To be most effective, these approaches demand landscape-scale perspectives on the mechanisms underlying invasive species spread and persistence over time . However, most studies on the effects of management on invasive species are conducted at the much smaller scale of 1-m2 plots. Such small plots can assess local responses to management but provide only a limited picture of landscape heterogeneity. Moreover, unless embedded within larger landscape assessments, small plots cannot readily quantify expansion, contraction, and persistence of invaded patches over time. At present, mapping with remote sensing generally requires that the invaders differ from the resident community in specific ways, such as in plant chemistry , texture/morphology, phenology, or canopy level. Thus, while remote sensing has been extremely effective in detecting changes in functional groups of plants , it is more difficult to detect invasion of a species that is similar in functional type to the resident community it is invading . Detecting the invasion of grasses into grassland communities is particularly challenging, with phenological differences so far proving to be the most helpful identifying features. However, when multiple images are required for phenological detection methods, drying rack for weed available low- or nocost public data has often come at a coarse spatial scale that fails to capture the smaller patch dynamics relevant to management . The use of data from commercial satellites in precision agriculture illustrates the power of finer-scale imagery , but such commercial satellite imagery has typically not been affordable for rangeland managers.

Even with the limitations of coarse-scale imagery, however, some studies have been able to use differences in phenology to map invasions, based on differences in season of green-up and timing of green-up after precipitation, or on seasonal variation in NDVI. In California’s semi-arid grasslands, two invasive weedy grasses have become particularly problematic, and are a high priority for mapping and control efforts: Elymus caput-medusae Nevski and Aegilops triuncialis L.. These annual grasses were introduced to California in the late 19th century from Eurasia and are now established throughout the Western United States. Both species produce unpalatable forage that is avoided by livestock, particularly as the plants mature, and generate a thick, mulching layer of litter that typically persists well into the next growing season . In California, broad-scale detection of these invaders is challenging because these annual grasses are invading a community already dominated by annual grass species, including Avena and Bromus spp. . In addition, for most of the growing season, the phenology of the invasive plants overlaps with the desirable annual forage grasses. Both groups germinate with fall rains in October and November and then grow throughout the rainy winter season into the spring months. However, the invasive weedy grasses differ notably from the forage grasses in their end-of-season phenology. The forage grasses typically reach peak greenness in March or early April and then senesce in late April and May , while the weedy species exhibit an extended late-season growing period that ends in late May or June. It is during late spring and early summer, when the invasive weedy grasses are green but the forage grasses are golden and senesced, that weed patches may be most easily identified on the ground by field observers . Our study focused on two key questions: How well can fine-grain phenologically-timed aerial imagery detect the invasion of medusa head and goat grass into naturalized California annual grasslands over time? and Using this method, how do the abundance and persistence of forage and weed-dominated patches vary in response to grassland management?

This study was conducted in partnership with private landowners and conservation practitioners, to assess the effectiveness of various weed mapping approaches in quantifying landscape-level impacts of land management actions on invasion in California grasslands .Our study examined invasive weedy grass distribution within a 6.8-km2 region of semi-arid grasslands on rolling hills on the west side of the Sacramento Valley, CA, USA. The study area included four different management units on three privately-owned ranches. The landowners of these private properties gave permission to conduct this work. These units had experienced different grazing intensities over recent years, ranging from none to intensive rotational grazing by sheep, goats, and cattle. The two weedy grasses medusa head and goat grass were well-established across all properties, alongside annual forage grasses . Much of the landscape matrix was thus a heterogeneous mixture of weedy and forage grasses, out of which emerged near-monospecific patches dominated by either weeds or forage. Our primary objective was to map the distribution of the strongly weed-dominated patches, which provide little forage or conservation value. At the study location, the dominant soil types are fine smectitic thermic Andic Haploxererts; fine, mixed, active thermic Typic Palexeralfs; and fine-silty mixed, superactive thermic Typic Haploxeralfs. The climate is Mediterranean, with a cool and rainy growing season that typically begins in September and extends into May when the summer drought begins. Peak precipitation typically occurs between December and February but patterns of precipitation are quite variable. Almost no precipitation falls during summer, when mean maximum temperatures can exceed 37 ˚C . The two growing seasons we studied differed both in total precipitation and its temporal distribution. In growing year 2008, total annual precipitation was close to average , with rains heaviest in January and February and very little falling thereafter . In contrast, total precipitation in growing year 2009 was only 81% of average , with the largest rain events occurring during the shoulder seasons and little in mid-winter; small rain events occurred later into late spring and early summer than in 2008 .

Our aim was to identify a robust method for mapping the distribution of weed-dominated patches that would work well even across years of different precipitation and then to use this approach to evaluate weed patch persistence or change across the four management units in our study site. The first challenge was to discern patches dominated by annual weedy grasses within the existing annual grassland, which is morphologically similar. To do this, we first characterized the phenological signature of the weeds based on subtle seasonal changes in their canopy greenness that could be discerned in contrast to the forage grasses or mixed communities of forage and weeds in which the weeds were not dominant. We considered the two weedy species as a group and did not attempt to distinguish between them. To characterize the weed group’s phenological signature, we evaluated how its greenness changed from the period of peak landscape greenness to the end of the growing season, and compared its signature to that of forage-dominated patches. To assess vegetation greenness, we used low-cost digitized color infra-red aerial photography familiar to many range managers and for which spatial resolution was fine enough to resolve small weed patches in this system. From this imagery, we derived values comparable to the Normalized Difference Vegetation Index , a classic index that identifies green vegetation. We then tested different combinations of imagery and classification approaches against ground truth data to identify the most robust mapping method, and then used this method to look in detail at weed patch distribution across the study site, as detailed below.Aerial imagery was acquired from a fixed wing airplane by Pacific Aerial Surveys between the hours of 12:30 and 1:30 pm, using Kodak Aerochrome III Infrared Film 1443 with a minus-blue filter at a scale of 1:34,200 or 1:35,000 . The imagery was acquired with a 22.9 cm x 22.9 cm negative format from a mapping camera with a focal length of 153 mm/6 inches. Flight lines were arranged north–south with less than 5% crab and less than 2 degrees tip and tilt. For each date, pipp mobile storage systems we used a single image that encompassed the entire study area, so the need for image mosaicking was avoided. The images were acquired on cloud-free days at a relatively low altitude, so atmospheric corrections were also not necessary. After processing, the film was scanned without color adjustment on a photogrammetric scanner to create a digital image with approximately 0.39–0.45-meter resolution. Spring images were taken on March 10 of 2008 and 2009, during the period of peak greenness when the canopy of forage grass species is typically greener than that of the invasive weeds . End-of-growing season imagery was acquired when the weed species were still green but forage grasses had senesced as judged from the ground . This date varied across years due to weather-driven variation in phenology. In 2008, the May image was acquired on May 13, and in 2009, images were acquired on both May 18 and 26 . Aerial images were orthorectified in the Leica Photogrammetry Suite in ERDAS Imagine 9.3 using recent camera calibration reports from the United States Geological Survey Optical Science Lab. Ground control points for orthorectification were georeferenced in the field with a Trimble GeoXH using a Trimble Beacon Receiver communicating with a U.S.

Coast Guard beacon and an antenna on a 2-m range pole. Point data were georeferenced in UTM10 with a WGS84 datum and then post-processed and adjusted for atmospheric conditions with reference data from the nearest National Geodetic Survey Continuously Operating Reference Station for greater positional precision. The navigational accuracy of point locations in the field and post-processed points was approximately 0.30 m, which was confirmed by benchmark tests. For effective orthorectification, a digital elevation model is also necessary. At the time of the project, the only existing DEM for the project area was the National Elevation Dataset produced by the USGS. This DEM had a spatial resolution of 30 m x 30 m. This meant that an elevation value was available every 30 m across the landscape. Because the project area contained considerable landscape relief, it was determined that the NED DEM would not allow for production of accurate orthoimagery needed for concurrent analysis of several image datasets. To ensure the accuracy of the final orthoimagery, it was necessary to create a new, more precise, DEM. The source for the elevation data was the georeferenced Digital Raster Graphic dataset created by the USGS from the USGS7.5-minute quadrangle maps . Individual contours were digitized in ESRI ArcMap 9.2 as polyline features. The elevation for each contour was entered into the database table as an attribute. After digitization was complete, a 2-meter DEM was created by interpolating the elevation values of the contours. The interpolation was completed using the Topo to Raster tool in ESRI ArcGIS Toolbox 9.2. The resulting 2-meter DEM represented a large increase in elevation precision over the National Elevation Dataset and was used in image orthorectification to correct for changes in terrain.The Normalized Difference Vegetation Index is a simple, well-tested metric derived from red and near-infrared radiances; it is calculated as . Green vegetation typically shows higher values than senesced or non-living materials. To minimize topographic and seasonal differences in illumination, we created an NDVI-like image from the red and near infrared bands of each digitized color infrared image, using ENVI 4.7 . The spectral properties of film differ from those of calibrated satellite instruments, but film-based NDVI estimates provide valuable utility. Historically, color infrared film was designed so that the red and infraredsaturation intensities appeared relatively similar to each other in order to produce an aesthetically pleasing image; as a result, NDVI values captured on film are typically lower than those captured by satellite sensors, in which the infrared saturation intensity is permitted to be greater. To evaluate different methods of capturing phenological signatures, we then used Modeler in ERDAS Imagine 9.3 to produce NDVI difference images for each year as ΔNDVI = NDVIMarch − NDVIMay . Mask. We created a 1-m resolution mask to remove from analysis the areas that did not include vegetation of interest . We digitized masked objects in ESRI ArcMap 9.3 and exported them to ERDAS Imagine, where a 5×5 neighborhood filter was used to remove spurious data and simplify the dataset. This image was then converted to vector polygons, compared to orthoimagery, and edited as appropriate. The polygon features were then simplified in ArcMap, converted to raster, and formatted for use in ENVI. Image layer stack for classification. To facilitate image classification, we created a layer stack in ENVI 4.7 using all of the NDVI and NDVI difference imagery, as well as the mask.