The use of satellite imagery provides a unique vantage point for observing seasonal dynamics of the landscape that have implications for global change issues. Recent work by Myneni and others (1997) presents satellite-derived evidence that the photosynthetic activity of terrestrial vegetation increased from 1981 to 1991 due to an increase in plant growth associated with an increase in the duration of the active growing season. There has also been an increase in the amplitude of the seasonal cycle of atmospheric carbon dioxide since the early 1970s, and an advance in the timing of the drawdown of CO2 in spring and early summer of up to 7 days (Keeling and others 1996). Myneni also points out that variations in the amplitude and timing of the seasonal cycle of atmospheric CO2 show an association with surface air temperature consistent with the hypothesis that warmer temperatures have promoted increases in biospheric activity outside the tropics. A likely cause is an increase in the length of the active growing season brought about by warmer temperatures.
Improved methods of documenting the seasonal ecosystem dynamics may provide additional evidence to support such hypotheses of the impacts of global change. This paper describes one such method and applies it over data covering the state of Colorado. Traditionally, ground observations of seasonal characteristics have provided information concerning specific plants over a limited spatial area. Satellite data analysis, however, provides the means to measure broad-scale changes at the ecosystem level. Multitemporal satellite-derived vegetation observations have allowed researchers to quantify seasonal events and to characterize vegetation according to its seasonal patterns (Reed and others 1994, Loveland and others 1993).
Vegetation indices such as the normalized difference vegetation index (NDVI) are derived from a simple ratio based on the contrast in spectral reflectance of photosynthetically active vegetation that characteristically has low reflectance in the visible (VIS) portion of the spectrum and high reflectance in the near-infrared (NIR). Thus, variations in green vegetation density can be analyzed with the following equation:
NDVI = (NIR-VIS)/(NIR+VIS)
In addition, a direct relationship between the integrated seasonal values of NDVI and seasonal
biomass accumulation has been documented (Tucker and others 1981; Daughtry and others 1983; Tucker
and others 1985). The combination of NDVI with daily coverage and 1-km spatial resolution
make the Advanced Very High Resolution Radiometer (AVHRR) satellite sensor well suited
for regional to global scale studies that assess ecosystem dynamics. Figure 1 is an animated
loop of the NDVI observed over Colorado from 1990 through 1996.
NDVI time-series smoothing
The NDVI is affected by a number of phenomena including cloud contamination, atmospheric perturbations, and variable illumination and viewing geometry, all of which usually reduce the NDVI value (Los and others 1994). To address these effects, NDVI data are often composited using the maximum value over a specified time period: usually a week, ten days (dekad), or two weeks. While maximum value compositing increases data quality, further processing-- in effect a smoothing-- of the temporal NDVI signal must be performed to facilitate some time-series analyses. The smoothing algorithm must serve as a rough interpolation between observations and, in order to remove the effects of the remaining NDVI-reducing phenomena, upwardly bias the results.
An iterative, compound median smoother has been implemented at the U.S. Geological
Survey (USGS) EROS Data Center that accomplishes these goals. The smoother iteratively
processes the time-series by applying median filters of various widths, then applies a "re-
roughing" by reintroducing the original NDVI time-series into the process. The upward bias,
or NDVI "peak-catching", is applied by reintroducing unsmoothed NDVI values that are
greater than the smoothed values. Further work is being conducted that investigates other
smoothing approaches, such as Fourier analysis similar to that used in the FASIR adjustments
(Sellers and others 1994) and other statistical approaches. In Figure 2, a three-year time series of
NDVI is illustrated in the solid line, while the result of the compound median smoother are
shown as a dashed line. The smoother eliminates cloud contamination (illustrated by the
extremely low NDVI value in Year 2), as well as NDVI reducing perturbations (illustrated
during the greenup during Year 1).
Once the database is smoothed of temporal discontinuities, methods can be applied to extract
a suite of seasonal characteristics from the time-series data set. Some of the more important
seasonal characteristics that are needed are the time of the start of the growing season, the
time of maximum photosynthetic activity, and the duration of the growing season. Reed and
colleagues (1994) developed a methodology to derive a set of 12 seasonal characteristics from
the smoothed NDVI time series that summarizes characteristics of ecosystem dynamics. The
methodology involves applying a moving average filter to the time series, which essentially
creates a new time series with a time lag. The moving average time-series (MATS) then can
serve as a predicted NDVI based on the previous n (user-defined) observations. When the
actual (smoothed) values are greater than the value predicted by the MATS, then a trend
change (start of growing season) is occurring (Figure 3). The end of the growing season can be
found similarly and the duration can be calculated as the difference between the two.
Other seasonal characteristics such as the time of maximum NDVI and the time-integrated
NDVI (using the value of NDVI at start of season as a baseline) are also important surrogate
measure of ecosystem characteristics.
Land characterization from AVHRR
Trends in the seasonal characteristics become more apparent when analysis is stratified according to land cover type. The land cover data used to stratify the Colorado data set was derived from a database developed by Loveland and others (1991 and 1993). Time-series AVHRR NDVI data from 1990 was used to characterize the land cover with methods that can be described as a multitemporal unsupervised classification of NDVI data with post- classification refinement using multisource earth science data.
The post-classification refinement is performed with the aid of digital elevation, ecoregions
data and a collection of other land cover/vegetation reference data (Brown and others 1993).
The interpretation is based on extensive use of computer-assisted image processing tools
(Brown and others, in press); however, the classification process is not completely automated
and resembles a traditional manual image interpretation technique. One hundred fifty-nine
seasonal land cover regions resulted from the analysis of the conterminous United States.
The seasonal land cover regions were then cross-referenced to several different classification
schemes, including an adaptation of the USGS/Anderson Level II scheme (Anderson and
others 1976) (see Figure 4).
Figure 5 shows smoothed time-series NDVI data for representative pixels of several land cover types in Colorado observed on a biweekly basis from 1990 to 1996. Between-year variability of the cropland pixel is apparent, with the low values in 1991 and 1994. The shrubland pixel also shows significant variability, with low peak values in 1990 and 1991. The grassland pixel shows relative consistency in the amplitude of the NDVI values, but exhibits differences in the length of the growing season. The length of the growing season in 1992 appears to be significantly longer than that illustrated in 1994. The coniferous forest curve exhibits a marked consistency, both in amplitude and in length of growing season. While the coniferous forest is green throughout the year, background effects (such as snow cover) affect the seasonal NDVI profile.
Figures 6 through 9 show images depicting four of the seasonal characteristics, including the time of the start of the growing season, time of maximum NDVI, duration of growing season, and time-integrated NDVI over a five year period from 1991 to 1995. Since the calculation of the seasonal characteristics requires data that both precede and follow the NDVI data for a single year, 1990 and 1996 are not shown.
Figures 6 and 10:
The time of the onset of the growing season in Colorado for 1991-1995 is depicted in Figure 6. Features that regularly stand out are the late onset of growing season for the coniferous forests, especially in 1995 and the croplands in northeastern Colorado. Shrublands and grasslands have an earlier onset of greenness. The bar chart in Figure 10 illustrates the interannual variability in the time of the start of the growing season over the five years, but does not exhibit any noticeable trend of earlier or later starts to the growing season.
Figures 7 and 11:
Figure 7 illustrates the time of maximum NDVI for 1991-1995. There is some consistency in the spatial pattern with the croplands in northeastern Colorado reaching their peak around the 17th or 18th biweekly period (July-August), while the surrounding grasslands reach their peak much earlier in the growing season. Similarly the coniferous forests reach their peak in mid- late summer while the shrublands and grasslands of western Colorado reach their peak greenness in May. Interannual differences are exhibited, especially in 1994 (with an overall earlier time of maximum) and 1995 (with a later maximum, particularly in the coniferous forest). Again, the bar chart (Figure 11) does not show any clear trends of earlier or later times of maximum NDVI emerging, but rather shows remarkable consistency in the timing of the maximum NDVI.
Figures 8 and 12:
Time-integrated NDVI over the five years of observation is shown in Figure 8. Of the four seasonal characteristics depicted in this paper, time-integrated NDVI (a surrogate for production) probably has the least variability over the years of observation. The images do show some distinct land cover patterns with the croplands in northeastern and southeastern Colorado and the coniferous forests standing out in comparison to the surrounding land cover types. The bar chart (Figure 12) does not show a high degree of interannual variability in the four land cover types, and no clear trends emerge in the five years of observation.
Figures 9 and 13:
Figure 9, depicting the duration of the growing season, shows relatively consistent growing seasons from 1991 to 1994, with 1993 having a slightly shorter growing season, particularly in the coniferous forests. The growing season in 1995 is noticeably longer, especially in the shrubland areas of western Colorado and the shrublands and grasslands of the eastern half of the state. While five years of data are certainly not sufficient to speculate on any trends toward increasing or decreasing lengths of growing season, the data do not show any discernible pattern (Figure 13).
This paper illustrates some methods that can be used to monitor the impacts of global change
upon the landscape. Recent work has shown a relationship between CO2
and both seasonal duration and productivity. The methods described in this paper can provide a means of
quantifying the seasonal characteristics of the land surface at a 1-km spatial resolution
worldwide. While the results of the paper do not establish any clear trends in the seasonal
duration of Colorado, due to the relatively short time of observation, the methods show
promise for observing trends.
Anderson, J.R., Hardy, E.E., Roach J.T., and Witmer R.E., 1976, A Land Use and Land Cover Classification System for Use with Remote Sensor Data. U.S. Geological Survey Professional Paper 964, Reston, VA: U.S. Geological Survey.
Belward A. S., and Loveland., T.R., 1995, The IGBP-DIS 1-km Land Cover Project: Remote Sensing in Action. In Proceedings of the 21st Annual Conference of the Remote Sensing Society, Southampton, United Kingdom, p. 1,099-1,106.
Brown, J.F., Loveland, T.R., Merchant, J.W., Reed, B.C., and Ohlen, D.O., 1993, Using Multisource Data In Global Land Cover Characterization: Concepts, Requirements and Methods. Photogrammetric Engineering and Remote Sensing 59:977-987.
Brown, J.F., Reed, B.C, and Huewe L., in press. Advanced Strategy for Multi-Source Analysis and Visualization in Land Cover Characterization. In Proceedings, Pecora 13, Human Interactions with the Environment: Perspectives From Space.
Daughtry, C.S.T., K.D. Gallo, and M.E. Bauer, 1983. Spectral estimates of solar radiation intercepted by corn canopies. AgRISTARS Tech. Report SR-PZ-04236. Purdue University, West Lafayette, IN.
Keeling, C. D., Chin, J. F. S. & Whorf, T. P., 1996. Increased activity of northern vegetation inferred from atmospheric CO2 measurements, Nature 382, 146-149, (1996).
Loveland, T.R., J.W. Merchant, D.O. Ohlen, J.F. Brown, 1991. Development of a land-cover characteristics data base for the conterminous U.S. Photogrammetric Engineering and Remote Sensing, 57, 1453-1463.
Loveland, T.R., Merchant, J.W., Brown, J.F., Ohlen, D.O., Reed, B.C., Olson, P., and Hutchinson, J., 1995, Seasonal Land-Cover Regions of the United States, Annals of the Association of American Geographers, 85(2): 339-355.
Los, S.O., C.O. Justice, and C.J. Tucker, 1994. A global 1 by 1 NDVI data set for climate studies derived from the GIMMS continental NDVI data, International Journal of Remote Sensing, Vol. 15, No. 17, 3493-3518.
Myneni, RB, C.D. Keeling, C.J. Tucker, G. Asrar and and R. R. Nemani. 1997. Increased plant growth in the northern high latitudes from 1981 to 1991. Nature 386: 695-702.
Reed, B.C., J.F. Brown, D. VanderZee, T.R. Loveland, J.W. Merchant, and D.O. Ohlen, 1994. Measuring phenological variability from satellite imagery. Journal of Vegetation Science. 5: 703-714.
Sellers, P.J., C.J. Tucker, G.J. Collatz, S.O. Los, C.O. Justice, D.A. Dazlich, and D.A. Randall 1994. A global 1 by 1 NDVI data set for climate studies. Part 2: The generation of global fields of terrestrial biophysical parameters from the NDVI, vol. 15, No. 17, 3519-3545.
Tucker, C.J., B.N. Holben, J.H. Elgin, Jr., and J.E. McMurtrey, 1981. Remote sensing of total dry-matter accumulation in winter wheat. Remote Sensing of Environment. 11:171-189.
Tucker, C.J., J.R.G. Townshend, and t.E. Goff, 1985. African land cover classification using satellite data. Science. 227.