| Introduction | Summary| Methods
| Acknowledgements | Results and Discussion
| Literature
| |
Grasslands represent the largest vegetative province in North America
(Samson and Knopf, 1994), and the tallgrass prairie is the most mesic and
productive grassland type in this biome. The tallgrass prairie once covered
67.6 million ha, extending from Kansas to Ohio and Texas to Canada, but
Samson and Knopf (1994) report that 90% of all grasslands has been turned
by the plow. The Flint Hills region of Kansas and Oklahoma has the largest
extent of remaining tallgrass prairie. The region extends in a north-south
orientation from northeast Kansas to northern Oklahoma, with its widest
extent being about 70 km--see Fig. 1.
The Flint Hills are dominated by C4 grasslands with
a typical midwestern continental climate characterized by warm, wet summers
and dry, cold winters. Mean annual precipitation in this area is sufficient
to support forest or savanna vegetation, but drought, fire and grazing
play important roles in maintaining this grassland (Axelrod, 1985; Anderson,
1990). In the Flint Hills, spring burning of grasslands is common, and
grazing by domestic livestock is the dominant land use because the relatively
steep slopes and rocky soils prevent extensive establishment of row-crop
agriculture. In Kansas these grasslands support a livestock grazing industry
second only to Texas in animal-unit-months. Seastedt and Knapp (1993) concluded that in tallgrass prairie, complex
interactions among multiple limiting resources in combination with a variety
of land use practices, can lead to a heterogeneous landscape. Thus, any
attempt to examine this landscape must consider these factors. In an attempt
to quantify the tallgrass prairie landscape of the Flint Hills and to determine
what is responsible for these patterns, we used remote sensing data of
the Advanced Very High Resolution Radiometer (AVHRR) to calculate Normalized
Difference Vegetation Index (NDVI). One of the most common applications of remote sensing is monitoring
and evaluating vegetation over the land surface. Normalized Difference
Vegetation Index is an important tool in this endeavor. NDVI is
based on the fact that growing vegetation has a low red reflectance due
to absorption by chlorophyll and other plant pigments and a high near-infrared
reflectance due to internal reflectances involving the structure of green
leaves. Thus, vegetated areas should yield high values of NDVI. Many researchers
have demonstrated that NDVI is related to a variety of vegetative parameters
including percent cover, leaf-area index, and green biomass. (e.g. Goward
et al., 1985; Justice et. al. 1985; Box et al. 1989). Early in the 1990
growing season the U.S. Geological Survey's EROS Data Center (EDC) started
acquiring NOAA-11 AVHRR 1-km resolution daily observations to produce weekly
and biweekly maximum NDVI composites of the conterminous United States.
The objective of the vegetation mapping program is to "compile, annually,
a comprehensive series of calibrated, georegistered, daily observations
and biweekly maximum NDVI composites" (Eidenshink 1992). As stated
by EDC, these data sets provide a comprehensive growing season profile
of ecosystems and are extremely useful for assessing seasonal variations
in vegetation condition and provide a foundation for studying long-term
changes resulting from human or natural factors. We used these NDVI values
coupled with abiotic factors and land cover types to explore spatial and
temporal patterns of vegetation within the Flint Hills of Kansas and Oklahoma.
The equation produces NDVI values in the range of -1.0 to +1.0, where
negative values typically represent clouds, snow, water, and other non-vegetated
surfaces, and positive values represent vegetated surfaces. NDVI
was calculated from calibrated data that were scaled to byte range and
geometrically registered. In order to scale the computed NDVI results to byte data range, we added
1 to the computed value to the NDVI value and multiplied that by 100. As
a result, NDVI values less than 100 now represent clouds, snow, water,
and other non-vegetative surfaces and values greater than 100 represent
vegetative surfaces. Temperature and precipitation data were derived from a network of measurement
stations operated by the National Weather Service Cooperative Observer
Network (NWSCON). The cooperative observer network consists of volunteer
observers who agree to maintain a station, as well as record, tabulate,
and report the data. All NWSCON stations are equipped with a cylindrical
gauge for recording precipitation in liquid form. In addition, calibrated
measurement probes are used to record snowfall amounts. Some NWSCON stations
are also equipped with temperature measurement equipment. These consist
of a standard mercury-in-glass thermometer equipped with bulb constriction
for measuring maximum temperature, and an alcohol index thermometer for
recording the daily minimum. Thermometers at NWSCON stations are mounted
within a protective screen to shield them from direct incoming solar radiation
Total annual precipitation (TAP) and growing degree days (GDD) were
calculated using data from these stations for each of the years. GDD was
calculated according to the following.
INTRODUCTION
METHODS
We obtained AVHRR NDVI data from 1990-1994, which was collected and
distributed by the U.S. Geological Survey's EROS Data Center (EDC) in Sioux
Falls, South Dakota. NDVI is the difference of near-infrared (AVHRR channel 2)
and visible(AVHRR channel 1) reflectance values divided by total reflectance.
IR(channel 2) - Visible(channel 1)
NDVI = ----------------------------------
IR(channel 2) + Visible(channel 1)
daily maximum temperature + daily minimum temperature
GDD = ----------------------------------------------------- - threshold temperature
2
Values were summed
on a daily basis. Values below zero were set to zero. The threshold temperature
was set at 10°C. This temperature was used since it best describes optimal
growing temperature for C4 grasses. For our analysis with GDD
and TAP, we created annual integrated NDVI images from the biweekly maximum
NDVI composites. GDD was used since Goodin and Henebry (1996) determined
it to be sensitive to NDVI curves when they are calculated using close-range
sensing data.
In an attempt to assess the importance of land cover on NDVI values, a land cover map from the Kansas Land Cover Mapping Project (Kansas Applied Remote Sensing Program) was obtained for the Kansas portion of our study site--see Flint Hills land cover classification. This land cover map, which was based upon Landsat Thematic Mapper data in 1990, produced 10 distinct cover types--see Table 1. Data were compared against manually interpreted 1985 National High Altitude Photography (NHAP) and 1986 Kansas reappraisal air photos. Coordinates of known homogenous (at least 2 AVHRR pixels) land cover areas (grassland, cropland and woodland) were placed into a geographical information system (GIS). The GIS coverages were then used to obtain sub-samples from the bi-weekly maximum NDVI composites for each year (19 in 1990 and 21 in 1991-1993) to determine how these cover types contributed to the variance in the NDVI over time.
| Land Cover Class | Percentage of Area |
|---|---|
| Residential | 0.66% |
| Commercial/Industrial | 0.29% |
| Urban Grasslands | 0.26% |
| Urban Woodlands | 0.01% |
| Urban Water | 0.01% |
| Cropland | 24.33% |
| Grasslands | 65.30% |
| Woodlands | 6.72% |
| Water | 2.07% |
| Other (mostly highways) | 0.36% |
During our four year study period, which included years of below average precipitation (1990) to near record precipitation (1993), NDVI values varied across the Flint Hill landscape. In every year, NDVI tended to be highest in the southeast section, which corresponds to known precipitation patterns. The 30-year mean of precipitation in the southeast is around 1110 mm (Pawhuska, Oklahoma) while in the northwest, the 30-year mean is around 835 mm (Manhattan, Kansas). Multiple linear regression models using a stepwise procedure determined that a combination of GDD or TAP had r² varying from 0.39 to 0.88--see Table 2 and Fig. 2. When all four years were combined, GDD was the only significant variable (r² = 0.32, Table 2). Thus, other factors (i.e. land use) are important in the spatial and temporal patterns measured in this landscape. Other abiotic variables (e.g. growing season precipitation, temperature or GDD based upon different threshold temperatures) did not improve any of the relationships.
| Year | NDVI | r² | P |
|---|---|---|---|
| 1990 | NDVI = 119.19 -3.003 * (GDD/TAP) | 0.49 | 0.002 |
| 1991 | NDVI = 82.69 + 0.00000714 * (GDD *TAP) | 0.83 | 0.001 |
| 1992 | NDVI = 69.33 + 0.009 * GDD | 0.61 | 0.001 |
| 1993 | NDVI = 42.0 + 0.168 * GDD | 0.88 | 0.001 |
| All Years | NDVI = 75.52 + 0.008 * GDD | 0.32 | 0.001 |
A two-way ANOVA of the slopes of the 'greenup' and 'senescence' phases of the NDVI curves was conducted to determine the effects of cover types on the NDVI slopes. Greenup phase equals the points from day of year 0 on the NDVI curve to the maximum value while senescence phase equals points from the maximum value on the NDVI curve to the last composite of that year. The slopes were then compared using a GLM with year and cover type as class variables.
Year was not significant, but cover type was significant (P < 0.05). The slope of the greenup phase (NDVI units/day) on grasslands and woodlands was not significantly different from each other (P > 0.05), but was different from cropland (P < 0.05). Similarly the slope of the greenup phase on woodlands and croplands was not significantly different from each other (P > 0.05) but was significantly different from grasslands--see Fig. 3.
Cover type had no significant effect on the slope, but significant differences occurred from year to year. 1990 and 1991 were not different from each other but were different from 1992 and 1993. In addition, 1992 and 1993 were significantly different from other years--see Fig. 4.
These differences between the greenup-phase and the senescense phase suggest that croplands are more sensitive to variation in temperature and precipitation, especially early in the growing season. This difference may be explained simply by the fact that farmers can exercise some option in planting time, crop-type, etc. However, during the senescence phase all cover types are more sensitive to seasonal precipitation rather than temperature.
At a single tallgrass prairie site (Konza Prairie; northwest section of the Flint Hills) over a 19-year time period, abiotic factors were used in predicting peak above ground biomass (Briggs and Knapp, 1995). They reported that although some factors were statistically significant, a large percentage of the variation was not explained. They concluded that no single factor is responsible for the interannual variability in peak aboveground biomass. It appears from our study using AVHRR that this is also true over a larger spatial area. Although a significant relationship between GDD and NDVI in the Flint Hills occurs in most years, overall a large percentage of the variance remains unexplained.
In contrast with many other North American grasslands where shortages of a single resource often dominate system responses, humid tallgrass is best viewed as being limited by multiple resources such as water, light, nitrogen (Knapp and Seastedt, 1986; Chapin et al., 1987; Seastedt and Knapp, 1993). The combination of a highly variable continental climate and various management practices (i.e., plowing, grazing and fire) can lead to a very heterogeneous landscape where the relative importance of limiting factors, and the structural and functional responses of the ecosystem, vary both in space and time. Thus, many ecological patterns and processes in tallgrass prairie are best considered from a non-equilibrium perspective, where frequent shifts in the relative importance of key multiple resources are crucial for maintaining both the diversity and productivity of these ecosystems (Seastedt and Knapp, 1993). Any attempt to examine the spatial and/or temporal patterns in this landscape should incorporate these ideas with an understanding of the non-equilibrium of this system.
We conclude that understanding land cover type in the Flint Hills is critical in defining the spatial and temporal patterns in this relatively homogenous vegetation cover (grassland). Management of the area (mostly grazing and burning) that produces these various land cover types is also very important and any land coverage analysis should try to incorporate land management practices. For example, in our analysis, over 65% of the area was simply classified as grassland (Table 1). We know from research conducted on Konza Prairie, that land management practices such as burning and grazing alter the spectral signatures of the landscape (Briggs and Nellis 1991; Turner et al., 1992; Goodin and Henebry, 1996). Thus any further research plans for examining the spatial patterns of vegetation in the Flint Hills must involve developing a land coverage, which incorporates land management practices such as grazing and burning history.
Anderson, R.C. 1990. The historic role of fire in the North American
grassland. In S.L. Collins and L. L. Wallace (eds.): Effects of
Fire on Tallgrass Prairie Ecosystems. University of Oklahoma Press, pp
8-18. Axelrod, D.I. 1985. Rise of the grassland biome, central North America.
The Botanical Review 51(2): 163-201. Box, E.O., B.N. Holben, , and V. Kalb. 1989. Accuracy of the AVHRR vegetation
index as a predictor of biomass, primary production and net CO2
flux. Vegetatio 80(2): 71-89. Briggs, J.M. and A.K. Knapp. 1995. Interannual variability in primary
production in tallgrass prairie: climate, soil moisture, topographic position
and fire as determinants of aboveground biomass. American Journal of Botany
82(8): 1024-1030. Briggs, J.M. and M.D. Nellis. 1991. Seasonal variation of heterogeneity
in tallgrass prairie: A quantitative measure using remote sensing. Photogrammetric
Engineering and Remote Sensing 57(4):407-411. Chapin, F.S. III, A. J. Bloom, C. B. Field, and R.W. Waring. 1987. Plant
responses to multiple environmental factors. BioScience 37(1):49-57. Eidenshenk, J. C. 1992. The 1990 conterminous US AVHRR data set. Photo-grammetric
Engineering and Remote Sensing 58(6):809-913. Goward, S.N., C.J. Tucker, and D.G. Dye. 1985. North American vegetation
patterns observed with the NOAA-7 Advanced Very High Resolution Radiometer.
Vegetatio 64(1):3-14. Goodin, D.E. and G.M. Henebry. 1996. Seasonal NDVI trajectories in response
to disturbance: toward a spectral-temporal mixing model for tallgrass prairie.
IEEE Trans. Of Geoscience and Remote Sensing, Volume IV. 215-217. Justice, C.O, B. L. Markham, J.R.G. Townshend, B.N. Holber and C.J.
Tucker. 1985. Analysis of the phenology of global vegetation using meteorological
data. International Journal of Remote Sensing 6(8):1271-1381. Knapp, A.K. and T.R. Seastedt. 1986. Detritus accumulation limits productivity
of tallgrass prairie. BioScience 36(10):662-668. Samson, F. and F. Knopf. 1994. Prairie conservation in North America.
BioScience 44(6):418-421. Seastedt, T.R. and A.K. Knapp. 1993. Consequences of non-equilibrium
resource availability across multiple time scales: The transient maxima
hypothesis. American Naturalist 141(4):621-633. Turner, C. L., T. R. Seastedt, M. I. Dyer, T. G. F. Kittel and D. S.
Schimel. 1992. Effects of management and topography on the radiometric
response of a tallgrass prairie. Journal of Geophysical Research 97(D17):18,855-18,666.
Return to Geospatial Snalysis Symposium
Comments to JMB@lter-konza.konza.ksu.edu
LITERATURE CITED