Characterizing the status, variability and temporal dynamics of ecological systems in the southwest

Profound changes are evident in our ecosystems: the urban footprint on the landscape is expanding rapidly; exotic plants are invading native systems; the polar ice caps are melting. In this time of rapid global change, it is important to recognize and monitor even subtle landscape changes, and necessary to separate patterns of normal variability from trends toward a new state of the ecological system.

The main objective of this research is to quantify and monitor ecosystem dynamics. To do this, we are characterizing the seasonal and interannual spectral variability of ecological systems as mapped in the USGS Southwest Regional Gap Analysis Program (SWReGAP; see http://www.gap.uidaho.edu/) using remotely sensed satellite data. Landscape dynamics are being characterized using Moderate Resolution Imaging Spectroradiometer (MODIS) Enhanced Vegetation Index (EVI) data. MODIS EVI satellite data were chosen because they capture the dynamics of vegetation distributions across the landscape (Huete et al., 2002). MODIS-EVI images are collected daily at 250-m resolution in the red and near-infrared bands, along with 34 additional bands collected at larger resolutions. The EVI is a spectral measure of the amount of vegetation present on the ground, calculated using the red, near-infrared, and blue bands Daily MODIS- EVI images are combined into 16-day image composites by retaining the maximum EVI value at each pixel designed to produce a cloud-free image of the photosynthetically-active vegetation on the surface. Each year is represented by 23 16-day composites.

We are focusing our study on 3 sites (Mojave National Preserve, Organ Pipe Cactus National Monument, Cobb Peaks Area) and 5 vegetation types. The vegetation types are from the ecological systems defined for Arizona by NatureServe, and include: Rocky Mountain Ponderosa Pine Woodland, Colorado Plateau Pinyon Juniper Woodland, Sonora-Mojave Creosotebush-White Bursage Desert Scrub, Sonoran Paloverde-Mixed Cacti Desert Scrub, and North American Warm Desert Riparian Woodland, Shrubland, and Bosque. Within these areas and vegetation types, we are:

• Extracting measures of landscape temporal dynamics from remotely sensed data.
• Evaluating the seasonal and interannual dynamics for the mapped ecological systems.
• Characterizing the variability of the ecological systems, both temporally and spatially.
• Exploring various tools that permit visualization of stable and changing landscapes.

The visualization tools and metrics developed in this study will describe the potential dynamic range of seasonal and interannual spectral characteristics for the mapped ecological systems. Such information will provide valuable layers of additional information to the new SWReGAP land cover maps from which subsequent changes may be recognized and evaluated. Informed management of our ecosystems requires knowledge of the inherent seasonal and interannual variability of the ecosystems as well as their expected evolution, both as a result of expected change under stable environmental conditions and as a result of directional changes in environmental conditions and external forcings.

In this time of apparent global change, it is important to recognize and monitor even subtle landscape changes, and necessary to separate trends toward a new state of the ecological system from patterns of normal variability. Effective management of our natural resources is enhanced by the ability to identify these trends as early as possible, especially if the observed changes can be attributed to anthropogenic pressures on the land that can be addressed and modified (e.g., grazing and recreational use).

We are evaluating the dynamics of the ecological systems mapped by the Southwest Regional Gap Analysis Project (SWreGAP) (Fig. 2). The SWreGAP project was initiated in 1999 as a multi-institutional cooperative effort to map and assess biodiversity for a five-state region (AZ, CO, NV, NM, UT) comprising approximately 560,000 square miles in the southwestern U.S. A key task was the development of a seamless landcover map for the region and the collection of other pertinent bio-physical spatial data. Through coordination from the USGS’s Gap Analysis Program (GAP) and the collaborative efforts of participating state institutions, a seamless landcover product was completed in September 2004. These data and related datasets are made available to the public with ‘provisional’ status by the SWReGAP consortium of institutions responsible for their development.

Figures 3 and 4: ECOLOGICAL SYSTEM PHENOLOGY For each of the five ecological system types selected for focused study, we extracted the MODIS-EVI data for 2001-5. The percent of each vegetation type (mapped at 30m resolution) within each MODIS cell was calculated and we selected only the MODIS cells that contained 100% of types 1 to 4 (forest and desert types) or 50% of type 5 (Riparian). A smaller percentage was chosen for riparian vegetation since it typically occurs along narrow features, and exhibits a high contrast with the surrounding desert. Within each masked ecological system type, we stratified the landscape into 10 sub-types based on their distinctive phenologies using the unsupervised classification routine in Erdas Imagine: 92 EVI composites (2001 to 2004) were input as a single multi-band image and an output of 10 classes was specified. Data for 2005 will be added to these profiles. Figure 3 shows the classified MODIS images for the five ecological system types and Figure 4 shows the average phenologic profiles for the 10 classes created within each system. Each image pixel is assigned to one of 10 classes based on its particular EVI phenology and each class in the map (Fig. 3) has the average phenology shown in the graph (Fig. 4).

The maps reveal that in all ecological systems, there is a strong gradient based on elevation. This is not surprising since vegetation type, precipitation, and temperature are strongly correlated with elevation in Arizona. The large magenta feature in the Type 1: Gap 34 map (Fig. 3) of Ponderosa Pine Woodland is the Rodeo-Chediski fire scar; Figure 4 shows the phenology of this feature, which displays a sharp decline in EVI in the summer of 2002, marking the transition from unburned to burned forest. The higher elevation types (Gap 34 and 36) typically have strong greenness in the winter with a broader, secondary peak in the summer. The sharp, January spikes evident especially in Gap 36 (Pinyon – Juniper) are suspicious and may represent erroneous EVI values due to an incorrect aerosol correction over bright snow cover (Reed, pers. comm); we are processing the MODIS-NDVI data to evaluate this phenomenon. The desert types (Gap 57 and 60) show a pulse of greenness in the spring except during 2002, which was an extremely dry year. Some of the sub-types also show a subtle pulse of greenness in the fall, likely due to monsoonal moisture. The phenologies of the Riparian types are quite distinctive. Classes 1, 2, and 3 show a strong, unimodal peak in the summer. These are also the most productive sub-types as evidenced in Figure 4 by profiles with the highest EVI values. It is likely that these sub-types contain agriculture; the pixels were selected to be at least 50% riparian and irrigated agriculture is often located adjacent to riparian areas. Several of the other classes (e.g., class 7) show a bimodal profile, with a strong post-monsoon peak in the fall and a subtler spring peak. These data have potential to delineate riparian areas most critical for habitat, depending on the species needs. Other phenological differences likely reflect differing amounts and species composition of vegetation contained in each SWreGAP ecological system type, as well as phenological differences due to environmental factors, such as landscape pose and substrate. Such profiles have potential to highlight areas with abundant annuals and invasives.

Figure 5: LANDSCAPE PHENOLOGY Measures of landscape temporal dynamics were derived from the MODIS-EVI data for the years 2001 through 2005. The EVI data for an individual year consists of 23 images; the temporal data were evaluated for each of the five years individually in order to capture the annual phenological cycle of vegetation green-up and senescence, and to allow comparisons between the years. A variety of measures that capture an expression of this phenological signal were derived using basic statistics as well as Fourier transforms. In Fourier analysis, the ordered values at each pixel, from layer 1 through layer 23, are input as a vector and the Fourier transform analyzes the basic waveform of the entire annual EVI profile at that location. The measures extracted by the Fourier transform are the additive term (i.e., the best fit of a flat line to the data) and the first frequency magnitude and phase (describing the best fit of a single sine wave to the data vector). These measures can be calibrated to represent total EVI (‘Prod’), range of EVI (‘Mag’), and season of high EVI activity (‘Phase’), respectively (Wallace 2002, Jakubauscas and Legate 2001, Moody and Johnson 2001). Figure 5 shows these measures for 2002 (a dry year, with Rodeo-Chediski fire scar) and 2005 (a recent, wet year). Additional higher frequency components were also extracted, but are not shown here.

The two left Productivity images reveal abundant vegetation along the Mogollan Rim, the north rim of the Grand Canyon (A), and the riparian zones (e.g., B). The phase (middle images) reveals a well-ordered peak of greenness in 2005, with the deserts of SW Arizona greening up earlier in the year and the higher elevation Colorado Plateau greening later. SE Arizona also greens later in the year apparently due to monsoonal rain. In contrast, the phase data for 2002 is quite patchy, reflecting the arbitrary phases produced when fitting a curve to a relatively flat profile. The two right Magnitude images show the contrast between a drier 2002 and wetter 2005. The 2002 image is typically darker than 2005: compare the low elevation deserts at points C and D, as well as the riparian features at E and F. The high magnitude at G is the Rodeo-Chediski fire scar, where dense Ponderosa Pine forest burned in June 2002

Figure 6 summarizes the Fourier metrics for the Sonoran Palo Verde – Mixed Cactii Desert Scrub Ecological System (Gap 57) for 2002 and 2005. In this graph, the annual Fourier metrics of each MODIS pixel of Gap type 57 is plotted as a point with the first frequency phase on the horizontal axis and the Fourier magnitude on the vertical axis. The colors reflect the Fourier productivity, so that high productivity (high vegetation amount) is dark green and low is brown. The dramatic difference in the vegetation behavior between 2002 and 2005 is readily apparent here. 2002 is dominated by low vegetation amounts and very subtle greenness dynamics; only a few pixels show small peaks of greenness in May, October and November (the scattered green dots in July are likely agriculture that are included due to land use change since the SWreGAP mapping or image misregistration). In contrast, the 2005 graph shows the ecological system has several pulses of green-up, with a strong, broad peak in the spring, a smaller late-spring peak, and a distinct late summer monsoon peak. The 2005 landscape is much more dynamic and much greener than 2002