Creation of NDVI images
from level 1B AHVRR data retrieved from
George Mason University antenna

Aubrey Weese

School of Computational Sciences,
George Mason University, Fairfax VA

March 11, 2003

Abstract

In June 2002 George Mason University acquired a Quorum satellite receiving station that is capable of picking up High Resolution Picture Transmission (HRPT) data from NOAA satellites. A total of four AVHRR level 1B datasets (three from NOAA 16 and one from NOAA 12) were retrieved by this antenna and processed into a final scaled 8-bit NDVI product. These NDVI maps were then compared to NDVI maps produced from 1B datasets downloaded from the Wallops Island, VA receiving station via the NOAA Satellite Active Archive. The two were found to be virtually identical. The only significant differences were between NOAA 16 and NOAA 12. These differences were caused by the application of the SMAC atmospheric correction algorithm to the NOAA 12 data. Finally, one NOAA 16 NDVI map was produced per month for the year 2002 to inspect seasonal change in Cape Cod, MA. It was found, as expected, that the maximum NDVI values in the area decreased in the winter and increased in the summer.

Introduction and Background

Vegetation indexes are a tool that can be used to map the presence of vegetation on a pixel basis as well as measuring the amount or condition of vegetation within each pixel. They are dimensionless numbers created by exploiting the unique spectral properties of plants, particularly in the red and near infrared portions of the spectrum. Plants reflect a very low amount of red energy because this energy is absorbed by chlorophyll in photosynthetic leaves, which have maximum absorption at 470 nm (blue) and 670 nm (red). At the same time leaves reflect a very high amount of near infrared energy because of the scattering process in healthy, turgid leaves.

Therefore, the contrast between red reflectance to near infrared reflectance can provide a good indication for the presence of plants. In the picture below, you can see how the reflectance curve for a plant (in this case soybeans) with a dip in the red region and a peak in the infrared, differs from the reflectance curve of bare soil.

Figure 1: Spectral reflectance signature of a photosynthetically active leaf with a soil signature to show contrast (Tucker and Seller, 1986).

This contrast can be expressed as a ratio. The simple ratio (SR) of X_nir/X_red was the first vegetation index to be used (where X stands for digital counts) (Jordan, 1969). However, this ratio was found to be inadequate for densely vegetated areas where X_red approaches zero, and the ratio increases without bounds. So it was normalized to the Normalized Difference Vegetation Index (NDVI) = X_nir-X_red / X_nir + X_red (Deering, 1978). This ratio can range from -1 to +1 but normal values fall between 0 and 0.8. A zero means no vegetation, and close to +1 (0.8 - 0.9) indicates the highest possible density of green leaves. This is illustrated in the figure below.

Figure 2. Ilustration by Robert Simmon, from Nasa Earth Observatory Website.

Healthy vegetation (left) absorbs most of the visible light that hits it, and reflects a large portion of the near-infrared light. Unhealthy or sparse vegetation (right) reflects more visible light and less near-infrared light. The numbers on the figure above are representative of actual values, but real vegetation is much more varied.

For low and medium amounts of vegetation, changes in the NDVI are caused by changes both in red and near-IR reflectance. At higher amounts of vegetation, the red band becomes saturated (approaches zero) because of chlorophyll absorption, and the NDVI changes only due to variations in the near-IR reflectance. In the real world, desert regions have been found to have the lowest, or "baseline" NDVI, followed by semi-arid and grassland regions. The highest NDVI values are found in closed forest canopies and open forests with green understories. Intermediate values usually indicate mixed regions. Over 70% of the earth's surface falls within this category, being open canopies with mixed background and vegetation signals (Graetz, 1990).

A major advantage of the NDVI is that because of its ratioing properties, it is able to naturally cancel out a large proportion of signal variations due to calibration, noise, and changing irradiance conditions caused by varying sun angles, topography, clouds, shadows and atmospheric conditions.

The NDVI contributes valuable data to the study of global climate change, because it is used as a first step in the calculation of land surface parameters in a general circulation model of the atmosphere. In order to be complete, this circulation model needs to incorporate information about the biosphere, because there are water, energy and carbon exchanges between vegetation and the lower boundary of the atmosphere. Originally, data for biosphere models was obtained from look up tables that assigned biophysical parameters to global land cover classifications.

Satellite data provides a much better source for this information. The temporal change NDVI can be used to depict seasonal activity, length of the growing season, peak greenness, onset of greenness, and leaf turnover or "dry down" period (Myneni et al, 1997). Monthly NDVI values can be used to calculate monthly fAPAR (fraction of absorbed photosynthetically active radiation), LAI (leaf area index), land surface albedo and roughness length (Sellers et al. 1992, Sellers et al. 1995). They can also be used to determine such things as carbon-fixation, canopy resistance, and potential evapotranspiration (Asrar et al., 1984; Baret and Guyot, 1991; Goward and Huemmrich, 1992; Sellers, 1985; Running and Nemani, 1988; Tucker et al., 1981; Curran, 1980). These values provide useful input for many atmospheric models. NDVI is also used in a variety of land applications, including natural resource management, agriculture, the Global Health and Human Monitoring Program, and Famine Early Warning Systems (Prince and Justice, 1991; Hutchinson, 1991).

The Advanced Very High Resolution Radiometer is the instrument that was first used, and still is most commonly used, to calculate NDVI. The sensor has five channels that are capable of providing day and night time information about ice, snow, vegetation, clouds and the sea surface.

Table 1. The 5 channels of the NOAA polar series AVHRR instrument

This data is obtained on a daily basis and used primarily for weather analysis and forecasting. However, another important use is the calculation of NDVI. Channel 1 in the sensor corresponds to red light (0.58 - 0.68 um) and channel 2 corresponds to near infrared (0.73 - 1.10 um). So the NDVI equation can be adapted to AVHRR as (channel 1 - channel 2) / (channel 1 + channel 2). Although vegetation monitoring was not the primary purpose of the instrument, it has been widely used in this capacity because of its high temporal resolution and global coverage.

The AVHRR was developed by ITT-A/CD and is flown on NOAA KLM spacecraft. It has a small field of view (1.3 milliradians by 1.3 milliradians) that is scanned across the earth from one horizon to another by the continuous 360 degree rotation of a flat scanning mirror. The scan direction is oriented perpendicular to the orbit track and the mirror rotates and a speed selected so that adjacent scan lines are contiguous at the sub satellite (nadir) position. In this way, the sensor produces strip maps of the globe from pole to pole as it orbits on the NOAA satellites at an elevation of approximately 833 km. These satellites circle the earth approximately 14 times per day, having orbital periods of about 102 minutes. The orbits are timed to allow complete global coverage twice per day, per satellite. This normally gives a daytime and nighttime view of the earth, in swaths of about 2,600 km in width. High resolution (1 kilometer per pixel) data are transmitted from the satellite continuously, and can be collected when the satellite is within range of a receiving station. This direct transmission is called HRPT, for High Resolution Picture Transmission.

As of June, 2002, George Mason University acquired such a station to pick up high resolution satellite data. This station is manufactured by Quorum Communications. The HRPT data is broadcast from the satellites over the L-Band frequency (1.7 GHz) and picked up by a tactical antenna located on the roof of the GMU Fenwick library. The antenna is positioned by a Totally Accurate Clock (TAC-2) device. The time must be exactly accurate so the instrument will know where to expect the satellite, since it moves a huge distance in just one second. Once the antenna is locked on to a satellite, it transmits data to a MetCom DSP (Digital Signal Processing) drive bay receiver. The receiver converts the signal to a 70 MHz IF and digitizes it at 50 million samples per second into a file with 10-bit resolution. The system is operated by a Dos based program called Qtrack that drives the antenna and records the received signal to disk drives. Each of the 5 NOAA satellites flies by about once every two hours and the antenna can pick up data ranging from Cuba to Northern Canada. The data files it picks up average around 100MB per swath.

The image below is from the indoor portion of a Quorum receiving station almost identical to the one at GMU.

Figure 3. A Quorum receiving station in Pioneer Junior College, Singapore

The data files picked up by the antenna are classified as Level 1B. The definition of level 1B, according to the NOAA Polar Orbiter User's Guide, is "raw data that has been quality controlled, assembled into discrete data sets, and to which earth location and calibration information has been appended (but not applied)."

This level 1B data can be imported into a software program called WinChips, developed by the Institute of Geography at University of Copenhagen, which will apply the appended location and calibration information and process the data into useful values such as NDVI.

Methods

Six datasets were collected from the antenna in summer 2002 on the dates June 20, July 31, and August 1, 2, 6 and 13. The July 31 set was collected from NOAA 12 and the rest were from NOAA 16. These datasets all contained 5 AHVRR channels, but only channels 1 and 2 were imported into WinChips because that is all that is needed to compute NDVI. The NOAA AVHRR import module was used to reduce these sets from 10-bit HRPT files to 8-bit WinChips image files. During import, image line synchronization is checked. Noisy lines are identified, and, if lines are missing, blank lines are inserted into data stream.

Also, calibration coefficients are determined from the information embedded in the HRPT data stream. For the non-thermal channels 1 and 2, this is done by defining the channel offset using deep space data. A calibration lookup table is then created as a text file. This table expresses the conversion of digital counts into physical units. It incorporates non-linear corrections, and the methods implemented follow the guidelines for NOAA AVHRR calibration described in the NOAA POD User's Guide and the NOAA KLM User's Guide. A Chips orbit file is also created by extracting embedded orbital elements from the data stream. Table 2 gives an example of the information contained in the orbit file and text file generated by WinChips for the June 20th dataset.

Table 2. Log file (left) and orbit file (right) used by WinChips in calibrating AVHRR data

Four of the six sets imported were useful and went on through further data processing. The June 20th set was thrown out because it showed only a narrow strip of land with no identifiable landmarks to provide a clue to its location.

Figure 4. Discarded June 20, 2002 antenna image

The August 2nd set was thrown out because it showed only ocean.

Figure 5. Discarded August 2, 2002 antenna image

The following images portray the visible and near-infrared channels of the four datasets that were retained. Generally, in visible light, vegetated areas are very dark, almost black, while desert regions are light. At near-infrared wavelengths, the vegetation is brighter and deserts are about the same.

	Visible Red Light (Channel 1)	Near Infrared (Channel 2)
July 31	Figure 5. July 31, 2002 visible light antenna image	Figure 6. July 31, 2002 near infrared antenna image
August 1	Figure 7. August 1, 2002 visible light antenna image	Figure 8. August 1, 2002 near infrared antenna image
August 6	Figure 9. August 6, 2002 visible light antenna image	Figure 10. August 6, 2002 near infrared antenna image
August 13	Figure 11. August 13, 2002 visible light antenna image	Figure 12. August 13, 2002 near infrared antenna image

After importation, the WinChips NOAA calibration module was used to convert the raw AVHRR pixel values into reflectance. An offset of 0 and a gain of 0.5 was used because these were the example values given in the WinChips help file for use with 8-bit images. These values produce an accuracy of +/- 0.5%.

The sun zenith angle correction was also applied to all four of the images. This correction normalizes the reflectance according to the zenith angle. The SMAC (Simplified Method for Atmospheric Correction) algorithm was also applied to the July 31 images. The July 31 data came from NOAA 12, which supports SMAC correction, but the other three images came from NOAA 16, which does not.

The SMAC algorithm (Rahman, H. and Dedieu, G., 1994) calls for input of aerosol optical depth at 550 nm, atmospheric water vapor and ozone content. The values assumed for these were as follows: aerosol optical depth: 0.05, atmospheric water vapor: 2.3 g/cm², ozone content: 0.319 cm-atm. These values are the same as those used by the Northern BIOsphere and Modeling Experiment in producing AVHRR land cover maps of Canada (Cihlar, J., J. Beaubien, and B. Park, 2001).

The calibration did not have visual effects on the images except to darken them slightly.

The final step was to perform the NDVI equation on the calibrated images. When the straight equation NDVI = (channel 2 - channel 1) / (channel 2 + channel 1) was used, very dark images with indistinguishable details resulted. 8-bit images can have pixel values ranging from 0 to 255. Using the straight NDVI equation, large portions of the continental landmass were covered by pixels that had a values under 20, interspersed with rare bright speckles near 255. The image below in an example.

Figure 13. Straight NDVI equation applied to August 6, 2002 image

The bright pixels appear to have come from clouds. These results could possibly be improved by implementing a cloud screen before applying the NDVI equation. However, the pixels are so rare they do not even show up on the histogram.

Histogram 1. Straight NDVI equation applied to August 6, 2002 image

It was found that applying a scale to the function to increase the NDVI values helped to produce a brighter image with a larger range of values. This scale is described by the Towson University Center for Geographic Information Sciences' "Chesapeake Bay From Space" website. The scale converts a number between -1.0 and 1.0 into a pixel value that is more appropriate on an 8-bit gray tone display. The equation is: Scaled NDVI = 100(NDVI + 1)

Using this technique, the NDVI computed value is scaled to the range of 0 to 200, where computed -1.0 equals 0, computed 0 equals 100, and computed 1.0 equals 200. For example, a pixel with an NDVI value of 0.43 would be scaled into a gray scale value of 143. As a result, NDVI values less than 100 now represent clouds, snow, water, and other non-vegetative surfaces, and values equal to or greater than 100 represent vegetative surfaces.

Applying this scaled equation to the images produced the following images and histograms:

Figure 14. Scaled NDVI applied to July 31, 2002 antenna image	Figure 15. Scaled NDVI applied to August 1, 2002 antenna image
Figure 16. Scaled NDVI applied to August 6, 2002 antenna image	Figure 17. Scaled NDVI applied to August 13, 2002 antenna image

Histogram 2. Scaled NDVI applied to July 31, 2002 antenna image

Histogram 3. Scaled NDVI applied to August 1, 2002 antenna image

Histogram 4. Scaled NDVI applied to August 6, 2002 antenna image

Histogram 5. Scaled NDVI applied to August 13, 2002 antenna image

Following the development of a satisfactory scale, 10 bit HRPT images were then downloaded from the Wallops Island, VA receiving station via the online NOAA Satellite Active Archive. These HRPT streams were processed in WinChips to produce NDVI in exactly the same manner as the antenna data.

First, a NOAA 16 image from August 1, 2002 was downloaded. This image should ideally be almost identical to the NOAA 16 GMU antenna image collected on the same date. Comparisons were made to ensure that that was the case. For NOAA 12, a suitable image could not be found on the same date as the GMU July 31 image, so August 8 was used instead. Finally, comparisons were made between the NOAA 12 and NOAA 16 images to determine the difference made by the SMAC atmospheric correction algorithm.

Secondly, one NOAA 16 image was downloaded per month for the year 2002. Ideally, the NDVI values should decrease in winter as vegetation becomes sparser and increase in summer as it becomes denser. Checking if this is the case is a good test if the algorithm is working properly.

The WinChips software produced a "NOAA Navigation" error when attempting to georeference the AVHRR data, so that step was not done. This made it difficult to do comparisons over the exact same geographic area by latitude and longitude. It had to be done by eye instead. Cape Cod, Massachusetts, was chosen as the site for comparison because it is by far the most recognizable feature in all the images. In each image, Cape Cod was outlined as a region of interest as shown below, and then corresponding histograms were created.

Results and Discussion

The following are the four cape cod images used for comparison, as well as their corresponding histograms.

Figure 18. NOAA 16 August 1, 2002 SAA image received at Wallops Island, VA (region of interest outline shown)

Histogram 6. NOAA 16 August 1, 2002 SAA image received at Wallops Island, VA

Figure 19. NOAA 16 August 1, 2002 image from GMU antenna

Histogram 7. NOAA 16 August 1, 2002 image from GMU antenna

Figure 20. NOAA 12 August 8, 2002 SAA image received at Wallops Island, VA

Histogram 8. NOAA 12 August 8, 2002 SAA image received at Wallops Island, VA

Figure 21. NOAA 12 July 31, 2002 image from GMU antenna

Histogram 9. NOAA 12 July 31, 2002 image from GMU antenna

For the two NOAA 16 datasets, one can see that the images are hardly distinguishable, and the histograms are very similar. This shows that what the GMU antenna is picking up is the same as what the Wallops Island receiving station is picking up. The NOAA 12 datasets look different because of the difference in date and the difference in cloud cover.

However, it is interesting to note the difference between the NOAA 12 and NOAA 16 images. Application of the SMAC algorithm would cause the NOAA 12 images to become at first much darker than their NOAA 16 counterparts. Then, after the NDVI algorithm was applied, the land surface would emerge much brighter than in the NOAA 16 images, while the ocean stayed almost pure black, and the clouds an intermediate dark gray. It is hard to say for sure whether the SMAC algorithm improves the accuracy of the NDVI, but it certainly appears to do so, based in the heightened contrast between land, water, and clouds. Also, it is interesting to note that the SMAC divides the land itself into discrete classes (note the sharp peaks on the histogram), whereas the NOAA 16 images have a more gradual transition between values. This would lend it well to creating landcover maps based on classification tables.

Below are the twelve NOAA 16 histograms produced for the Cape Cod area, used to determine how NDVI changes seasonally over the year 2002.

Histogram 10. NOAA 16 Cape Cod January 5, 2002

Histogram 11. NOAA 16 Cape Cod February 2, 2002

Histogram 12. NOAA 16 Cape Cod March 1, 2002

Histogram 13. NOAA 16 Cape Cod April 2, 2002

Histogram 14. NOAA 16 Cape Cod May 1, 2002

Histogram 15. NOAA 16 Cape Cod June 1, 2002

Histogram 16. NOAA 16 Cape Cod July 4, 2002

Histogram 17. NOAA 16 Cape Cod August 1, 2002

Histogram 18. NOAA 16 Cape Cod September 5, 2002

Histogram 19. NOAA 16 Cape Cod October 1, 2002

Histogram 20. NOAA 16 Cape Cod November 2, 2002

Histogram 21. NOAA 16 Cape Cod December 4, 2002

The most telling thing to note in these histograms is the movement of the maximum pixel value. In general, the maximum value grades from just below 150 in the winter, to at 150 in the spring and fall, to just above 150 in the summer. Remember, according to our scale 150 equals an NDVI of 0.5. This is what would be expected.

According to the Woods Hole Research Center, 50.3% of the cape cod area is forested, woody, and open land, which would show seasonal change. The majority of the rest (31%) is commercial and residential, as is shown on the land cover map below.

Figure 20. 1990 Cape Cod Land Cover map

On top of having significant wooded areas, the Cape's weather changes significantly with the seasons, according to the National Park Service. Spring can often be cool and damp with temperatures ranging in the mid-40s. Summer usually provides warm days, ranging between 70 and 80 degrees, and cool nights. Winter on Cape Cod is milder than inland, with temperatures ranging between 30 and 40 degrees in mid-winter.

This all points to the fact that the NDVI there should change seasonally, which it indeed did. And, given the high error involved in comparing values using non-georeferenced images, the fact that this trend still emerged is a very good sign.

Sources of error

Error in this experiment comes from both the satellite data itself, and the processing methods. Errors in the satellite data can be caused by:

1. Cloud contamination. NDVI datasets distributed over the internet (such as the 1 deg X 1 deg global monthly average set distributed via the NASA DAAC) generally include cloud screening as part of their NDVI derivation process. This can be done by eliminating values with a channel 5 brightness temperature below 273 degrees K (Los et al. 1994).

2. High scan-angles tend to produce erroneous values. The NASA DAAC dataset also eliminates any vales obtained at an angle of over 42 degrees.

3. Atmospheric constituents (especially in the images that were unable to be corrected using the SMAC algorithm) tend to decrease NDVI values. The red signal normally increases as a result of scattered upwelling light, whereas the near IR signal decreases because of scattering and water vapor absorption. This occurs in a fashion that varies greatly over space and time.

4. The solar zenith angle can cause a difference of up to +/- 0.1 in the NDVI value between low and high scan angles. This can lead to an inconsistency between data collected at the same location over a different period of time (such as summer vs. late autumn) (Los et al. 1994).

5. Difference between lower NDVI values can sometimes be caused by variations in the composition of the soil background, rather than differences in the amount of vegetation. This problem in inherent in the formulation of the NDVI equation itself. The EVI (Enhanced Vegetation Index), developed by Huete et al. (1994), is an index that helps to correct for this problem.

6. Sellers (1985) has found that the NDVI is severely altered by the canopy structure. Specifically, variables such as the presence of dry and dead plant material, the leaf angle of the vegetation, and the distribution of vegetation vs. bare ground in a pixel can throw off the value. The best way to correct for this is to create site-specific regression plots.

Sources of error in the processing methods comes from the fact that in each image, Cape Cod was shaped slightly differently, due to the angle of the satellite view at that moment. Sometimes these differences could be quite large. So, a different shape file had to be created for each image, and it is inevitable that the exact same portion of land was not covered each time.

Conclusion

The results of this experiment could be highly improved by georeferencing the data, applying a cloud mask, eliminating high scan-angle values, and using only atmospherically corrected data. However, as it is stands, the experiment shows that the GMU antenna data combined with WinChips processing software can be used to produce an NDVI product that has some validity.

References

"An Electronic Atlas of the Cape Cod Environment" The Woods Hole Research Center. http://www.whrc.org/ccatlas/landcover/landcover.htm [Accessed March 10, 2003]

Asrar, G., Fuchs, M., Kanemasu, E. T., and Hatfield, J. L. (1984), Estimating absorbed photosynthetic radiation and leaf area index from spectral reflectance in wheat, Agron. J., 76:300-306.

Baret, F. and Guyot, G. (1991), Potentials and limits of vegetation indices for LAI and APAR assessment, Remote Sens. Environ., 35:161-173.

"Cape Cod National Seashore" National Park Service. http://www.nps.gov/caco/ [Accessed March 10, 2003]

Cihlar, J., J. Beaubien, and B. Park, (2001), BOREAS Follow-On DSP-01 NBIOME Level-4 AVHRR Land Cover, Canada, Ver[sion] 1.1, 1995. Data set. Available on-line [http://www.daac.ornl.gov] from Oak Ridge National Laboratory Distributed Active Archive Center, Oak Ridge, Tennessee, U.S.A.

Curran, P. J. (1980), Relative reflectance data from preprocessed multispectral photography, Int. J. Remote Sens., 1:77-83.

Deering, D. W. (1978), Rangeland reflectance characteristics measured by aircraft and spacecraft sensors. Ph.D. Dissertation, Texas A & M University, College Station, TX, 338 pp.

Goward, S. N. and Huemmrich, K. F. (1992), Vegetation canopy PAR absorptance and the normalized difference vegetation index: an assessment using the SAIL model, Remote Sens. Environ., 39:119-140.

Graetz, D. (1990), Remote sensing of terrestrial ecosystem structure: an ecologist's pragmatic view, In Remote Sensing of Biosphere Functioning; Ch. 2.,pp 5-30. Springer Verlag, N.Y., 312pp.

Huete, A. and Justice, C., (1991), Modis Vegetation Index Algorithm Theoretical Basis Document, ver. 3. http://modis.gsfc.nasa.gov/data/atbd/atbd_mod13.pdf

Huete, A., Justice, C. and Liu, H. (1994), Development of vegetation and soil indices for MODIS-EOS, Remote Sens. Environ., 49:224-234.

Hutchinson, C. F. (1991), Use of satellite data for Famine Early Warning in sub-Saharian Africa, Int. J. Remote Sensing, 12(6):1405-1421.

Jordan, C. F. (1969), Derivation of leaf area index from quality of light on the forest floor, Ecology, 50:663-666.

Kaufman, Yoram. "Measuring Vegetation (NDVI & EVI)" NASA Earth Observatory Library. http://earthobservatory.nasa.gov/Library/MeasuringVegetation/measuring_vegetation_2.html [Accessed: March 10, 2003]

Keong, L. KWOH and Hwa, K. LIM, (2001), Establishment of a NOAA and Fengyun Receiving Station in a Singapore High School, Centre for Remote Imaging, Sensing and Processing (CRISP), National University of Singapore.

Kidwell, Katherine B., (2000), NOAA KLM User's Guide. http://www2.ncdc.noaa.gov/docs/klm/

Kidwell, Katherine B., (1998), NOAA Polar Orbiter Data User's Guide. http://www2.ncdc.noaa.gov/docs/podug/

Los, S.O., C.O. Justice, C.J. Tucker, (1994). A global 1 by 1 degree NDVI data set for climate studies derived from the GIMMS continental NDVI data. International Journal of Remote Sensing, 15(17):3493-3518.

Myneni, R. B, Keeling, C. D., Tucker, C. J., Asrar, G., and Nemani, R. R. (1997),

Increased plant growth in the northern high latitudes from 1981 to 1991, Nature 386:698-702.

Prince, S. D. and Justice, C. O. (1991), (Editorial) Special issue on coarse resolution remote sensing of the Sahelian environment, Int. J. Remote Sensing, 12(6):1137-1146.

"Producing an NDVI Product" Chesapeake Bay and Mid-Atlantic from Space. Towson University Center for Geographic Information Sciences. http://chesapeake.towson.edu/data/ndvi2.asp [Accessed: March 10, 2003]

Rahman, H. and Dedieu, G., (1994), SMAC: a simplified method for the atmospheric correction of satellite measurements in the solar spectrum, International Journal of Remote Sensing, 15:123-143.

Running, S. W. and Nemani, R. R. (1988), Relating seasonal patterns of the AVHRR vegetation index to simulated photosynthesis and transpiration of forest in different climates, Remote Sens. Environ., 24:347-367.

Sellers, P. J. (1985), Canopy reflectance, photosynthesis and transpiration, Int. J. Remote Sensing, 6:1335-1372.

Sellers, P.J., J.A. Berry, G.J. Collatz, C.B. Fields, F.G Hall, (1992). Canopy reflectance, photosynthesis and transpiration. III. A reanalysis using improved leaf models and a new canopy integration scheme. Remote Sensing of Environment, 42:187-216.

Sellers, P.J., S.O. Los, C.J. Tucker, C.O. Justice, D.A. Dazlich, G.J. Collatz, and D.A. Randall, (1995). A revised land surface parameterization (SiB2) for atmospheric GCMs. Part 2: The generation of global fields of terrestrial biophysical parameters from satellite data. Submitted to Journal of Climate.

Tucker, C. J., Holben, B. N., Elgin, J. H., and McMurtrey (1981), Remote sensing of total dry matter accumulation in winter wheat, Remote Sens. Environ., 11:171.

Tucker, C. J. and Sellers, P. J. (1986), Satellite remote sensing of primary productivity, Int. J. Remote Sensing, 7:1395-1416.