RECONSTRUCTION OF REMOTE
SENSING TIME SERIES DATA
Javier Rivera
VIP Laboratory
REMOTE SENSING
Remote sensing can be defined as the acquisition of information
about the state and condition of an object with sensors without
taking a physical sample.
Remote sensing technology is used, among others things, for:
Agricultural management
Land cover and land use status and change
Weather (data collection)
Study of environmental changes
Fires monitoring
Phenology of plants, etc…
REMOTE SENSING SYSTEM
EXAMPLE PLATFORMS
Multispectral coarse resolution Global Imager
Study of the whole Earth System (multi-purpose data records)
AVHRR (Advanced Very High Resolution Radiometer )
1981-Present
Local Area Coverage (LAC @ 1.1 km) vs. Global Area Coverage
(GAC @ 4km)
MODIS (Moderate-resolution Imaging Spectroradiometer)
(Terra : 2000-Present & Aqua 2002 – Present)
250m, 500m and 1km
SPOT VGT
1998 to present
1km
Multispectral Medium resolution
Landsat TM/ETM+ series
The full record is now free (use to be $500/scene)
High resolution and hyperspectral
HYPERION, ALI, EO1, IKONOS, QuickBird, etc…
INTRODUCTION (VEGETATION INDICES)
Spectral Vegetation Indices are
radiometric measures of the
amount, structure, and condition of
vegetation, and a precise monitoring
tool
They provide key data to:
famine early warning systems
Hydrologic and Biogeochemical
studies
Phenology and Seasonality
Land cover and land cover change
(interannual variation)
Natural resource
management, monitoring and
sustainable development
Climate studies
Improves
the signal
0.0
0.1
0.2
0.3
0.4
0.5
0.6
350 550 750 950 1150
% R
efl
ecta
nce
Wavelength (nm)
VEGETATION INDICES
1*5.7*6*5.2
bluerednir
rednirEVI
rednir
rednirNDVI
1*4.2*5.22
rednir
rednirEVI
ρred=reflectance in the red region
ρnir=reflectance in the NIR region
ρblue=reflectance in the blue region
The objective of VI is to extract the “green” signal only (what is the green signal
is highly debatable, but it is a proxy for many vegetation conditions).
PHENOLOGY
Vegetation phenology can be defined as the plants study of the biological cycle events throughout the year and the seasonal and interannual response by climate variations. Phenology products, produced daily or on 16-day compositing period, provided different parameters which describe the seasonal behavior of the vegetation.
WHAT WE ACTUALLY DESIRE IN DATA
RECORD
What are the best characteristics of a data record
Resolution
Frequency
Accuracy
Fidelity
Long term
CURRENT METHODS
Threshold-based – Sacrifice temporal resolution
Fourier-based – Sub-estimate VI
Gaussian functions fitting – Time consuming and
cannot be applied to all types of data
PROBLEM STATEMENT
Time series data records exhibit multiple spatial and temporal discontinuities that inhibit the proper and accurate study of the natural and gradual process of growth of land surface vegetation.
These discontinuities, or gaps, are a result of: cloud cover
atmosphere contaminants
Shadow
less than ideal viewing
geometry
Sensor acquisition geometry
OBJECTIVES
Develop a new Data GAP filling method that
improves the temporal resolution and preserves
the data quality
To develop a gap filling technique that improves the
temporal characterization of the land surface and
preserves the integrity of data in support of accurate
change studies.
To evaluate the performance and reliability of this
gap filling method. To test the performance of this
data gap filling technique/Method
Impact on Phenology
Impact on Time series profiles
DATA - MODIS• Daily Global Data
• AQUA : 2002 – Current
• TERRA: 2000 - Current
• Resolution: CMG, 0.05 degrees (5600 meters)
• Projection: Unprojected, Latitude/Longitude
• Image dimensions: 3600 Rows, 7200 Cols
METHODOLOGY
Obtain long term time series data (from MODIS and AVHRR)
Data Filtering
Determine the average of the long term time series data and its confidence interval
Determine the Maximum Compositing Period
Fill Remaining GAPs using Inverse Distance Weighted method
DATA FILTERING - DECISION TREE
Rank=5IsCloud
y
Yes
No
START
Cloud
Shadow
Vz<=30
Rank=1 Rank=3
Yes
No
Rank=2
Data
valid
Yes
NoRank =7
Rank=4IsSnow
Yes
Yes
Low
Aerosol
No
No
Yes
No
Low
Aerosol &
Vz<30
Rank=2
No
Yes
This process depends heavily on the input data, and aims at limiting and eliminating problematic data from the ESDR records.
Cloudy Data, data with high Aerosols loads (MODIS only), and out of range data are all removed. Large spatial gaps result from this process.
Rank Description
1 Ideal Data, Good and Useful
2 Good to Marginal
3 Marginal to Questionable
4 Snow
5 Cloud
6 Estimated
7 NO DATA
COMPOSITING PERIODS
Compositing periods (days): 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20, 25, 30, 40
and 50
DETERMINE AN OPTIMAL COMPOSING NUMBER OF
GLOBAL DAILY IMAGES
0
10
20
30
40
50
60
70
80
90
100
0
10
20
30
40
50
60
70
80
90
100
0 5 10 15 20 25 30 35 40 45 50 55
Fre
qu
en
cy
Qu
ali
ty (
%)
Glo
ba
l V
I g
oo
d d
ata
(%
)
Compositing Period (days)
S80
S260
S175
FQ
MEAN OF MISSING POINTS (DAYS WITHOUT VI DATA)
BETWEEN TWO CONSECUTIVE KNOWN POINTS.
Jan, Feb, Mar Apr, May, Jun
Jul, Aug, Sep Oct, Nov, Dec
LONG TERM DATA
Long term average records are necessary for gap filling, data
characterization and for constraining various data processes, as well
as for long term phenology extraction.
The long term average can be used for constraining the gap filling
algorithm.
When using the long term averages to constrain the gap filling
algorithm preference is for the shorter periods (to minimize long term
biases), then longer periods are used depending on the algorithm
performance ( ± .5, 0.75 and 1.0 ).
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 100 200 300 400
ND
VI
Day of Year
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 100 200 300 400
ND
VI
Day of Year
GAP FILLED
Where:
VIi is the vegetation index value of the known points
dij is the distance to the known point
VIj is the vegetation index value of the unknown
point
n is a user defined power parameter (often 1, 2 or 3)
in
ij
in
ij
i
j
d
d
VI
VI1
Spatial Gaps resulting from data filtering are filled, QA
labeled, and stored in the final data records. The Gap filling
algorithm uses 2 methods:
1. Linear Interpolation for the long term average data records
2. Inverse Distance Weighting for the daily and multi-day data
records
3. To eliminate outliers and minimize under/over performance of
the Gap filling algorithm we’re constraining the estimates by
the 5-year moving window of the long term average. One
standard deviation is used to restrict the boundaries of the
values. Values outside these boundaries are replace with the
longer term average value and properly labeled in the output
QA (Rank parameter).
GAP FILLED - COMPARISON
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 100 200 300 400
Ve
ge
tati
on
In
de
x
(ND
VI)
Day of Year
VIFilled
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 100 200 300 400
Ve
ge
taio
n I
nd
ex
(ND
VI)
Day of Year
Piecewice's
Method
INPUT/OUTPUT
CONCLUSIONS
It is simpler and less computer intensive.
The method looks only at the data around the temporalgap which helps eliminate bias that may result frommethods that simultaneously need the full annual cycle,and
It balances between providing higher frequency andhigher quality data without the noise associated withdaily data while avoiding the excessive smoothing ofother methods.
Impacts phenology research, where change is measuredwith days over decades (so higher frequency makes adifference)
Better characterization of the VI Time Series improvesthe quantitative analysis. For example change in Carboncycling is measured in fraction of a percent and multi-daydata will not provide such accuracy
ACKNOWLEDGMENT
QUESTIONS?
Dr. Kamel Didan
Dr. Muluneh Yitayew
Armando Barreto