.0724

P(B|A) = 0.0920

P(A|B) = 0.4947

P(A)

= 0.3893

P(B) = 0.0724 P(C|A,B) = 0.2526

P(B|A) = 0.0920 P(C|B) = 0.1878

P(A|B)

= 0.4947

P(A|B,C) = 0.6655

)(

)|()()|(

BP

ABPAPBAP NO TAGS

Add Surface Temperature Tag

Add Lifted Index Tag

)|()(

),|()|()(),|(

BCPBP

BACPABPAPCBAP

),|()|()(

),,|(),|()|()(

),,|(

CBDPBCPBP

CBADPBACPABPAP

DCBAP P(A) = 0.3893

P(B) = 0.0724

P(C|A,B) = 0.2526

P(D|A,B,C) = 0.7543

P(B|A) = 0.0920

P(C|B) = 0.1878

P(D|B,C) = 0.6028

P(A|B)

= 0.4947

P(A|B,C) = 0.6655

P(A|B,C,D) = 0.8327

A – Snowfall Rate > 1mm/hrA – Snowfall Rate > 1mm/hrB – 150 GHz Brightness Temperatures B – 150 GHz Brightness Temperatures from 215 – 230 Kfrom 215 – 230 KC – Surface Temperature from 273 – 279 K C – Surface Temperature from 273 – 279 K D – Lifted Index from 1 - 8 D – Lifted Index from 1 - 8

Summary of CDRD ResearchSummary of CDRD Research• Global CDRD is operational

– 3D cloud resolving model simulations of precipitation in randomly chosen (precipitating) locations around the globe

– 4 daily runs– Dynamics and thermodynamics tags placed in

database along with profiles of state parameters, microphysics and radiative transfer

– Available online http://cup.aos.wisc.edu/CDRD• CDRD promises to be useful with a Bayesian approach for

precipitation assessment combining traditional retrieval and tags

• We are investigating the additional information content present in the tags and are working to optimize the choices of tags

• CDRD may be useful to predict overall success of microwave radiometer channels over a wide cross section of storm types

Advanced Microphysics Prediction System (AMPS)

• The overarching goal is to take advantage of modern computing power to bring 3D models of microphysics processes to the next level in order to:– Realistically model the evolution and associated

evolved structures of liquid and ice hydrometeors in complex cloud forms

– Facilitate the realistic modeling of radiative transfer in cloudy air to better understand the relationship between satelliute observed brightness temperature and precipitation

Advanced Microphysics Prediction System (AMPS)

• Explicitly evolve characteristics of shape, density, phase and size distribution of aerosol, liquid and ice particles in a 3D Eulerian framework.

• Computationally efficient enough to run for operational use in cloud resolving models.– Goal is to reduce the computation, while

keeping the degree of freedom in resulting physical properties.

Spectral Aerosol Prediction System

• This can be expanded to bin approach, or more size distributions can be added to describe nucleation mode and coarse mode as done by Wilson (2001).

Lognormal distribution

Uniform distribution

• Currently the accumulation mode is modeled with the size distribution.

Pure insoluble AP

Partially soluble AP

Aerosol Microphysics

• Nucleation scavenging and evaporation of hydrometeors are considered.• Neither interaction among aerosols nor between aerosols and hydrometeors are considered.

Spectral LIquid Prediction System (SLIPS)

• About 20 bins seems to be optimal• The vapor deposition is assumed to transfer mass of

activated droplets into multiple bins to compensate the time step.

Liquid microphysics

• CCN activation process– Kohler equation

• Vapor deposition process– Capacitance approach

• Collision-coalescence process– Quasi-stochastic model

• Collision-breakup process– Low and List (1982) formulation

• Aerosol mass prediction in the liquid hydrometeors

• Auto-Conversion….future work

Spectral Habit Ice Prediction System (SHIPS)

No use of categorization!

• Integrated based on local conditions and history of particles• Each bin has different properties of ice particles. • The properties change in time and space.

PPVs

Bin modelBulk micro. par.

Outputs of SHIPS

• Concentration, mass content, and Particle Property Variables (PPVs) for a bin.

• Habit of ice crystals and type of solid hydrometeors in the bin can be diagnosed with PPVs.

• Predicted maximum dimension, circumscribing volume, aspect ratio, bulk density of solid hydrometeors.

• Aerosol distribution outside and inside hydrometeors, and solubility of the aerosols.

Ice microphysics

• Ice nucleation process– deposition-condensation nucleation, contact freezing, immersion

freezing, secondary nucleation.

• Vapor deposition process– Capacitance analogy, empirical mass growth rate, probabilistic

growth

• Collision-coalescence process– Quasi-stochastic approach for aggregation process and riming

process

• Hydrodynamic breakup process• Melting-shedding process• Aerosol mass prediction in the solid hydrometeors

2D orographic snow storm simulation – IMPROVE-2 (13-14 Dec 2001)

From WMO Cloud Modeling Workshop (http://www.rap.ucar.edu/~gthompsn/workshop2004/) IMPROVE 2 website(http://improve.atmos.washington.edu/)

Snow crystals obs

sounding

Key microphysical processes for precipitation on the ground

• Aggregation• Riming

Habit dependent!

Woods et al (2005)

Observed ice particles

Woods et al. (2005)

Simulation with all processes

EXP1 Ice crystal habit of pristine and rimed crystals

• Plates and columns are forming in high level due to immersion freezing.• Bullet rosettes are forming in upper level and grow large due to less concentration.• Columnar crystals dominates in lower levels.• Dendrites were consumed by aggregation and riming before.

12 hours of only liquid mic +30 minutes vd and agg of ice mic +1 hour of all the processes

Simulation with all processes

• Aggregates forming in higher level from bullet rosettes.• Immersion process supplies crystals to the aggregation.• Rimed crystals exist in high level due to small consumption by vapor deposition.

EXP1

Type of solid hydrometers 12 hours of only liquid mic +30 minutes vd and agg of ice mic +1 hour of all the processes

Simulation with all processes

UNI230Ice crystal habit of pristine and rimed crystals

• Plates dominate in upper level due to vapor competition among high concentration of ice crystals.• Dendrites form on plates falling from the above.• Columnar crystals dominate in lower levels.• Irregular crystals were only seen at very beginning of simulation.

12 hours of only liquid mic +30 minutes vd and agg of ice mic +1 hour of all the processes

UNI230

• More aggregates at 3-5 km due to active formation by dendrites.• Ice crystals are available for aggregation once the process starts.• More active riming process in lower level• The altitude of active riming corresponds to observation.

Simulation with all processesType of solid hydrometeors 12 hours of only liquid mic +

30 minutes vd and agg of ice mic +1 hour of all the processes

Precipitation rate and accumulation

Woods et al (2005)

UNI230EXP1 UNIMAX

UNI230-NIM EXP1-DST230

Tim

e (

ho

ur)

Acc

um

ula

tion

(m

m)

EXP1-NIM

Tim

e (

ho

ur)

Ele

vatio

n (

km)

Acc

um

ula

tion

(m

m)

Low level riming Middle level aggregation Secondary nucleation

Case Study : Genoa, 1992 Floods

500 kPa height contour, surface wind vectors equivalent potential temperature (shaded over low elevations) and topography (shaded over higher elevations) for Genoa simulation valid at 1500 UTC, 27 October 1992.

From Tripoli et al (200?)

Experiment DesignWeather Prediction Model• UW-NMS (University of Wisconsin-Non-hydrostatic Modeling System)

Tripoli (1992)

Two categories of aerosols predicted: CCN and IN• nucleation scavenge and evaporation of hydrometeors considered.

Four vertical profiles for IN

CCN vertical initial profile IN vertical initial profile

Habit of predicted solid hydrometeorsCL 4 hours simulation

Habit of predicted solid hydrometeorsDL

• More supply of moisture.• More active depositional nucleation mode.

• More cloud droplets available in convective core for riming for the aggregates

Stronger Updraft

4 hours simulation

Type of predicted solid hydrometeors

CL

2 hours 30 min

Type of predicted solid hydrometeors

DL

2 hours 30 min

CL DL

DA

Mean concentration of ice particles (150<y<300km)

Surface precipitationCL DL

6 hours

CL DL

Maximum vertical motion (150<y<300km)

Horizontally averaged properties of ice particles

CL DL

Horizontally averaged properties of ice particles

CL DL

Summary

• Convective cloud system shows sensitivity to different vertical profiles of Sahara dust layer.– The case with Sahara dust layer indicate stronger updraft in

earlier time than clean case.– More active aggregation process and subsequent riming

process led to more precipitation in the dust case.– The dust case produced surface precipitation than clean case in

the early stage of cloud development, but 10 hour accumulated surface precipitation is similar.

• The simulation supports the results of sensitivity test for Florida storms (CRYSTAL-FACE) by Van Den Heever et al. (2006) qualitatively.

Future Work with Mugnai

• Calculate radiative transfer from explicit microiphysics– Explicit size bins– Structure characteristics of each bin defined

• Progress from simple treatment to more complex treatment

– Spheres– Equivalent spheres– Complex shapes

– Multiple phase of each bin

CRSDAS(Will Lewis)

• Early work with Nexrad In Space (Eric Smith, Eastwoord Im)

• Ensemble Kahlman Filter

• Assumed Geostationary reflectivity and Doppler velocity data

• Applied to simulation of– Supercell– Hurricane Genesis

Conclusions

• CDRD, AMPS and CRMSDAS all show great promise for methodologies to measure – CDRD promises major returns in short term– AMPS is providing new insight into how microphysics

works and promises to catapult the science of microwave radiatiative transfer to a new level

– CRMSDAS is how it should be done in the long run, but it is expensive and we have a lot to learn about error characyteristics and the correct representation of microphysics and radiative transfer before we can expect forward models to be competent.