< Back  ||  Home > Reports > Grazing

Evaluating the Relationship between Spectral Indices and Grazing

Introduction

The RangeView project at the V-bar-V Ranch has utilized satellite imagery to evaluate the relationships between transect-based ground cover observations and AVHRR NDVI data, and to characterize the phenological and interannual dynamics of the NDVI for major vegetation types on the ranch. From a management perspective, understanding the feedbacks from livestock to apparent greenness is as important as evaluating and predicting the forage base within and between years. This analysis explores the potential for detecting change in spectral indices due to presence of livestock in a pasture by comparing AVHRR and Landsat-7 imagery from before and after the grazing period.

This analysis focuses on the Middle Snake Ridge pasture on the V-bar-V Ranch where grazing was scheduled for 23 days between 29 July and 20 August 1999 (Figure 1). Data from the 11-year AVHRR NDVI dataset and Landsat-7 ETM sensor are used to evaluate change in the spectral behavior of the surface during this time frame (Figure 2). The AVHRR NDVI time series for the pasture (average for 9 km^2) and for a single pixel corresponding to the TM subset are also used to provide a context for longer-term variability in the NDVI at these locations. Summer 1999 monthly rainfall totals for stations surrounding the ranch and Ikonos 1-m resolution panchromatic imagery are incorporated in the interpretations of temporal and spatial patterns of variability and change.

Figure 1. Location of the Middle Snake Ridge pasture on the V-bar-V Ranch. Area of detail shows the cloud-free subset area for TM spectral indices and the corresponding 1-km^2 AVHRR pixel; UTM coordinates (zone 12, NAD83). Image is the 26 April 2000 Ikonos 1-m panchromatic image with the TM subset area shown in red; mosaicking artifacts are present in the middle one-third of this image but have no bearing on the present analysis.

Figure 2. Temporal relationships between grazing period in the Middle Snake
Ridge pasture and imagery contributing spectral indices to the analysis.

Methods

Landsat-7 ETM 6-band images for 22 July 1999 and 23 August 1999 were orthorectified to a USGS-orthorectified reference image using a subset of the 30-m NED DEM as the elevation source (refer to <http://edcnts12.cr.usgs.gov/ned/>). Total RMSE were .315 and .303, respectively, using 29 and 21 image control points distributed across the full scene. The two scenes were atmospherically corrected using a modified dark object subtraction method based on the COST model (Chavez, 1996) which incorporates the date-specific Landsat-7 Calibration Parameter File band gain and offset values (see NASA, 2001).

Spectral indices, TM NDVI and the Kauth-Thomas "tasseled cap" transformation (Crist, 1985; Huang et al., 2000), were computed for each scene using the reflectance output from the atmospherically-corrected images. The original index values (ranging from -1 to +1) were utilized, rather than being rescaled, for subsequent analyses. Change images were derived for each index by subtracting the July image from the August image. Due to the presence of monsoonal clouds in the ETM+ imagery, only a subset of the Middle Snake Ridge pasture (399 pixels = .36 km^2) was analyzed. The pixel values of the cloud-free subset area for the original images were extracted and the distributions were compared (i.e. change was evaluated) statistically using the t-test for paired samples in SPSS 10.0 for Windows, with two-tailed significance at P<.025.

The AVHRR rescaled NDVI values (see Eidenshink, 1992) for the area of interest were extracted for each compositing period over the 11-year span of the dataset and used to develop long-term average and standard deviation time series for each pixel for each biweekly time frame. Additionally, the NDVI values for the 1-km^2 pixels within the Middle Snake Ridge pasture (n=9) were averaged to create a time series for the pasture as a whole. The long-term average and 1999 datasets for the pasture and for a pixel corresponding to the TM subset area (pixel 16980) were used in this analysis. The compositing time frames corresponding to before, during, and after grazing are the 1999 layers 15-17, and the long-term average layers for compositing time frames ending approximately on 29 July, 12 August, and 26 August. Differences in pasture NDVI values a) between dates in 1999, and b) between 1999 and the LTA were evaluated using the t-test for paired samples in SPSS 10.0 for Windows, with two-tailed significance at P<.025.

Results and Discussion

Spectral Change

The differences between July and August Landsat images were statistically significant for all four of the TM spectral indices (Table 1). Mean change in the NDVI, TM greenness, and TM wetness was negative, while mean change in TM brightness was positive (Figure 3).

Table 1. Results of t-test for paired samples comparing the means of the July and August
TM subsets (n=399 pixels) for each of the four spectral indices. Two-tailed significance
at P<.025 is flagged.

Figure 3. Summary of changes in spectral indices (TM NDVI and TM7 Kauth-Thomas
brightness, greenness, wetness) between 22 July and 23 August 1999 for the cloud-free
subset area (n=399 pixels) of Middle Snake Ridge pasture. Error bars indicate +2 standard
deviations (95% confidence interval of the mean). The change in each of these indices is
statistically significant at P<.025.

Temporal Variability

Change in the AVHRR NDVI for the pixel corresponding to the location of the TM subset area (pixel 16980) is significant and positive between the "before grazing" layer (about July 29) and the layer composited during the grazing period (about August 12), and is significant and negative between the "during" layer and the layer composited after the grazing period (about August 26). The net change in the NDVI between "before" and "after" grazing for this pixel is +3 rescaled NDVI units and is not statistically significant (Table 2). There is zero net change in the NDVI averaged for the pasture as a whole in 1999, although the long-term average change for the pasture in this time frame is comparable to the net change exhibited by pixel 16980. Both the pixel and pasture-level changes in the NDVI are contrary to the direction of change in the NDVI exhibited in the Landsat images, however the magnitude of change exhibited by the TM NDVI is not significantly different from the change in AVHRR NDVI between the two time frames (Table 3).

When compared to the long-term average (LTA) for pixel 16980, the changes to the NDVI over this time frame in 1999 are not significant for the compositing time frames corresponding to "before" or "after" grazing, but are significant for the "during" time frame (Table 2). This peak in the NDVI around mid-August may be anomalous in the general phenological pattern for this pixel and, indeed, for the pasture as a whole (Figure 4a). Qualitative comparison with the rainfall time series for the area around the ranch suggests that there may be a relationship between rainfall of the preceding month and this peak (Figure 4b). Precipitation and herbaceous vegetation differences, as well as other potential contributing factors to apparent greenness (e.g., differences in cloudiness and differences in timing of grazing between 1999 and other years), should be considered in any further evaluation of this peak and its potential relationship to rainfall.

Table 2. Two-tailed significance for t-tests comparing the NDVI means for nine pixels in Middle Snake
Ridge pasture between dates in 1999 (i.e. between "before", "during", and "after" grazing) and between
1999 and the long-term average (LTA) for the same compositing time frame. In both sets of comparisons,
only the differences with the middle (August 12) date are significant.

Table 3. Rescaled NDVI values for the Middle Snake Ridge pasture (average of 9 1-km^2 AVHRR pixels),
AVHRR pixel 16980 approximately corresponding to the TM subset area, and for the TM subset area
(average of 399 30-m^2 TM pixels; ETM+ images of 22 July and 23 August). TM NDVI are "rescaled" here
for the purpose of comparison by adding 1 and then multiplying by 100. Long-term averages are also provided
for the AVHRR datasets.

Figure 4. Time series for the AVHRR NDVI at pixel 16980, corresponding approximately to the location of
the TM subset in Middle Snake Ridge pasture. Timing of grazing in this pasture is indicated by the symbols.
The pasture-level average NDVI and the long-term average + 1 SD for pixel 16980 are also shown. Monthly
precipitation for the corresponding time frame at four weather stations around the V-bar-V Ranch is plotted
beneath (data: NCDC,2001).

Spatial Contrast

The direction and magnitudes of change are not uniform across the subset area. In general, there is greater canopy cover in the eastern half and less in the western half (Figure 5). This is apparent to varying degrees for the change images in each of the spectral indices (Figure 6), particularly for the two green vegetation images (NDVI and TM greenness) and, even prior to enhancement, for TM brightness. The change in TM brightness was the greatest and most statistically significant of the four indices. Previous analyses under the RangeView project involving correlation of percent cover data with TM spectral indices yielded highest correlations between percent soil cover and TM brightness (R^2=.67 to .75, P<.05). These and the current results, in conjunction with the spatial detail provided by the Ikonos imagery, suggest that exposure of soil may be the process underlying the changes that are apparent. Spatially corresponding decreases in green vegetation (Pearson's R for brightness with TM greenness = -0.53, and with TM NDVI = -0.265, P<.01) may be one factor contributing to the exposure of soil during the grazing period.

Figure 5. Ikonos 1-m panchromatic image of 26 April 2000 in the TM subset area (red outline).

Summary and Conclusions

Landsat ETM spectral indices have revealed significant changes in green vegetation, brightness, and wetness during a period of grazing in a pasture on the V-bar-V Ranch. The net change in green vegetation and wetness at this spatial resolution was negative, while change in brightness was positive. Corresponding changes in the 1-km resolution AVHRR NDVI at this location and for the pasture as a whole were positive, but not statistically significant. High-spatial resolution Ikonos panchromatic imagery provides a context for interpreting the changes observed as related to differences in canopy cover across the subset area. In particular, change in brightness appears to be related to changes in exposure of soil, due at least in part to decreases in the amount of green vegetation.

Several sources of environmental variability may underlay these changes, notably rainfall, phenology, and spatial differences in soil, herbaceous, and canopy cover. Comparison between the higher-temporal resolution AVHRR time series and monthly precipitation time series for the area surrounding the ranch suggest that rainfall variability may, at a larger spatial scale, play a role in the behavior of the NDVI.

At the same time, there may be several potential feedbacks from grazing that can influence the appearance of greenness and/or brightness of an image. These may include removal of non-photosynthetic vegetation (NPV, e.g., dry grass), consumption of green vegetation, compaction of vegetation, disturbance of soil, and differential use of space and/or vegetation resources by livestock. More detailed, ground-based knowledge of these processes is required to fully account for apparent spectral change, which could contribute to the use of remote sensing imagery as a more powerful management tool.

References

Chavez, P. S., jr. (1996) Image-based atmospheric corrections - Revisited and improved. Photogrammetric Engineering and Remote Sensing 62 (9): 1025-1036.

NCDC (2001) National Climatic Data Center - NNDC Climate Data Online. Monthly precipitation data for Arizona stations: 020670, 023828, 025635, 027708. Internet source (accessed August 2001): <http://cdo.ncdc.noaa.gov/plclimprod/plsql/poemain.poe>

Crist, E. P. (1985) A TM tasseled cap equivalent transformation for reflectance factor data. Remote Sensing of Environment 17: 301-306.

Eidenshink, J. C. (1992) The 1990 conterminous U.S. AVHRR data set. Photogrammetric Engineering and Remote Sensing 58 (6): 809-813.

Huang, C., B. Wylie, L. Yang, C. Homer, & G. Zylstra (2000) Derivation of a tasseled cap transformation based on Landsat 7 at-satellite reflectance. USGS EROS Data Center, Internet source: <http://landcover.usgs.gov/pdf/tasseled.pdf>

NASA (2001) Landsat 7 Science Data Users Handbook: Chapter 9 - Calibration Parameter File. Internet source (accessed Nov 2001): <http://ltpwww.gsfc.nasa.gov/IAS/handbook/handbook_htmls/chapter9/chapter9.html>