Comparison of land surface temperatures from land surface models and microwave and infrared satellit

1
Comparison of land surface temperatures from
land surface models and microwave and infrared
satellite retrievals

• J.-L. Moncet, P. Liang, A. Lipton,
• J. Galantowicz
• AER, Inc.
• C. Prigent
• Observatoire de Paris, LERMA

2
Overview

• Surface temperature is an important diagnostic
  variable used to assess land surface models
• Model-produced land surface temperatures differ
  significantly regionally and in response to
  changes in forcing
• Satellite-derived surface temperature is useful
  for providing a global sanity check and
  identifying anomalous behaviors
• Recent comparisons between satellite LST sources
  (MODIS, ISCCP, and AIRS) show that much work
  still needs to be done to make the satellite
  products agree

3
MODIS/ISCCP Mean LST comparison (July 2003)

• MODIS LST Aqua L3 daily V4 5km grid, equatorial
  crossing time 130am and 130pm
• ISCCP LST composite of geostationary and polar
  products, DX level 30km equal-area grid at 3-hour
  intervals
• Spatial regrid MODIS 5km LST is averaged to
  ISCCP 30km grids
• Temporal interpolation ISCCP 3-hourly LST is
  linearly interpolated to MODIS view time
• Clear condition defined as 98 of MODIS grids
  with one 30km ISCCP grid passed MODIS cloud mask
• Monthly Mean Average of daily clear LST
  measurements during a month

4
MODIS/ISCCP Mean LST comparison (July 2003)
ISCCP MODIS LST difference monthly mean
Gray areas Grids of fewer than three clear
nighttime measurements over a month according to
MODIS LST clear ratio
ISCCP LST is higher than MODIS LST during both
day and night over arid and semi-arid areas, and
daytime difference is larger then nighttime
difference
5
MODIS/ISCCP diurnal cycle amplitude comparison
(July 2003)
MODIS
Larger discrepancies between MODIS and ISCCP in
arid regions
ISCCP
6
MODIS/ISCCP temporal LST variability night
MODIS
ISCCP
LST temporal variability Monthly standard
deviation for clear conditions

• MODIS and ISCCP LST temporal standard deviations
  are comparable for nighttime data, with values
  being slightly higher for ISCCP
• Both LSTs are generally stable at nighttime. 97
  of the MODIS grid points having a standard
  deviation less than 4 K and 93 for ISCCP

7
MODIS/ISCCP temporal LST variability day

Mean(SD) 2.43k
MODIS
MODIS
DashedCDF
K

Mean(SD) 3.98k
ISCCP
ISCCP
DashedCDF
K

• MODIS and ISCCP LST temporal LST standard
  deviations have larger difference for daytime
  data, MODIS LST remains stable for most of the
  coverage and ISCCP standard deviation increase
  greatly
• 92 of the MODIS grid points having a standard
  deviation less than 4 K and 65 for ISCCP

8
Possible factors causing differences in temporal
variability

• Random errors in LST retrieval, due to ancillary
  data, spatial and temporal interpolation
• Cloud contamination increases LST standard
  deviation
• If cloud screening is overly conservative, it
  might exclude uncontaminated unusual LSTs and
  decrease variability

MODIS
ISCCP

• Many areas where ISCCP LST monthly standard
  deviation is much bigger than MODIS are arid,
  where cloud errors are less likely to be a factor
• Spatial interpolation errors would have biggest
  impact where spatial inhomogeneity of LST is
  biggest daytime areas with high solar heating
  (arid areas)
• Similar for time interpolation errors, arid areas
  have high rate of time change (temporal
  inhomogeneity)

9
Can interpolation of ISCCP to MODIS time explain
differences?
ISCCP MODIS LST difference vs. time difference
between MODIS overpass time and nearest
bracketing ISCCP data
The daytime mean ISCCP?MODIS LST difference is
maximum at ?hour0, and decreases by 2K at
?hour?1.5 No clear trend in the LST difference
standard deviations versus time difference
The nighttime mean ISCCP?MODIS LST difference at
?hour?1.5 is slightly higher than at ?hour0 No
clear trend in the LST difference standard
deviations versus time difference
Conclusion The slight increasing or decreasing
trend of mean LST difference versus time
difference are consistent with linear
interpolation over a typical diurnal cycle, and
the interpolation error is secondary to the
discrepancies between the MODIS and ISCCP LST
products from other causes.
10
Using independent microwave data to validate
infrared LST products

• Heritage previous work of C. Prigent with SSM/I
• AMSR-E provides excellent timeliness and
  co-location with MODIS
• AMSR-E has 6 and 10 GHz, the low-frequency
  channels on AMSR-E are very valuable for
  monitoring temporal changes in surface emissivity
  
• Emissivity retrieval system ? Clear retrieval
  mode
• 
  with assumption
• then
• Emissivity variability semis increases if
  correlation between TB and Tskin is poor

11
Isolating temporally stable surfaces

• The 11-GHz channels have little sensitivity to
  the atmosphere and the ratio is quite insensitive
  to fluctuations in surface temperature
• Any change in the state of the surface (due to
  e.g. soil moisture, vegetation greening or
  harvesting) significantly affecting the retrieved
  AMSR-E emissivities at a given location over
  homogeneous and RFI free areas be captured by
  analysis of the temporal evolution of

Filled ? clear sample Half open ?partly
cloudy sample Open ? overcast sample
Night
? Daytime ? Nighttime
Single grid R11 time series plot ? an example of
R11 change caused by surface property change, ?
circled by black squares are outliers.
Analysis of relation between emissivity temporal
variability and LST temporal variability should
exclude unstable surfaces, marked by bright
colors where monthly stand deviation of R11 gt
0.01.
12
Temporal variability in AMSR-E emissivity MODIS
vs. ISCCP LST
AMSR 19V monthly SD difference retrieved with
ISCCP LST retrieved with MODIS LST for clear
condition July 2003
AMSR 19V monthly SD retrieved with ISCCP LST vs.
retrieved with MODIS LST for clear condition July
2003
Night
Day
Higher temporal consistency between independent
AMSR-E Tbs and MODIS LSTs than with ISCCP LSTs
13
AMSR-E day/night emissivity differences and land
surface type
14
AMSR-E day/night emissivity differences and LST
source
AMSR 19V emissivity difference day-night stable
surface MODIS LST
Emissivity diurnal difference is unexpected large
for the circled areas with ISCCP LST
MODIS
Areas with larger ISCCP emissivity diurnal
difference coincide with the areas of larger
daytime LST difference between ISCCP and MODIS.
ISCCP
15
AMSR-E day/night emissivity differences Regional
Penetration at 89 GHz is expected to be high only
over sandy deserts, emissivity diurnal difference
is expected to be minimum with uncontaminated LST
(by cloud, dust) over stable surfaces
The narrow peak of emissivity difference
histogram with MODIS LST indicates that the
relationship (in a monthly average sense) between
day and night AMSR-E TBs and MODIS is
remarkably similar across geographically distinct
regions.
The main peak of emissivity difference histogram
with ISCCP LST is much broader and contains a
significant fraction of positive day/night
emissivity difference.
16
Consistency assessment between microwave
measurements and LST data source

• Independent AMSR-E observations validate temporal
  MODIS LST variations and mean diurnal cycle
• MODIS better than ISCCP for validating land
  surface models in the clear-sky (tight quality
  control required for cloud contamination gt high
  quality cloud mask and timeliness)
• AMSR-E LSTs can be used in cloudy conditions
  over vegetated surfaces also useful for IR/QC
  in clear conditions
• 1D variational retrieval algorithm uses
  emissivity database from clear AMSR/MODIS
  retrievals as a constraint and operates on clear
  and cloudy measurements
• Emissivities useful for diagnosing regional/long
  temporal biases and response to sudden events
  (correlation with R11)

17
MODIS/AGRMET Mean LST comparison (July 2003)
AGRMET MODIS LST difference monthly mean for
clear condition
Night
Day

• AGRMET LST Air Force Weather Agencys version of
  NOAH LSM model, uses a combination of
  precipitation estimates, including gauge, SSM/I,
  and geostationary satellite Infrared channel
  precipitation estimates, LST data are at 0.5
  degree resolution and 3-hour intervals
• Spatial temporal interpolation Bilinearly
  interpolated to ISCCP grid and MODIS view time
• Clear condition MODIS clear ratio 98

18
MODIS/AIRS Mean LST comparison (July 2003)
AIRS MODIS LST difference monthly mean for
clear condition
DashedCDF
Bias -0.5K
Night
Night
DashedCDF
Bias -0.6K
Day
Day

• AIRS LST AIRS L3 V5 daily gridded standard 1x1
  degree product, equatorial crossing time 130am
  and 130pm
• Spatial temporal interpolation Bilinearly
  interpolated to ISCCP grid and no temporal
  interpolation
• Clear condition MODIS clear ratio 98

19
MODIS/NCEP Mean LST comparison (July 2003)
NCEP MODIS LST difference monthly mean for
clear condition
DashedCDF
Bias 0.6K
Night
Night
DashedCDF
Bias -7.3K
Day
Day

• NCEP LST NCEP global final analysis dataset at 1
  degree resolution and 6- hour intervals
• Spatial temporal interpolation Bilinearly
  interpolated to ISCCP grid and MODIS view time
• Clear condition MODIS clear ratio 98

20
Microwave-only 1D-var retrieval examples
July 18 2007 daytime
July 28 2007 daytime
1D-var retrieval for July 2003 in a region of
300km x 300km from West Virginia, the test is
chosen over vegetated surface where no surface
penetration issue. The surface emissivity
constraint is our monthly mean clear retrieval
database, and the above two examples show 1D-var
retrieval results over cloudy days.
21
Retrieval example Time Series LST

Retrieved LST time series (black triangles)
compared to KCKB (W. Va.) surface air observations

• Clear days with available MODIS clear LST
  measurement (blue squares) High agreement with
  MODIS measurements, the mean difference is
  0.15k.
• Cloudy days without MODIS clear LST measurement
  The retrieved LSTs are close to the station
  surface 2m shelter temperatures (green stars) and
  catch the variation trend over the whole month.

22
Future work

• Assess cloud contamination impact on LST data and
  retrieved emissivities
• Validate emissivity and effective temperature
  estimation algorithm over arid areas
• Examine the impact of emissivity bias/
  uncertainties on LST by our 1D-var retrieval
  algorithm