Robin Hogan

1
Evaluation statistics of cloud fraction and
water content

• Robin Hogan
• Ewan OConnor
• Damian Wilson
• Malcolm Brooks

2
Overview

• Cloudnet level 3 data
• A solution to the problem of evaluating high
  cloud?
• Summary of errors in model cloud fraction and
  water content climatologies over Europe
• ECMWF model
• KNMI Regional atmospheric climate model (RACMO)
• Met Office mesoscale and global
• SMHI Rossby Centre atmospheric model (RCA)
• Meteo France ARPEGE model
• DWD Lokal Modell
• Forecast skill

3
Cloudnet level 3 data

• Level 3 files summarise the comparison of a
  observations and model over a certain period
• Long-term mean of a quantity versus height
• Separation into freq. of occurrence and amount
  when present
• PDFs in height ranges 0-3 km, 3-7 km, 7-12 km and
  12-18 km
• Skill scores versus height for different
  thresholds
• Separate level-3 files/quicklooks are produced
  for
• Each variable cloud fraction, LWC, IWC, high
  cloud fraction
• Each site 4 European, 4 ARM (so far)
• Each model 7 so far, plus persistence/climatology
  forecasts
• Each month and each year
• Different forecast lead times (Met Office meso
  and DWD only)
• In principle different model resolutions /
  parameterisations
• Over 5000 files so far!

4
Cloud fraction
Observations Met Office Mesoscale
Model ECMWF Global Model Meteo-France ARPEGE
Model KNMI RACMO Model Swedish RCA model
5
What can we do about high cloud?

• All models see more cirrus than observed
• We use the known radar sensitivity to remove
  clouds from model that we would not expect to
  detect (affecting heights gt 7 km)
• Does not usually remove enough cloud to bring
  into agreement
• Are all models wrong?
• Or does radar miss more IWC than it thinks due to
  small particles?

6
ARM Nauru 8 Nov 2003

• Radar
• 35 GHz MMCR
• Lidar
• Merged ceilometer and micropulse lidar

Night-time
7
October 2003 Normal processing

• No periods when rain rate gt 8 mm/h
• Large difference between observations and ECMWF
  model, whether model is modified for radar
  sensitivity or not

8
only periods of high lidar sensitivity

• Consider only night-time and periods when lidar
  is unobscured by liquid cloud, rain or melting
  ice
• Liquid clouds removed from comparison
• Cloud fraction OK but peak 2 km too high

9
One month later
Model grossly overestimates high cloud
fraction To evaluate high clouds in models
need a high sensitivity lidar and appropriate
sampling of data (both model and observations)
10
ECMWF cloud fraction

• Cabauw 2002
• Amount when present is good
• Mean cloud fraction and frequency of occurrence
  too high in the boundary layer
• Need to treat snow as cloud in the model

11
ECMWF water content
Chilbolton 2004 LWC

• Mean LWC and IWC accurate to observational
  uncertainties
• Freq. of occurrence too high amount when present
  too low
• Inconsistent with cloud frac.?
• PDF shows occurrence of low values is too high

Chilbolton 2004 IWC
12
RACMO

• Cloud fraction errors similar to ECMWF before
  2003
• Water content errors (mean, frequency of
  occurrency) much as ECMWF
• Lower IWC in high cirrus

13
Met Office mesoscalecloud fraction

• Mean amount when present too low through most of
  atmosphere
• Largely due to inability of model to simulate
  100 cloud fraction, as shown by the PDFs
• Error in high cloud needs to be checked using
  high sensitivity lidar

Cabauw 2004
14
Met Office global cloud fraction
Cabauw 2004

• Observations show greater frequency of cloud with
  increased gridbox size opposite in model
• PDF error unchanged

15
Met Office mesoscale water content
Chilbolton 2004 LWC

• Liquid occurrence very good
• Boundary layer perhaps too low
• Mean LWC underestimated above 3 km
• Similar to previous result found for occurrence
  of supercooled layers

16
Met Office global water content
Chilbolton 2004 LWC
Chilbolton 2004 IWC

• Mean LWC similar but frequency of occurrence much
  lower
• IWC generally higher

17
SMHI Rossby Centre model
Palaiseau 2004

• Amount when present reasonable but frequency of
  occurrence and overall mean much too high
• Similar picture for LWC/IWC mean overestimated
  due to cloud too often

18
Meteo France cloud fraction

• Before Apr 03
• Amount when present far too low
• High values rarely predicted

Cabauw 2002
19
Meteo Fr. water content
Chilbolton 2004 LWC

• Boundary-layer LWC too low
• Frequency of supercooled liquid much too high
• Need to change the T-dependent ice/liquid ratio
• PDF of LWC and IWC too narrow
• Mean IWC too low in mid-levels

Chilbolton 2004 IWC
20
DWD cloud fraction

• Cloud fraction generally very good
• But frequency of occurrence always overestimated
  by 20-30
• PDFs particularly well simulated

Chilbolton 2004
21
DWD water content
Chilbolton 2004

• Frequency of liquid cloud occurrence too high
• LWC in supercooled clouds too high

22
Equitable threat score

• Measure of skill of forecasting cloud
  fractiongt0.05
• Persistence and climatology shown for comparison
• Lower skill in summer convective events

23
Skill versus lead time

• Unsurprisingly UK model most accurate in UK,
  German model most accurate in Germany!

• Typically 500-mb geopotential height used in
  operational forecast verification
• Cloud fraction a more challenging test more
  rapid loss of skill with time