SYNOPTIC GLOBAL REMOTE SENSING OF LAND SURFACE VEGETATION: CHALLENGES ... · environmental related...
1
Armando Barreto 1* , Kamel Didan 1 SYNOPTIC GLOBAL REMOTE SENSING OF LAND SURFACE VEGETATION: CHALLENGES AND OPPORTUNITIES VIP Lab., ECE Dept & Institute for the Environment & Society. The University of Arizona, Tucson, AZ 85721, USA *[email protected] Introduction Continuous collection of global satellite imagery over the years has contributed to the creation of a long data record from AVHRR, MODIS and other sensors. These records account now for 30+ years, and as the archive grows it becomes an invaluable source of data for many environmental related studies dealing with trends and changes at local and global scale. For atmospheric research this is perfect, however for studies dealing with phenomena at the surface of the earth, the presence of clouds, aerosols, spatial gaps, viewing issues, and less than consistent atmosphere correction and reprocessing makes it difficult to obtain good quality data everywhere and every time. These issues change by location and time, making it difficult to know at a glance the availability and quality of the data. Objectives Our goal was to assess: The global data records quality The probability of getting good and enough data to perform a study at any given location and/or time The accuracy and certainty of these data We generated global maps showing the problems and limitations of synoptic global remote sensing, the level of noise and errors in these records, and where science results could be suspect. In doing so we also identified opportunities for further work on these data records. Data and Methodology Terra & Aqua 16–day CMG Vegetation Index record from 2000 to 2009 Additionally we used the Terra MODIS CMG land cover product Quality information was extracted from the CMG product and the new C5 pixel reliability measure The analysis was stratified per latitude, season, land cover and geographic location 2000 NDVI profiles for different biomes and different locations. These profiles illustrate the noise levels in the VI signals throughout the year. The noise will have implications on phenology estimation and subsequently vegetation change research. Almost all biomes are subject to noise and uncertainty, however the rainforest biome shows the largest level of noise. Conclusions Establishing the quality of the data should be the first step for accurate and successful data analysis. The global spatial and temporal distribution of data, quality, cloud and aerosol indicates that two major biomes, rainforest and boreal forests, are the most challenging owing to excessive cloud, snow/Ice and aerosol cover and a host of other issues (ex: poor atmosphere correction). This indicates that research using synoptic remote sensing over these areas needs to account for this uncertainty. These issues are further complicated by the lack of continuity across the various global imagers. Average yearly good data yield 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 NDVI (PM Sensor) NDVI (AM Sensor) Evergreen needleleaf forest Evergeen broadleaf forest Deciduous needleleaf forest Deciduous broadleaf forest Open shublands Woody savannas Savannas Barren or sparsely vegetated Jan-Feb-Mar Apr-May-Jun Jul-Aug-Sep Oct-Nov-Dec AM versus PM overpass 0.00% 1.00% 2.00% 3.00% 4.00% 5.00% 6.00% 7.00% 1/25/2000 5/4/2000 8/12/2000 11/20/2000 2/28/2001 6/8/2001 9/16/2001 12/25/2001 4/4/2002 7/13/2002 10/21/2002 1/29/2003 5/9/2003 8/17/2003 11/25/2003 3/4/2004 6/12/2004 9/20/2004 12/29/2004 4/8/2005 7/17/2005 10/25/2005 2/2/2006 5/13/2006 8/21/2006 11/29/2006 3/9/2007 6/17/2007 9/25/2007 1/3/2008 4/12/2008 7/21/2008 10/29/2008 2/6/2009 5/17/2009 % Global Average Cloud Percentage (16 days composite) Typical global NDVI map (August 2008) showing vegetation at its peak. Only highest quality data retained. The two major biomes, Rain forest and Boreal forest , represent the bulk of global vegetation yet they are the areas most prone to problems . A morning versus afternoon overpass impacts the data, complicating continuity and accurate trend analysis. Deciduous needle leaf forest shows the largest difference, while open shrublands have the strongest correlation High Aerosol Cloud distribution Seasonal good data yield Results Yearly cloud distribution by latitude 90 30 60 0 -30 -60 0% 5% 10% 15% 1 3 5 7 9 11 13 15 Latitude (degrees) 0% 5% 10% 15% 1 3 5 7 9 11 13 15 0% 2% 4% 6% 8% 1 3 5 7 9 11 13 15 0% 5% 10% 15% 20% 1 3 5 7 9 11 13 15 0 25 100 % 50 75 0% 5% 10% 15% 20% 1 3 5 7 9 1 1 1 0% 10% 20% 30% 1 3 5 7 9 1 1 1 0% 5% 10% 15% 1 3 5 7 9 1… 1… 1… 0% 5% 10% 15% 20% 3 5 7 9 1 1 1 0% 5% 10% 1 3 5 7 9 1 1 1 Jan-Feb-Mar Apr-May-Jun Jul-Aug-Sep Oct-Nov-Dec 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 30.4 60.8 91.2 121.6 152 182.4 212.8243.2 273.6 304 334.4 364.8 NDVI Month Savannas: South Africa J F M A M J J A S O N D 0.5 0.6 0.7 0.8 0.9 0 30 61 91 122 152 182 213 243 274 304 334 365 NDVI Month Evergreen broadleaf forest: Brazil J F M A M J J A S O N D 0 0.1 0.2 0.3 0.4 0 30 61 91 122 152 182 213 243 274 304 334 365 NDVI Month Shrublands: Australia J F M A M J J A S O N D 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 30 61 91 122 152 182 213 243 274 304 334 365 NDVI Month Deciduous broadleaf forest: Eastern USA J F M A M J J A S O N D 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0 30 61 91 122 152 182 213 243 274 304 334 365 NDVI Month Woody Savannas: Asia J F M A M J J A S O N D 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0 30 61 91 122 152 182 213 243 274 304 334 365 NDVI Month Evergreen needleleaf forest: Russia J F M A M J J A S O N D Acknowledgements: This work was supported by NASA CA# NNX08AT05A, PI Kamel Didan
Transcript of SYNOPTIC GLOBAL REMOTE SENSING OF LAND SURFACE VEGETATION: CHALLENGES ... · environmental related...
Armando Barreto1*, Kamel Didan1
SYNOPTIC GLOBAL REMOTE SENSING OF LAND SURFACE VEGETATION: CHALLENGES AND OPPORTUNITIES
1 VIP Lab., ECE Dept & Institute for the Environment & Society. The University of Arizona, Tucson, AZ 85721, USA
Introduction
Continuous collection of global satellite imagery over the
years has contributed to the creation of a long data record
from AVHRR, MODIS and other sensors. These records
account now for 30+ years, and as the archive grows it
becomes an invaluable source of data for many
environmental related studies dealing with trends and
changes at local and global scale. For atmospheric research
this is perfect, however for studies dealing with phenomena
at the surface of the earth, the presence of clouds, aerosols,
spatial gaps, viewing issues, and less than consistent
atmosphere correction and reprocessing makes it difficult to
obtain good quality data everywhere and every time. These
issues change by location and time, making it difficult to
know at a glance the availability and quality of the data.
Objectives
Our goal was to assess:
The global data records quality
The probability of getting good and enough data to
perform a study at any given location and/or time
The accuracy and certainty of these data
We generated global maps showing the problems and
limitations of synoptic global remote sensing, the level of
noise and errors in these records, and where science
results could be suspect. In doing so we also identified
opportunities for further work on these data records.
Data and Methodology
Terra & Aqua 16–day CMG Vegetation Index record from
2000 to 2009 Additionally we used the Terra MODIS CMG
land cover product
Quality information was extracted from the CMG product
and the new C5 pixel reliability measure
The analysis was stratified per latitude, season, land cover
and geographic location
2000
NDVI profiles for different biomes and different locations. These
profiles illustrate the noise levels in the VI signals throughout the
year. The noise will have implications on phenology estimation and
subsequently vegetation change research. Almost all biomes are
subject to noise and uncertainty, however the rainforest biome
shows the largest level of noise.
Conclusions
Establishing the quality of the data should be the first step for accurate and
successful data analysis. The global spatial and temporal distribution of
data, quality, cloud and aerosol indicates that two major biomes, rainforest
and boreal forests, are the most challenging owing to excessive cloud,
snow/Ice and aerosol cover and a host of other issues (ex: poor
atmosphere correction).
This indicates that research using synoptic remote sensing over these
areas needs to account for this uncertainty. These issues are further
complicated by the lack of continuity across the various global imagers.
Average yearly good data yield
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
ND
VI
(PM
Se
nso
r)
NDVI (AM Sensor)
Evergreen needleleaf forest Evergeen broadleaf forest Deciduous needleleaf forest Deciduous broadleaf forest
Open shublands Woody savannas Savannas Barren or sparsely vegetated
Jan-Feb-Mar Apr-May-Jun Jul-Aug-Sep Oct-Nov-Dec
AM versus PM overpass
0.00%
1.00%
2.00%
3.00%
4.00%
5.00%
6.00%
7.00%
1/2
5/2
00
0
5/4
/20
00
8/1
2/2
00
0
11
/20
/20
00
2/2
8/2
00
1
6/8
/20
01
9/1
6/2
00
1
12
/25
/20
01
4/4
/20
02
7/1
3/2
00
2
10
/21
/20
02
1/2
9/2
00
3
5/9
/20
03
8/1
7/2
00
3
11
/25
/20
03
3/4
/20
04
6/1
2/2
00
4
9/2
0/2
00
4
12
/29
/20
04
4/8
/20
05
7/1
7/2
00
5
10
/25
/20
05
2/2
/20
06
5/1
3/2
00
6
8/2
1/2
00
6
11
/29
/20
06
3/9
/20
07
6/1
7/2
00
7
9/2
5/2
00
7
1/3
/20
08
4/1
2/2
00
8
7/2
1/2
00
8
10
/29
/20
08
2/6
/20
09
5/1
7/2
00
9
%
Global Average Cloud Percentage (16 days composite)
Typical global NDVI map (August 2008) showing vegetation
at its peak. Only highest quality data retained.
The two major biomes, Rain forest and Boreal forest,
represent the bulk of global vegetation yet they are the
areas most prone to problems.
A morning versus afternoon overpass impacts the data, complicating
continuity and accurate trend analysis. Deciduous needle leaf forest shows
the largest difference, while open shrublands have the strongest correlation
High Aerosol
Cloud distribution
Seasonal good
data yield
Results
Yearly cloud distribution by latitude
90
30
60
0
-30
-60
0% 5% 10% 15%
1
3
5
7
9
11
13
15
Lati
tud
e (
deg
rees)
0% 5% 10% 15%
1
3
5
7
9
11
13
15
0% 2% 4% 6% 8%
1
3
5
7
9
11
13
15
0% 5% 10% 15% 20%
1
3
5
7
9
11
13
15
0 25 100 %50 75
0% 5% 10% 15% 20%
1
3
5
7
9
1
1
1
0% 10% 20% 30%
1
3
5
7
9
1
1
1
0% 5% 10% 15%
1
3
5
7
9
1…
1…
1…
0% 5% 10% 15% 20%
1
3
5
7
9
1
1
1
0% 5% 10%
1
3
5
7
9
1
1
1
Jan-Feb-Mar Apr-May-Jun Jul-Aug-Sep Oct-Nov-Dec
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0 30.4 60.8 91.2 121.6 152 182.4 212.8243.2 273.6 304 334.4 364.8
ND
VI
Month
Savannas: South Africa
J F M A M J J A S O N D
0.5
0.6
0.7
0.8
0.9
0 30 61 91 122 152 182 213 243 274 304 334 365
ND
VI
Month
Evergreen broadleaf forest: Brazil
J F M A M J J A S O N D 0
0.1
0.2
0.3
0.4
0 30 61 91 122 152 182 213 243 274 304 334 365
ND
VI
Month
Shrublands: Australia
J F M A M J J A S O N D
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 30 61 91 122 152 182 213 243 274 304 334 365
ND
VI
Month
Deciduous broadleaf forest: Eastern USA
J F M A M J J A S O N D 0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0 30 61 91 122 152 182 213 243 274 304 334 365
ND
VI
Month
Woody Savannas: Asia
J F M A M J J A S O N D
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0 30 61 91 122 152 182 213 243 274 304 334 365
ND
VI
Month
Evergreen needleleaf forest: Russia
J F M A M J J A S O N D
Acknowledgements: This work was supported by NASA CA# NNX08AT05A, PI Kamel Didan
