The LES ARM Symbiotic Simulation and Observation (LASSO) Activity

Today’s presenter: William I. Gustafson Jr.

​

Core LASSO Team: William I. Gustafson Jr.¹, Andrew M. Vogelmann², Xiaoping Cheng³,

Mark Delgado², Satoshi Endo², Tami Fairless², Krista Gaustad ¹, Karen Johnson²,

Carina Lansing¹, Zhijin Li³, John Rausch², Eddie Schuman¹, Heng Xiao¹, Damao Zhang¹

​

¹ Pacific Northwest National Laboratory; ² Brookhaven National Laboratory; ³JPL/UCLA; ⁴U. of Maryland Baltimore County

https://www.arm.gov/capabilities/modeling/lasso

1

20 May 2023

What is LASSO?

https://www.arm.gov/capabilities/modeling/lasso

2

20 May 2023

LASSO = Large-Eddy Simulation (LES) Atmospheric Radiation Measurement (ARM) Symbiotic Simulation and Observation activity

Big picture motivation for LASSO

• Intent of LASSO: leverage high-resolution modeling to enhance the value of ARM’s suite of observations for researchers

​

• Motivated by goal of bridging the gap between observations and scales used in forecast and climate models
• Increase understanding of linkages between different observations
• Advancement of cloud and boundary layer theories and parameterizations
• Enable model development and better link observations and models

​

• Vetted large-eddy simulation (LES) provides a plausible proxy for unobservable details in the context of the observations

https://www.arm.gov/capabilities/modeling/lasso

3

20 May 2023

LASSO’s library approach

• Targets specific weather regimes with libraries of LES cases

​

• Curates a suite of relevant observations associated with the chosen regime

​

• Provides a basic comparison between the observations and the LES to aid selection of simulations

​

• Simplifies downloading and use, e.g., bundling of data, online tools targeting key details and download LASSO data

https://www.arm.gov/capabilities/modeling/lasso

4

20 May 2023

Current LASSO scenarios

• Several scenarios proposed at the LASSO Expansion Workshop; details in workshop report: https://www.arm.gov/publications/programdocs/doe-sc-arm-19-023.pdf

​

1. Shallow-convection: Generated 5 years of cases
• Focus on daytime, continental, shallow convection
• Southern Great Plains, Oklahoma

​

• LASSO-CACTI: Beta release was available with full release later in 2023
• Focus on convective initiation and early stages of convective upstage growth for deep convection
• Córdoba, Argentina

​

• LASSO-ENA: In development with staged data release in 2024-25
• Focus on maritime stratiform and related precipitation processes
• Graciosa Island, Azores

https://www.arm.gov/capabilities/modeling/lasso

5

20 May 2023

LASSO being applied to a range of atmospheric topics

• Published examples using LASSO
• Quantifying shallow convection statistics
• Understanding shallow convection processes
• Cloud-radiation connections
• 1D vs. 3D radiation methodologies
• Parameterization development & evaluation
• Radar scan strategies for shallow clouds
• Understanding dust transport in boundary layer
• Included within the Global Modeling Testbed (GMTB) for parameterization evaluation and single-column model use
• Also a great tool for teaching!

​

https://www.arm.gov/capabilities/modeling/lasso

6

20 May 2023

Neggers et al. (2019, JAS)

Time

Large-eddy simulations for shallow convection: LASSO-ShCu

Gustafson et al., 2020, BAMS, https://doi.org/10.1175/BAMS-D-19-0065.1

Nitty gritty details: LASSO-ShCu Technical Description: https://www.arm.gov/publications/tech_reports/doe-sc-arm-tr-216.pdf

https://www.arm.gov/capabilities/modeling/lasso

7

20 May 2023

Available shallow-convection simulations

• 5 years of cases at the Southern Great Plains (SGP) atmospheric observatory
• ShCu season lasts roughly from April to September
• Cases picked based on presence of shallow convection and avoidance of synoptic complications
• Ensembles generated for each date

https://www.arm.gov/capabilities/modeling/lasso

8

20 May 2023

Available ShCu Cases

>760 simulations

LASSO’s shallow-convection LES methodology

• Use WRF model with added LES capabilities (early years also include SAM runs)
• Provide a statistical representation of the clouds; a traditional LES approach with doubly periodic boundaries and homogeneous surface
• dx = 100 m, domain width = 25 km (early runs use 14 km)
• Initialized with observed sounding at ~6 LST
• Large-scale forcing ensemble to inform forcing uncertainty and cloud sensitivity
• Output saved every 10 minutes

https://www.arm.gov/capabilities/modeling/lasso

9

20 May 2023

Domain configuration

• Grid spacing: dx=100 m; dz=30 m up to 5 km, then stretches to 300 m at domain top
• Captures shallow convection processes sufficiently to provide a good representation of the clouds
• Resolution chosen to capture the clouds—grid spacing too large to capture stable conditions and boundary layer transitions, particularly during evening
• Implications
• Simulations good to use during midday
• Avoid first several hours due to model spin-up and evenings when turbulence does not decay accurately
• Timing of cloud decay can be off due to not capturing turbulence decay timing

https://www.arm.gov/capabilities/modeling/lasso

10

20 May 2023

10-Jun-2016, Sim. ID 20

Spin-up period

PBL collapse

ARSCL

WRF

Handling of the lower boundary conditions

• Flat, uniform lower boundary for the LES
• No terrain or vegetation variability
• Time-dependent, specified sensible and latent heat fluxes imposed from Variational Analysis VAP
• Averages surface flux measurements from around SGP
• Eddy correlation system (ECOR) uses a sonic anemometer with an open-path infrared gas analyzer to measure vertical wind and water vapor
• Energy balance Bowen ratio station (EBBR) calculates bulk aerodynamic fluxes from soil and atmospheric measurements
• VARANAL: https://www.arm.gov/capabilities/science-data-products/vaps/varanal

​

• Implications for the LES results
• No feedback between land and atmosphere
• Radiation has no impact on surface fluxes or energy balance
• Surface gradients across the region are not represented

https://www.arm.gov/capabilities/modeling/lasso

11

20 May 2023

Lateral boundary conditions

• Doubly periodic lateral boundary conditions
• Assuming a scale separation between larger, mesoscale conditions and the LES domain
• Connection to larger scales achieved through a uniform, domain-wide, large-scale forcing tendency
• Implications
• Largest scales of simulated variability and motion are on the order of the domain size, 25 km (14 km for early years)
• Each model column is statistically identical
• Cannot think in terms of one-to-one comparisons to point locations
• LES represents the average conditions around the SGP

https://www.arm.gov/capabilities/modeling/lasso

12

20 May 2023

LASSO employs an ensemble of forcings to capture the range of possible conditions

• Large-scale forcing (LSF) datasets generated from 3 sources

​

• Variational Analysis: ARM product, 300 km spatial scale

​

• ECMWF IFS model: ~16, 115, & 413 km spatial scales

​

• Multiscale Data Assimilation (MSDA): 75, 150, & 300 km scales;
• GSI 3DVar data assimilation package at dx=2.5 km
• Directly incorporates ARM observations into the analysis
• RWP wind profiles
• Surface meteorology

​

• Total of 8 LSF ensemble members per case (includes 1 no-LSF)

https://www.arm.gov/capabilities/modeling/lasso

13

20 May 2023

Gustafson et al. (2020, BAMS)

Spread in Cloud Fraction

Observations: ‘data scales’

https://www.arm.gov/capabilities/modeling/lasso

14

20 May 2023

Observations: variables

https://www.arm.gov/capabilities/modeling/lasso

15

20 May 2023

*COGS available for 2018 & 2019 only

Observations: cloud fraction sources and comments

Cloud fraction

• Sources
• Total-Sky Imager (TSI)
• ARSCL (z < 5 km)
• COGS (Romps and Öktem, BAMS, 2018)
• Comments
• TSI can be contaminated by upper-level cloud
• ARSCL too sensitive and is pencil-beam obs
• Good correlation between TSI and ARSCL time series

indicates no upper-level cloud influencing the TSI

• COGS (when available) reliable for cloud frac < ~0.5-0.6

2-D time-height cloud mask

• Sources
• ARSCL
• COGS (when available)
• Comments
• ARSCL is a pencil-beam obs and can be contaminated by insects (correction method in progress by Christopher Williams et al.)

16

20 May 2023

LASSO Metadata Table for identifying simulations

• LASSO data bundles identified by combination of
1. Case date
2. Simulation ID

​

• Identify the simulation ID when publishing!

​

• Metadata associated with each simulation ID can change between case dates

​

• Metadata Table helps identify metadata and search for relevant simulations

17

https://adc.arm.gov/lassometadata

The LASSO Bundle Browser

https://adc.arm.gov/lassobrowser

​

Developed and/or maintained by

• Kyle Dumas
• Michael Giansiracusa
• Bhargavi Krishna

Enables:

• Interactive querying of LASSO sims & skill
• Contains diagnostic plots
• Ordering of data bundles

​

​

https://www.arm.gov/capabilities/modeling/lasso

18

20 May 2023

Deep convection during CACTI:�LASSO-CACTI

William Gustafson

Pacific Northwest National Laboratory

https://www.arm.gov/capabilities/modeling/lasso

19

20 May 2023

CACTI & RELAMPAGO field campaigns

• CACTI = Cloud, Aerosol, and Complex Terrain Interactions
• DOE ARM funded
• October 2018 through April 2019

​

​

​

​

​

​

• RELAMPAGO = Remote sensing of Electrification, Lightning, And Mesoscale/Microscale Processes with Adaptive Ground Observations
• NSF funded
• June 2018 through April 2019

https://www.arm.gov/capabilities/modeling/lasso

20

20 May 2023

Flights

RELAMPAGO Assets

Why Córdoba, Argentina?

• Córdoba region has a very high frequency of initiating mesoscale convective systems (MCSs)

21

20 May 2023

TRMM Statistics for MCS Frequency

Freq. of large MCS initiation based on T_b

Freq. of IR T_b<235 K for initiated MCSs

6 hours later…

12 hours later…

9%

50%

50%

50%

Varble et al. (CACTI Science Plan, 2018)

CACTI Approach

• Varble, A. C., et al., 2021: Utilizing a Storm-Generating Hotspot to Study Convective Cloud Transitions: The CACTI Experiment. BAMS, 102, E1597-E1620, doi:10.1175/BAMS-D-20-0030.1.

​

• Nesbitt S. W., et al., 2021: A Storm Safari in Subtropical South America: Proyecto RELAMPAGO. BAMS, 102, E1621-E1644, doi:10.1175/BAMS-D-20-0029.1.

​

• AMS special collection: https://journals.ametsoc.org/collection/RELAMPAGO-CACTI

​

https://www.arm.gov/capabilities/modeling/lasso

22

20 May 2023

The CACTI �Observing Facilities

• Core instrumentation from the ARM Mobile Facility #1 (AMF1)
• Surface suite
• Total-sky imager
• Micropulse lidar
• Doppler lidar
• Ceilometer
• Radiometer suite
• Aerosol suite

​

• G-1 aircraft

​

• Radars: e.g., C-band scanning, Ka/X-band scanning, Ka-band zenith pointing, radar wind profiler

​

• See list in Varble et al. (BAMS, 2021, doi:10.1175/BAMS-D-20-0030.1)

https://www.arm.gov/capabilities/modeling/lasso

23

20 May 2023

Science drivers guide LASSO-CACTI scenario design

• Convective cloud dynamics
• e.g., thermal-like structures, updraft strength, and entrainment; the relationship to critical features like updraft and downdraft mass fluxes, vertical transport, and the shallow-to-deep convective transition
• Convection-environment interactions, e.g., cold pools
• Convective drafts in turbulent flow
• Microphysics-dynamics interactions
• Especially in the context of cloud-scale eddies and smaller-scale turbulence

​

• Science drivers chosen to balance relevant science with computational capacity
• LES resolution governed by cloud core requirements
• Domain size determines portion of lifespan simulated
• Limiting ensembles to mesoscale simulations with the potential for a small number of LES ensemble members for specific cases

https://www.arm.gov/capabilities/modeling/lasso

24

20 May 2023

Case dates target a selection of convective behavior

• Chosen days have convection form and grow within view of ARM’s scanning radar

​

• Identified 20 case dates meeting our criteria
• Convection ranges from short-lived convection to large MCSs

25

Simulated

Brightness Temperature

at 19 UTC

Δx = 2.5 km (D02)

AMF

Mesoscale ensembles for case selection and LES boundary condition choices

• Mesoscale ensembles run for each case date (example for 10-Nov-2018 at right)
• 33 ensemble members based on ERA5, ERA5 Ensemble, FNL, and GFS Ensemble
• Nested down to 2.5 km grid spacing
• Best performing ensemble members identified based on cloud comparison to GOES-16 IR data
• Down-selected ensemble members get final vetting using bulk CSPAR2 statistics, e.g., 20 dBZ echo-top height

26

AMF

Modeling stages to achieve Δx=100 m

• Stage 1: Mesoscale ensembles with Δx=7.5 & 2.5 km
• For selecting boundary conditions and case selection

​

• Stage 2: LES setup with Δx=500 & 100 m
• For selected cases, some with several LES per case

​

• Stage 3: Post-process data to simplify usage

27

20 May 2023

Domain Sizes

Meso

LES

LES domains

• “Ndown” from D02 to D03
• Nesting permits starting domains at different times to save resources and smooth spin-up process
• D03 starts at 6 UTC
• D04 can start at 12 UTC
• Primary LES run based on best-performing mesoscale ensemble member(s)
• Some additional LES runs for testing other BCs or physics
• ~25 h wall time per model hour on 7168 cores of Cumulus-2

https://www.arm.gov/capabilities/modeling/lasso

28

20 May 2023

Comparison of 500 hPa Vertical Velocity at Each Grid Spacing

Summary of what is available for LASSO-CACTI

• Simulations
• 33-member ensembles of kilometer-scale simulations to evaluate forcings and sensitivity of convection for 20 days, ∆x = 7.5 and 2.5 km
• 9 days with LES simulations for a handful of forcings/physics options each day, ∆x = 500 and 100 m

​

• Model data
• Raw model output for restarts and traditional output
• Variable subsets by category of variable, e.g., clouds, radiation, PBL
• Input files for reproducing each run

​

• Skill scores and quick-look plots

https://www.arm.gov/capabilities/modeling/lasso

29

20 May 2023

Multiscale Observational Datasets

• Regional: Satellite-based
• Sources
• VISST: IR brightness temperatures (11.2 μm channel)
• Application
• Time-dependent areal coverage of the convective cores
• Local: Scanning C-Band Radar-based
• Sources
• CSAPR-2/Taranis
• Applications
• Locate AMF-storm position within the LES grid
• Time series of surface rain rates, and of radar echo-top heights
• Point Measurements
• Sondes (ARM & RELAMPAGO)

30

Skill scores to evaluate simulations

• Satellite brightness temperatures (Tb) for convective area development

​

• Radar echo-top heights for local convective intensity

​

• Radar-retrieved precipitation averaged over region

31

Where to find more information

• LASSO-ShCu
• Primary citation: Gustafson et al., 2020, BAMS, https://doi.org/10.1175/BAMS-D-19-0065.1
• Nitty gritty details: LASSO Technical Description: https://www.arm.gov/publications/tech_reports/doe-sc-arm-tr-216.pdf
• LASSO-CACTI
• Technical description: https://lasso-cacti-doc.arm.gov/latest/index.html
• Web page: https://www.arm.gov/capabilities/modeling/lasso
• Technical support
• lasso@arm.gov (goes to Andy and Bill)
• Works for any sort of question, best for reporting something broken
• Not publicly searchable
• LASSO Discussion Forum on new ARM Discourse website: https://discourse.adc.arm.gov/
• Encouraging folks to use this for general questions and discussion
• Content will build over time and become a community asset
• LASSO email list: http://us11.campaign-archive1.com/home/?u=74cd5b8a5435b8eca383fc18c&id=38f02e1568

32

Additional Slides�LASSO-ShCu

LASSO’s version of WRF model

• Available at https://code.arm.gov/lasso/lasso-wrf
• In the “wrf_faster” branch of the repository
• Traces heritage back to the Fast-physics System Testbed and Research (FASTER) project at BNL circa 2010
• Based on WRF v3.8.1
• Added features over default WRF in a new ‘em_crm’ build category
• Extra LES-related output
• Time averaging between output times
• Volumes, profiles, and column-integrated quantities; spatial averaging
• Specialized LES diagnostics, e.g., vertical fluxes of heat and moisture, in-cloud vs. all-domain averages
• Streamlined handling of LES initialization and large-scale forcing
• Ability to use large-scale forcing and/or nudging toward profiles
• Customized Morrison microphysics with 4-mode aerosol inputs (used during LASSO testing phase)

https://www.arm.gov/capabilities/modeling/lasso

34

20 May 2023

Physics configuration settings for “production years”

https://www.arm.gov/capabilities/modeling/lasso

35

20 May 2023

• Ice microphysics activates even though we focus on shallow convection because of tall model top
• Attempt to capture cirrus clouds with varying success
• Some ensemble members transition to deep or mid-level cloud

Model data available in LASSO data bundles

• LASSO “data bundles” organize data into a series of tar files
• sgplassodiagconfobsmod#C1.m1.YYYYMMDD.tar
• sgplassodiagraw#C1.m1.YYYYMMDD.tar
• Other obs-related bundle files
• sgplassohighfreqobsC1.c1.YYYYMMDD.000000.tar
• sgplassocogsC1.c1.YYYYMMDD.000000.tar

https://www.arm.gov/capabilities/modeling/lasso

36

20 May 2023

Model inputs

Post-processed

model output

Raw model

output

~70 MB

~70 GB

config directory of “confobsmod” tar

• Contains input files necessary for reproducing the simulation
• Initial sounding: wrfinput_d01.nc
• Large-scale forcing: input_ls_forcing.nc
• Surface fluxes: input_sfc_forcing.nc
• WRF’s configuration file: namelist.input

​

• Script to convert WRF inputs to SAM inputs: sam_input_generation.py

​

https://www.arm.gov/capabilities/modeling/lasso

37

20 May 2023

Formatted for em_crm option

raw_model directory of “raw” tar

• Contains all the raw model output from WRF
• Every 10 minutes
• Separate wrfout files for each model hour, i.e., 6 times per file
• One wrfstat file per run with all times in it

​

• wrfout_d01… = Standard WRF output format in netCDF
• See WRF ARW Users Guide for most details, http://www2.mmm.ucar.edu/wrf/users/docs/user_guide_V3/user_guide_V3.8/ARWUsersGuideV3.8.pdf
• Some variables split into base and perturbation values, e.g., geopotential height = PHB + PB
• Staggered variables: U, V, W, PH+PHB

https://www.arm.gov/capabilities/modeling/lasso

38

20 May 2023

Model inputs

• Initial conditions
• Temperature, moisture, and wind profiles from 12 UTC sounding (5 or 6 LTC depending on day of year)
• Random perturbations to potential temperature in lowest 33 model layers (~990 m), 0.1 K max. amplitude
• Surface forcing
• Regionally averaged sensible and latent heat fluxes from EBBR and ECOR stations via Variational Analysis VAP, updated hourly
• Large-scale forcing
• Provided as a domain-wide tendency for temperature and moisture profiles
• Represents large-scale horizontal advection and heating/drying due to subsidence
• Nudging to observations not used for LASSO
• Implications
• Synoptic changes impact whole domain at the same time—they do not propagate across the region
• Large-scale winds not updated after initial time

https://www.arm.gov/capabilities/modeling/lasso

39

20 May 2023

wrfstat_d01 files = LES diagnostic output

• Variables averaged/accumulated between 10-min. output times
• Meteorological state, turbulent fluxes, cloud details, radiation, etc.
• Variable suffixes indicate type of averaging (CS refers to “CRM statistic” from the em_crm build mode)
• CSV = volume variable, i.e., cell-by-cell for whole volume, only time averaged and no spatial averaging
• CSP = domain-wide horizontally averaged profile
• CSS = slab variable, i.e., a horizontal surface either from column integrated or a surface variable
• CST = time series, i.e., single point per time from horizontal averaging of column-integrated or surface variables

https://www.arm.gov/capabilities/modeling/lasso

40

20 May 2023

Types of wrfstat Variables

CSV is to CSP as CSS is to CST

LASSO Domain

obs_model directory of “confobsmod” tar

• Contains post-processed and summarized model data
• Generated via the LASSO-O workflow that will be presented in Part 3

​

• Consists of a subset of values used to calculate the skill scores

​

• Three file variations
• sgplassomod = “ingested” wrfout data on original time 10-min. interval; diagnostic variables
• sgplassodiagobsmod = adds time reduction to hourly time sampling & side-by-side with observations
• sgplassodiagobsmodz = height-dependent cloud fractions, model+observations, 10-min. interval

​

• Much easier to download confobsmod file if you do not need detailed raw output

https://www.arm.gov/capabilities/modeling/lasso

41

20 May 2023

LASSO Bundle Browser: Upper half (A, B, D)

A. Select date(s), Model config, Measurement 🡪

B. Static plots for selected date summarizing model behavior

• Heat maps, Skill score plots for all simulations and variables, GOES vis satellite loop

D. Interactive plots showing results for selected Measurement and choices in (C)

• Clockwise: Taylor diagram, Plot of Taylor score vs. Relative Mean, Regression, Time series

​

​

42

https://adc.arm.gov/lassobrowser

LASSO Bundle Browser: Lower half (C, E)

C. Slide rulers to choose the range of net skill scores that are displayed in (D) and (E)

• Measurement Skill, 1D Cloud Skill, 2D Cloud Skill, Total Cloud Skill

E. Tabulated results are given for the selections from (A) and (C)

• Save table values
• Summary of the simulation run configuration
• Diagnostic plots per simulation: Time series, Taylor diagram, Regression, 2D Cloud Mask, Soundings
• Ordering data

43

20 May 2023

https://adc.arm.gov/lassobrowser

Where to find more information

• High-level description and primary citation: BAMS article
• Gustafson et al., 2020, BAMS, https://doi.org/10.1175/BAMS-D-19-0065.1
• Nitty gritty details: LASSO Technical Description
• https://www.arm.gov/publications/tech_reports/doe-sc-arm-tr-216.pdf
• Web page: https://www.arm.gov/capabilities/modeling/lasso
• Technical support
• lasso@arm.gov (goes to Andy and Bill)
• Works for any sort of question
• Best for reporting something broken
• Not publicly searchable
• LASSO Discussion Forum on new ARM Discourse website: https://discourse.adc.arm.gov/
• Encouraging folks to use this for general questions and discussion
• Content will build over time and become a community asset
• LASSO email list: http://us11.campaign-archive1.com/home/?u=74cd5b8a5435b8eca383fc18c&id=38f02e1568

44

Additional Slides�LASSO-CACTI

Soil initial conditions for LASSO-CACTI

• Soil initialized with WRF-Hydro to establish a spun-up soil state consistent with WRF physics
• Continuous WRF-Hydro run from August 2018 to April 2019 using Δx=2.5 km
• Driven by ERA5

https://www.arm.gov/capabilities/modeling/lasso

46

20 May 2023

Soil temperatures are similar between ERA5 and WRF-Hydro

Soil comparison for 26-Oct-2018 0 UTC:

ERA5

​

​

​

​

​

WRF-Hydro

Topmost Soil Layer Temperature [K]

Grid spacing choice for deep convection

• ∆x ~ 250 m is a threshold for cloud behavior
• > 250 m is more plume-like
• < 250 m is more bubble-like
• Too coarse and updrafts are not mixed sufficiently, making them more like hoses to the free troposphere

47

20 May 2023

Winds for Different Grid Spacing

2 km

1 km

500 m

250 m

100 m

66.7 m

33.3 m

Lebo & Morrison (2015)

Color = w

Contour = u

Model output strategy

• Mesoscale ensembles have 33 members (20 dates), LES have several per case (9 dates)
• Total dataset for the scenario ~2 PB

https://www.arm.gov/capabilities/modeling/lasso

48

20 May 2023

WRF Model Data

We want to make it as easy as possible for users,� but…

• Using these runs will be non-trivial due to the data size!
• Raw output sizes
• Mesoscale ensemble for D02
• ~325 GB per ensemble member
• >100 TB for full set of cases and members
• LES runs for D04
• Raw output >35 TB per run
• >1 PB raw model output for 10 cases & 2 LES/case
• Subsets add to above sizes

​

https://www.arm.gov/capabilities/modeling/lasso

49

20 May 2023

Rough File Sizes for Each Domain

Subsets generated in post-processing

• Goal of reducing file sizes for users not needing whole raw files
• Extra diagnostics provided, e.g., LWP, CAPE, destaggered winds, heights, pressure
• Variable subsets grouped by theme in separate files*:

https://www.arm.gov/capabilities/modeling/lasso

50

20 May 2023

• Static data, constant in time like terrain height, 0.1 GB
• Meteorological state, 28 GB (with staggered variables interpolated to cell centers)
• Meteorological state for staggered variables, 8 GB
• Cloud data, 2 GB
• Surface data, 0.4 GB
• Boundary layer data, 5 GB
• Radiation data, 0.2 GB
• Aerosol data, 4 GB
• Tendency data, 10 GB (e.g., microphysics tendencies & process rates)
• Tracer data, 8 GB

• Subsets available on different height coordinates

• Height above ground level
• Height above sea level
• Pressure levels
• Raw model levels

* File sizes given are per output time for a typical D04 subset file on raw model levels. Note that a wrfout_d04 is 171 GB.