Techniques to measure and map riparian habitat and determine landcover change in the delta in 2002

Measuring and mapping riparian habitat

   "Preliminary research (Nagler et al., 2001c, submitted) established a strong correlation between percent vegetation cover measured on the ground and the normalized difference vegetation index (NDVI) of Red/NIR band images taken from a low-level(150 m) airplane survey of the flood plain (r2 = 0.84). Further, the major plant associations (groundcover species, saltcedar-arrowweed, emergent plants and native trees) contributing to different habitat values could be visually differentiated on the images. Comparison of NDVI values for common landscape features (water, soil, vegetation) for aerial images and a TM of the same scene taken near the time of the flight gave nearly identical values, showing that results can be accurately scaled from ground, to aerial and then satellite images (Zamora, Nagler et al. 2001, in press).

   We will build on this data to develop vegetation maps and habitat maps at four seasons of the year, corresponding to summer (June 21), fall (Sept. 21), winter (Dec. 21) and spring (March 21), each year for the three years of the study. We will use the MQUALS sensor package (Huete et al., 1999) to acquire overlapping, 1,000 m aerial images of the floodplain, using multi-band (blue, red and NIR) and visible-band digital cameras. The MQUALS system incorporates paired, ground and airborne sensors such that voltages can be converted into actual reflectance values for radiometric measurements. We will use visual interpretation of the images to construct a vegetation base map, dividing each image (representing ca. 100 ha) into 100, 1-ha mapping units, categorizing each unit as soil, water or vegetation, and if vegetation as groundcover, shrub, native tree or emergent aquatic species. Using NDVI and other vegetation indices and cluster analyses, we will develop methods to accurately map units based on spectral properties, then we will scale up to larger-scale mapping units for TM images of the same scenes. Ground truthing will consist of locating representative sample areas on the ground and quantifying vegetation (by type), soil and water using line-intercept methods. Other biophysical measurements on the ground will include: global and local Leaf Area Index by LiCor2000 calibrated by physical measurement of LAI for subsamples of each species and, reflectivity spectra of leaves of each plant type as well as soil, water and litter at 490-990 nm in 10 nm increments by hand-held radiometer.

   A larger scale habitat map will be developed by classifying each image representing 100 ha into habitat classes as defined by Ohmart et al. (1988) for the lower Colorado River. These classes include open and closed gallery forest, shrub-dominated, aquatic and emergent marsh associations, each divided into sub-categories and each with particular wildlife habitat values for relevant species. This type of mapping has been conducted for the lower Colorado River in the United States using visual interpretation of conventional aerial photographs, but has not been done by spectral methods or scaled to satellite images.

Within two weeks of each seasonal flight, a TM-3 and MODIS image of the floodplain will be acquired. Digital numbers will be converted to exoatmospheric reflectance values, and NDVI values calculated for each pixel of the image. The TM will not have sufficient detail to differentiate different plant associations, but when overlaid on the photomosaic there will be a correspondence between NDVI values and underlying vegetation units. The photomosaic base map and corresponding classified ETM+ and MODIS images can then be used for subsequent change analysis, over seasons and years. This analysis assumes that the basic vegetation structure in the delta changes only slowly, an assumption which has proven valid over twenty years of monitoring on the US stretch of the lower Colorado River. Therefore, an aerial-based photomosaic used to classify vegetation types will be valid for many years, while more frequently acquired satellite images can be used to detect changes in biomass coverage and intensity with vegetation units (Glenn et al., 2001c)."

Change detection

   "Change detection involves the use of multitemporal data sets to discrminate areas of land cover and/or soil related change. We are interested in separating out the changes of interest from all changes taking place. We expect that the use of vegetation indices will isolate changes in temporal and spatial vegetation variations. We will also search for the appropriate band combinations and to differentiate flooding events and vegetation responses. Careful attention will be placed to ensure that the changes in spectral reponses and/ or indices are indeed attributable to a change in land cover and not due to extraneous factors such as solar angle (time of day) or atmospheric conditions.

   BoR reports of the historic flow events in the last 20 years will provide information on the period and magnitude of the anthropogenically-caused flooding. Correlating these pulse floods with images of the delta in the post-flooding period gave an overview of the effect of the flooding on semiarid riparian vegetation extent and habitat, and provided a high correlation (r2 = 0.931) between percent vegetation and the years of flow (Zamora et al., in press). That study concluded that less than 1% of the base flow of the river is required to support the regeneration of native trees by washing salts from the banks and allowing mesophytic species to be established, and thus it is possible to maintain a biodiverse ecosystem in this inherently variable semiarid zone. However, in years between flood events, when there is little to no water available, salts build up in the soils producing a salinity effect which reduces biodiversity; salt tolerant species such as the exotic Tamarix ramosissima (salt cedar) begin to fill the delta in a uniform pattern. We will produce maps of semiarid, riparian vegetation (habitat) land cover and land use change in response to available water (3-D model) contained in the flood plain at different flood stages. A predictive model of these land-cover, land-use change dynamics (cause and effect) will be built on the last 20 years of available data (historic flood flow event reports, and vegetation indices from satellite images). It will be used to make current assessments of endangered species habitat and a mosaic showing areas of fragmentation. The model will be useful for predicting the future status of biodiversity in the delta based on a minimum water requirement (Glenn et al., 2001c)."