< Back  ||  Home > Reports > Vegetation Dynamics

Temporal Dynamics of the NDVI for Vegetation Types on the V-V Ranch

Introduction

One of the goals of RangeView at the V-bar-V Ranch has been to develop a means for predicting range vegetation in the context of climatic variability. This requires, first, developing an understanding of the annual (phenological or seasonal) dynamics of vegetation types and then evaluating the responses of different vegetation types to various potential sources of interannual variability in the NDVI.

Study Area, Data Sets, and Methods

This analysis utilizes the 11-year AVHRR NDVI dataset for an area including and immediately surrounding the V-bar-V Ranch (Figure 1; refer to Eidenshink, 1992, for description of the processing methods used by the USGS to create this dataset). Temporally and spatially aggregated NDVI data were derived from this original dataset and used in analyses of annual and interannual vegetation change (see Figure 2).

Figure 1. Vegetation and topography of the V-bar-V Ranch and vicinity treated in this analysis. Pasture
boundaries, 1-km^2 AVHRR pixels, and TES vegetation types are shown. Pastures >10 km^2 (i.e. those including
10 or parts of 10 pixels) are labeled. Vertical exaggeration in the elevation profile ~5x; profile derived from a
30-m resolution DEM. Note that linear distance scales are slightly different between the two figures.

Figure 2. Procedures for temporal and spatial aggregation of the original 11-year dataset of biweekly composite
NDVI images. Intermediate and final datasets referred to in the text are shown in bold: LTA NDVI, COV NDVI,
TES NDVI, TES NDVIann, LTA NDVItes, COV NDVItes, and TES NDVImax.

Long-term average NDVI (LTA NDVI) and interannual coefficients of variation (standard deviation as a fraction of the mean, COV NDVI) are computed for each date and pixel over the 11-year data set. These values are then averaged for the set of pixels comprising each major TES vegetation unit (see Miller et al., 1995) that occurs on the V-bar-V Ranch (TES 500: N=395, TES 400: N=519, TES 200+300: N=194), yielding LTA NDVItes and COV NDVItes. Averages for each date in each year are also computed for the three TES units, giving TES NDVI. The maximum value for each date in the 11-year time frame is extracted for each TES unit to yield TES NDVImax. The annual average NDVI (TES NDVIann), used in comparisons with ENSO phase, is computed as the average of 26 dates each year, then averaged for the pixels in each TES unit. All computations are performed using simple spreadsheet functions in Microsoft Excel.

The average June-November Southern Oscillation Index (SOI) is used as a proxy for ENSO phase for the ensuing year (e.g., June-November 1990 indicates the ENSO phase in the winter-spring of 1990-1991). Correlation coefficients are derived by simple linear regression using the Data Analysis extension in Microsoft Excel. Correlations are considered significant when R^2>.40 at p=.05.

Results and Discussion

Long-term average NDVI: I. Time series for TES units

Each of the three major vegetation types on the V-bar-V Ranch exhibits a bimodal greenness pattern in the LTA NDVItes (Figure 3a). Additionally, the three curves are largely in-phase with one another with regard to timing of minima and maxima and the rate of increase and decrease in the LTA NDVItes between these extremes (Figure 3b). The differences in magnitude of the LTA NDVItes between vegetation types (i.e. across space) are structured by an underlying elevational gradient across the ranch (previous analyses of this dataset indicate consistent correlations between NDVI and elevation ranging from R^2=.71 to .86).

The first peak in the annual time series follows the winter rainy season and lasts from about early May to early July, with the highest LTA NDVItes values occurring in mid- to late June. The LTA NDVItes declines to its mid-summer minumum in late July and remains low until late August. The second mode occurs after the summer monsoon, lasting from about mid-September to mid-October, and is slightly higher and narrower than the late spring peak. The autumn maximum LTA NDVItes falls around the first of October, just about the time of the first killing frost at the latitude of the V-bar-V Ranch, and declines sharply thereafter. The lowest LTA NDVItes values of the year occur from late December to mid-March.

Figure 3. a) Time series of long-term average AVHRR-NDVI (LTA NDVItes) based on an 11-year data set
(1989-1999), showing the bimodality of NDVI for the three major vegetation types on the V-bar-V Ranch.
b) First derivative (computed as the difference in LTA NDVItes between successive dates divided by
14 days) for the LTA NDVItes time series, illustrating the relative synchrony of change and similarity in
rates of change in LTA NDVItes between the three vegetation types. c) Interannual coefficient of variation
(COV NDVItes; multiply by 100 for percent), showing the pattern of increased variability in the winter and
summer storm seasons. Tic marks are at 14-day intervals corresponding to the approximate end dates of
the NDVI compositing periods.

The COV NDVItes series display marked intra-annual temporal variability and relatively little difference spatially between vegetation types (Figure 3c). Values during the winter (December to March) are higher than at any other season for each of the three vegetation types, ranging from about 4% to 6% of the LTA NDVI. This pattern reflects the influence of clouds and snow which can cause NDVI values to plummet sharply, sometimes to nearly zero (100 rescaled) when vegetation is completely obscured during the biweekly compositing period. In this dataset, this phenomenon is never apparent for more than one biweekly compositing period in a row. Interannual variability in the NDVI arises when the timing and duration of individual, major winter storm events is inconsistent between years. At the same time, the cloudiness and snowiness of the winter season may vary from year to year with larger-scale climatic phenomena such as the ENSO cycle.

Variability remains high at upper elevations (ponderosa pine vegetation) through April, possibly reflecting the influence of late-season storms and snow events. Variability at lower elevations (pinyon-juniper and desertscrub vegetation) declines rapidly through March and April, and for all areas of the ranch it is at a minimum (about 2% of the LTA NDVI) from late April to late May. This time frame, the "arid foresummer", is the driest of the year for most of Arizona and, probably related to static high pressure, experiences the least cloudiness of any similar time frame.

Variability increases again in June and remains at about 4% of the LTA NDVI through the summer. The summer pattern of variability probably arises from differences in the timing of the inception of the monsoon and in the intensity of the monsoon (see Douglas et al., 1993; Adams & Comrie, 1997) which contributes to cloud formation over the Mogollon Rim and, consequently, over the V-bar-V Ranch. A second minimum in early September is followed by a period of variability between 3% and 3.5% of the LTA NDVI lasting from late September to December, after which there is a marked increase in variability during the winter season. The period of moderate variability in the autumn occurs after most herbaceous growth has stopped and likely relates to cloud effects on the NDVI. Variation in the location of high- and low-pressure cells (i.e. changes in atmospheric circulation patterns) between years may be a contributing factor during this period of time and in the ensuing winter months.

Long-term average NDVI: II. Correlation of annual average NDVI with ENSO indicators

There is a weak but statistically insignificant negative relationship between TES NDVIann and the Southern Oscillation Index (SOI) - an indicator of ENSO phase - for all vegetation types (R^2 at p=.05: TES 500 = -.21, TES 400 = -.24, TES 300 = -.21). Moderate, negative rank correlation coefficients (r(r)) (TES 500 = -.52, TES 400 = -.45, TES 300 = -.45) are significant at between the 80% and 90% confidence levels (p=.10-.20). These results suggest that annual average greenness may, to some extent, mirror the well-known negative relationship between precipitation and SOI where negative SOI values generally result in higher-than-average winter precipitation and positive SOI values generally result in lower-than-average winter precipitation in the southwestern U.S (Redmond & Koch, 1991). Figure 4 illustrates the interannual relationships between the SOI and the TES NDVIann for each of the three vegetation types on the V-bar-V Ranch for the 11-year time frame of this analysis.

Figure 4. Southern Oscillation Index (SOI) and TES NDVIann for the three major Terrestrial Ecosystem Site
(TES) units on the V-bar-V Ranch. SOI of the preceding June-November is plotted for the succeeding year
(symbols are colored to indicate ENSO phase). Note that the SOI scale is inverted. Climate data: WRCC, 2001.

Visual inspection of scatterplots of these data (Figure 5a) reveals a pattern of within-year difference in the TES NDVIann between vegetation types that is consistent across most climate years. For all years but 1992-1994 - including two El Niño years (1995 and 1998), two La Niña years (1989 and 1999), and four non-ENSO years -, the difference between TES 300 and TES 400 is about 2.5 rescaled NDVI units, and the difference between TES 400 and TES 500 is about 5 rescaled NDVI units. These differences compare to the year-long differences between LTA NDVItes curves illustrated in Figure 3a. In each of the El Niño years 1992-1994, however, these differences between vegetation types are at least doubled, and it is the pinyon/juniper (TES 400) and ponderosa pine (TES 500) TES NDVIann values that are markedly higher than in other years.

Figure 5. Relationship between TES NDVIann and SOI as an indicator of ENSO phase: (a) scatterplot, (b) residuals plot.
Symbols are keyed to vegetation type: TES 500 (ponderosa pine/oak), TES 400 (pinyon/juniper), TES 300 (desertscrub).

The residuals plot for the regression of TES NDVIann with SOI for each TES unit (Figure 5b) illustrates a difference in predictability of greenness between El Niño years and other years. There is also variability between TES units within the same or similar climate year: in El Niño years, particularly, greenness is relatively more predictable at lower elevations, occupied by desertscrub, and increasingly less predictable at higher elevations occupied by ponderosa pine forest. The direction of deviation from predicted NDVI also varies inconsistently between similar climate years - for both La Niña and El Niño years, as well as for non-ENSO years, positive and negative deviations from predicted values obtain.

The influence of the summer monsoon - which, in north-central Arizona contributes 30-40% of annual precipitation (Douglas et al., 1993) - on the NDVI during this time frame has not been evaluated here explicitly, however this important circulation pattern should not be neglected in the interpretation of annual average NDVI. While there is no consistent relationship between the ENSO cycle and the summer monsoon in Arizona (Adams & Comrie, 1997), some authors suggest that there may be teleconnections between sea surface temperatures (SSTs) in the North Pacific and atmospheric circulation related to the summer monsoon over Arizona, and that these relationships to North Pacific SSTs may not be independent of the ENSO cycle (see Carleton et al., 1990; Higgins & Shi, 2000).

>view plots showing Pasture-level Average NDVI for ENSO Years

Long-term maximum value NDVI: Time series and relationships to ENSO phases

In light of the pattern of interannual variability encountered in the NDVI, particularly as the coefficient of variation experiences maxima during the winter and summer storm seasons, an attempt was made using the long-term maximum average NDVI (TES NDVImax) to describe the phenological behavior of the vegetation types on the V-bar-V Ranch. The premise underlying this approach is similar to that applied in the production of biweekly composites, namely to minimize or eliminate the influence of clouds and/or snow on apparent greenness. In proceeding with this discussion, and because the TES-level average NDVI (TES NDVI) values represent an average NDVI for of all of the pixels in each vegetation unit, it is presumed that, for each biweekly compositing date, the maximum NDVI value represents the least obscured conditions at a landscape scale in the 11-year time frame. The TES NDVImax values for each vegetation type and compositing period in the 11-year dataset are plotted in Figure 6 along with the long-term average NDVI (LTA NDVI) time series that were introduced above.

Figure 6. Time series of TES NDVImax and LTA NDVI for the 11-year data set (1989-1999). Tic marks
are at 14-day intervals corresponding to the approximate end dates of the NDVI compositing periods.

Immediately conspicuous in the TES NDVImax are the values for the winter (mid-December through early April). Where the LTA NDVI declines to a distinct annual minimum in January-February, TES NDVImax values remain fairly constant and markedly higher (by about 10 rescaled NDVI units) than the long-term averages. The bimodal greenness pattern is preserved in the TES NDVImax data, although the peaks in June and October are broader than in the LTA NDVI curves. It is tempting to hypothesize that the winter portions of the TES NDVImax curves are representative of the perennial, evergreen vegetation present in each vegetation type, and that the April to November portions represent the herbaceous and/or deciduous components of each vegetation type. In some way, perhaps, TES NDVImax represents the greatest potential photosynthetic activity of each vegetation type and provides an indication of the relative amounts of green vegetation present in each of these units.

The source years for maximum values are summarized in Table 1. Two years - 1998 and 1999, representing the two extremes of the ENSO cycle (refer to Figure 4) - are the predominant contributors of maximum average values to the TES NDVImax time series. When the TES NDVImax time series curves are regarded with respect to the ENSO phase of the source year for each compositing date, a seasonal pattern emerges (Figure 7).

Table 1. Number of maximum NDVI values contributed by each year in the dataset. Note that the
total for each vegetation type may exceed 26 (the number of biweekly composites in each year) due
to ties between years in the maximum NDVI. ENSO phase (based on SOI) is also indicated.
Climate data: WRCC, 2001.

Source Year	ENSO Phase	TES 200+300 (desertscrub)	TES 400 (pinyon/juniper)	TES 500 (ponderosa pine)
1989	La Niña	0	0	0
1990	0	0	0	1
1991	0	0	0	0
1992	El Niño	0	0	0
1993	El Niño	0	0	4
1994	El Niño	0	0	0
1995	El Niño	1	1	1
1996	0	0	0	0
1997	0	0	1	3
1998	El Niño	13	15	11
1999	La Niña	15	13	13

 

Figure 7. Long-term average and long-term maximum NDVI for the major vegetation types on the V-bar-V Ranch.
Colored symbols indicate the ENSO phase of the year from which the maximum values derive. Climate data: WRCC, 2001.
Tic marks are at 14-day intervals corresponding to the approximate end dates of the NDVI compositing periods.

In the winter, from about mid-December through late March (or even late April), the TES NDVImax occurs during the La Niña phase. Since it is under these SOI conditions that the southwest experiences lower-than-normal storm activity, and thus less cloud cover, this segment of the time series may be reflecting less interference by clouds rather than, or at least in conjunction with, a truly elevated NDVI. In the late spring through early summer (about April through mid-July), maximum values are contributed predominantly by El Niño years. This time frame receives relatively little precipitation, even during an El Niño year, so is relatively cloud-free in any year and, indeed, varies little in NDVI between years (refer to Figure 3c). That the maximum values for spring "green-up" occur just after El Niño winters suggests that, perhaps, there is a contribution to spring germination, and/or to leafing-out of perennials, and/or to soil moisture by enhanced winter rains and/or snow that results in greater vegetation cover and persistence through the foresummer.

The TES NDVImax pattern through the summer, during the monsoon (late July through late August), continues to derive from El Niño years, although the influence of ENSO phase on the summer monsoon is inconsistent. Once again, if there is a connection, it may be through soil moisture and/or enhanced early-season growth that contributes to greater apparent greenness (presumably reflecting the presence of more vegetation) later in the year. The autumn segment of the time series (about September through December) appears to be influenced by ENSO years, but inconsistently with regard to which or which combination of ENSO phases. The contribution of the summer monsoon may, in fact, be more important to the NDVI during the late summer/autumn time frame, and apparent relationships to ENSO phase may be merely superficial.

Summary and Conclusions

Both annual and interannual variability in the NDVI are influenced by (or, more pointedly, are overprinted upon) landscape-scale transitions in vegetation type - from desertscrub to pinyon/juniper woodland, to ponderosa pine/oak forest - across a strong elevational gradient on the V-bar-V Ranch. The annual pattern of the three major vegetation types is very similar in terms of the timing, rate, and magnitude of major NDVI fluctuations in the late spring/early summer and late summer/early fall. Interannual variability is greatest in the winter and summer for all three vegetation types, and is the greatest at higher elevations in the winter. This pattern likely reflects the influence of clouds and/or snow, particularly differences in occurrence, duration, and timing of these events between years.

There appears to be little, and certainly no consistent, relationship between annual average NDVI (TES NDVIann) and the ENSO indicator, the Southern Oscillation Index (SOI). The annual average values are influenced by several potentially confounding factors, notably the effects of clouds and snow: in El Niño winters - which tend to be wetter - greater vegetation might be expected, but the landscape is more often obscured and NDVI values are much depressed; in La Niña winters - which tend to experience fewer storms and less precipitation - less vegetation might be expected, but the landscape is less obscured and NDVI values appear to be relatively elevated. Without more detailed knowledge of actual precipitation and/or atmospheric conditions during these years in the study area, more substantive interpretations really cannot be made. At the same time, the contribution of summer monsoon precipitation to summer and autumn vegetation and the annual average NDVI may have no direct relation to the ENSO cycle and, if it is varying independently or even inversely, introduces additional uncertainty in searching for a relationship with the SOI.

Despite the difficulties associated with relating the spatio-temporally aggregated TES NDVIann to ENSO phase, there does appear to be a seasonal relationship in the maximum average NDVI time series to ENSO phase: maximum average NDVI values in the winter and early spring months occur during La Niña conditions while, in the late spring and early summer, maximum NDVI values occur during El Niño conditions (i.e. immediately following a winter during which El Niño conditions prevailed). While the winter pattern is certainly due, at least in part, to landscape visibility as suggested above, the summer pattern cannot be attributed to this as the April-June time frame is the least cloudy in any year. Rather, it may be that favorable winters contribute immediately to spring growth and/or to persistence through the foresummer due to stored soil moisture from the winter.

References

Adams, D. K. & A. C. Comrie (1997) The North American monsoon. Bulletin of the American Meteorological Society 78: 2197-2213.

Carleton, A. M., D. A. Carpenter, & P. J. Weser (1990) Mechanisms of interannual variability of the southwest United States summer rainfall maximum. Journal of Climate 3: 999-1015.

Douglas, M. W., R. A. Maddox, K. Howard, & S. Reyes (1993) The Mexican monsoon. Journal of Climate 6: 1665-1677.

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

Higgins, R. W. & W. Shi (2000) Dominant Factors Responsible for Interannual Variability of the Summer Monsoon in the Southwestern United States. Journal of Climate 13: 759-775.

Miller, G., N. Ambos, P. Boness, D. Reyher, G. Robertson, K. Scalzone, R. Steinke, & T. Subirge (1995) Terrestrial Ecosystem Survey of the Coconino National Forest, USDA Forest Service Southwestern Region.

Redmond, K. T. & R. W. Koch (1991) Surface climate and streamflow variability in the western United States and their relationship to large-scale circulation indices. Water Resources Research 27: 2381-2399.

WRCC (2001) Western Regional Climate Center, Classification of El Niño and La Niña Winters <http://www.wrcc.dri.edu/enso/ensodef.html>.