Improving Vegetation Monitoring and Assessment
Researchers complete a comparison of AVHRR and SeaWiFS satellite data for land vegetation assessment.
By David J. Conrad and Thomas W. Wagner

Introduction
The SeaStar satellite is a space-based remote sensing system scheduled to be launched into a sun-synchronous orbit in early 1997 by Orbital Sciences Corp. (OSC). It will be operated by OSC subsidiary Orbital Imaging Corp. (ORBIMAGE) based in Dulles, Va. The SeaStar satellite will carry a single earth-observing system, the Sea-viewing Wide-Field-of-view Sensor (SeaWiFS) Ñ the successor to the Coastal Zone Color Scanner (CZCS) that flew on Nimbus-7 from 1978-1986. Using six visible and two near infrared spectral bands, SeaWiFS is designed to provide high quality images of the oceans surface Ñ including images of its chlorophyll and transparency. However, during development, the sensors gains were modified to a bilinear profile which gives SeaWiFS the added capability of obtaining usable data from the more highly reflecting land surfaces.
      One potential land application for SeaWiFS is in providing improved Vegetation Index (VI) data. One vegetation index is to measure variations in the greenness of vegetation (percent cover, biomass, leaf area index, crown closure) while diminishing other variations such as slope, soil type, or soil moisture. VI data has been useful for the stratification of vegetation for large areas into rangeland, crop land, or forest.
      This article describes a simulation of SeaWiFS data to provide a VI in comparison with one from NOAAs Advanced Very High Resolution Radiometer (AVHRR). The AVHRR has been operational on NOAA spacecraft for many years. While the sensor was designed to meet a meteorological mission, its data has been routinely used to provide several vegetation indices, including the well-known Normalized Difference Vegetation Index (NDVI).
      Vegetation indices derived from AVHRR data are normally based on only two reflective bands Ñ one visible and one near-infrared. Alternatively the SeaWiFS sensor, with a similar spatial resolution (approximately 1 km) and temporal coverage (daily), offers an opportunity to exploit the advantages of greater spectral diversity in vegetation monitoring applications. The SeaWiFS also has a full on-board calibration suite.
      Researchers at the Environmental Research Institute of Michigan (ERIM) recently completed a comparison of AVHRR and SeaWiFS for land vegetation assessment. The study made the following comparisons:
• scene/sensor dynamic range, in relation to band saturation radiance;
• the effect of sensor quantization on vegetation assessments; and
• the effect of spectral diversity, including additional spectral bands in calculating a vegetation index.
      This article focuses on illustrating the third element, the effect of additional spectral bands in calculating vegetation indices.

Toward Improved Vegetation Indices
The presence of only two reflective bands on AVHRR (red and near infrared (NIR)) limits the type of vegetation indices possible from these data. The NDVI, shown below, has become the standard metric for vegetation monitoring.
      While this metric may be correlated with vegetation cover and condition, it may vary significantly with atmospheric haze, a source of confusion for vegetation monitoring. The Atmospherically Resistant Vegetation Index (ARVI) reduces the variance related to haze by using a blue band in the computation.
      The large haze influence in the blue band is used to correct the lesser haze influence in the red band.
      Crist and Horvath, 1995, provide the basis for an illustration of the improvement that SeaWiFSs can have in producing VIs in the presence of atmospheric haze. Specifically, imagery collected by the SeaWiFS predecessor, the Coastal Zone Color Scanner (CZCS), was processed to derive both NDVI and ARVI values. While having fewer spectral bands (6 as opposed to 8), the CZCS is similar to SeaWiFS in its spatial resolution and area of coverage.
      The imagery shown here was collected on June 11, 1980 and shows most of Michigans upper peninsula, along with smaller portions of Canada on the far right. The large body of water is Lake Superior. Figure 2 shows atmospheric haze which is oriented north-west in this scene, perpendicular to the east-west trending northern hardwood forests which cover much of the area. Additionally, the haze varies from moderate to little (left to right) across the scene. Figures 3a and 3b show the results of NDVI and ARVI calculations, respectively. In these two processed images the same color coding was used, as characterized by each respective VI, with dark green corresponding to the most densely vegetated areas. Comparison of these two images shows differences in the hazy areas of the scene. From ground truth and forest maps, the ARVI image is more accurate. This result suggests that ARVI may be used to produce a more stable vegetation index than NDVI. Since calculation of ARVI requires a blue band, it is not available from AVHRR data.
      Our analyses to date have focused on the potential utility of SeaWiFS data to improve vegetation monitoring and assessment, and since they are based on simple and limited simulations, they ought not to be taken as absolute. They do provide clear indications, however, of the potential advantage afforded by SeaWiFS over existing sources for producing accurate, global estimates of green vegetation cover and development. Additionally, this is only one of a number of approaches that may make use of SeaWiFS for land applications. It is easy to speculate that other approaches may be more useful, including spectrum matching procedures developed for land applications with other sensors such as Landsat and AVHRR. Even more powerful approaches are likely to employ multi-temporal data acquisitions and multi-stage sampling with other sensor systems.

About the Authors:
The authors are both employed by the Environmental Research Institute of Michigan in Ann Arbor, Mich. David J. Conrad is a research engineer in the Information and Materials Analysis Laboratory and Thomas W. Wagner is a research scientist in the Center for Earth Science. For further information concerning this article, either may be reached at 313-994-1200.

Back