Notes on Global Smoothed NDVI Datasets Usage
(Updated: 01/25/2006)
The sets of sub-global and whole global weekly smoothed NDVI
data is released for
evaluation purpose, or for preliminary use. The multi-year weekly
datasets come originally from different generations of AVHRR sensors
onboard NOAA polar-orbiting satellites (NOAA-7, -9, -11, -14, -16, -17,
-18) over the past two
decades; then fixed by the Adjusting Cumulative Distribution Function
(ACDF) approach (using the best known quality data from NOAA-11 and
NOAA-14), thus has the overall improved quality and consistency
comparing to the original post launch calibrated, and mathematically
smoothed NDVI datasets produced at NOAA/NESDIS. The NDVI value should
be viewed as Top-of-Atmosphere (due to lack of atmospheric radiative
transfer corrections), and NOAA-11 or -14 "equivalent" values.
Given the nature of polar-orbiting satellites, for high latitude (e.g.
above 60N), NDVI derived from visible and near infrared channels are
often unreliable, which is true especially in winter. Thus special
treatment for winter weeks was applied to derive winter datasets (See
e. below)). Other details see
below.
a. Spatial & Temporal Range, Resolution & Projection
1). Sub-Global Dataset
- Adopted the operational global vegetation index (GVI) data format,
covering 75N to 55S, and 180W to 180E, weekly, 0.144-degree resolution,
latitude-longitude projection (or the so called Plate Carree
Projection).
- Pixel cell (latitude, longitude) is assumed to represent the cell
center location. The exact spatial range covers from (75.024N to
54.008S and 179.856W to 180.000E), starting from the North-West corner
(75.024, -179.856), then following col -> row sequence.
2). Whole Global Dataset
- Has same project and spatial resolution as the Sub-Global Dataset,
but extends to cover the polar regions by putting the above Sub-Global
array into the whole global land/sea mask. In other words, the
core information for Sub-Global and Whole Global datasets for NDVI are
exactly the same, except that the Whole Global datasets have land/sea
masks for regions not covered by the Sub-Global datasets.
- The exact spatial range covers from (90.000N to 89.856S and 179.856W
to 180.000E), starting from the North-West corner
(90.000, -179.856), then following col -> row sequence.
b. Data Format
1). Sub-Global Array
- 2500-col by 904-row byte array with each byte representing a pixel
value.
- To obtain the value of a certain grid centered at (lon, lat), one
needs to calculate the sub-global array index (col, row) by:
col=integer part of [(lon+179.856)/0.144+0.5]
row=integer part of [(75.024-lat)/0.144+0.5]
2). Whole Global Array
- 2500-col by 1250-row byte array with each byte representing a pixel
value.
- To obtain the value of a certain grid centered at (lon, lat), one
needs to calculate the sub-global array index (col, row) by:
col=integer part of [(lon+179.856)/0.144+0.5]
row=integer part of [(90.000-lat)/0.144+0.5]
c. Value Extraction
- Since the data is stored in byte (8-bit is one byte) array flat file,
each pixel has the value within the range of 0 to 255, called Count
Value; (Hints: some programming languages define Byte type value ranges
from -126 to +127. Care should be taken when interpreting data after
directly reading from file. It is trivial to do further conversion
depending on the programming language you use).
- Count Value 255 is reserved for water and ocean pixels; Count Value
254 is reserved for un-used land pixels (i.e. land pixel that has no
NDVI value, such as those located North of 75.024N, or South of
55.008S, etc.);
- To obtain NDVI value from Count Value, simply apply: NDVI=(240.0 -
Count) / 350.0 - 0.05
d. Filename Convention and Time Range
- The experimental datasets are not generated from operational
production stream. Therefore, to keep it simple, there are 52 weeks for
each year (though this is a deviation from reality). In reality, there
are years that have 53 weeks, week 53 is dropped out (user could use
interpolation if data for that week is needed anyway). There are
also years with 52 full weeks but week 53 starts within this year and
the whole week should really be counted as the first week of the
following year (We decided to count a week as a week in a certain year
if great than or equal to 4 out of the 7 days fall into that year).
- The datasets for a whole year are zipped together as a .zip file
(usable for both Windows and Linux/Unix systems). After
extraction (or unzip), data files ending with ".GVI2" are for
Sub-Global Array, data files ending with ".WGVI" are for Whole Global
Array.
- The timestamp is part of each filename (after extraction or unzip),
which includes year, julian day, and week number and has the format
"_yyyyddd_YYww", where "yyyy" is the four-digit year number, "ddd" is
the Julian day number ("ddd" is always on Monday given the operational
data processing scheme), "YY" is the two-digit year number that the
week should be belong to (Note: in most cases, YY is the same as the
last two digits of "yyyy", however, there are occassions that the last
week of this year is counted as the first week of the next year.
For example, the timestamp portion of dataset
"SMN_CDF_fixed_2003363_0401.GVI2", stands for the Monday (Julian day
363) of year 2003, and the dataset is for week 1 of 2004.
- The weekly data is valid on Monday (or the Julian day "ddd"), and
good for the 7 days (including Monday) after Monday. Since the
operational production of our GVPS (Global Vegetation Processing
System) keep this convention, it is the most reasonable
interpretation. For years with more than 52 weeks, we suggest users use
interpolation (using week 52 of this year & week 1 of the following
year) to obtain the Smoothed NDVI for week 53.
e. Winter Weeks
- week 1 to 10, and week 43 to 52, NDVI values above 60N are assigned
zero.
f. Climatology Datasets
- The 24-yr (1982 to 2005) global weekly climatology of smoothed NDVI
was derived which includes mean, standard deviation, maximum, minimum
of NDVI for each pixel. The datasets are stored as Count Value
arrays (in Sub-Global and Whole Global formats separately) as described
above. Count Values should be converted using the same formula as
described in c. above to
obtain the physical values (Special note for the standard deviation
datasets: after count values being converted to physical values, all
negative physical values should be thrown away as they either represent
oceans or land masks).
Q
& A
Questions, Thoughts, Comments:
contact Le.Jiang@noaa.gov, 301-763-8348x330, IMSG at NOAA/NESDIS/ORA
(01/25/2006)