This tool integrates valley bottom and vegetation mapping into a single workflow. The workflow is integrated into a toolbar that automates many preprocessing and analysis tasks through Python scripts.
(Ratings are given on a 1-5 scale. Click on any rating for an explanation)
The RSAC Riparian Mapping Toolbar is a stand-alone toolbar that utilizes several software packages to model the extent and composition of riparian areas. The goal of this toolbar is to bridge the gap between advanced remote sensing techniques and expert field knowledge to provide land managers a user friendly workflow for delineating and mapping riparian areas. The toolbar automates the data preparation and modeling procedures, allowing the user to focus on the aspects that require expert field knowledge or photo interpretation skills. This allows a larger audience the ability to perform this technical process. The workflow is divided into three major steps, each of which are composed of smaller sub-steps. The major steps are:
Figure 1. The RSAC Mapping Toolbar.
This toolbar is designed to integrate all steps necessary to model the riparian extent and composition. Individual tools within the toolbar can be used for a broader array of applications such as general data preparation for any project involving Landsat TM/ETM+, SPOT 5, NAIP, or DEM data; general valley bottom modeling; and general vegetation mapping. Greater detail of the methods used within each step is provided in the individual sections below.
The toolbar performs a number of data preparation steps for valley bottom modeling, and also for vegetation mapping. The procedures include: at-sensor reflectance correction for SPOT 5 and Landsat TM/ETM+ data; calculation of various spectral indices from Landsat TM/ETM+, SPOT 5, and NAIP data; mosaic images; and various topographic derivatives such as slope and height above stream channel. These steps eliminate the timely manual data preparation tasks inherent with most remote sensing-based modeling procedures.
The logistic regression-based valley bottom tool provides a highly adaptable fuzzy classification or probability that an area is a valley bottom. This tool relies on user-provided training points that are classified as either valley bottom (1) or not valley bottom (0). These training data and the associated values of user-selected independent topographic variables, such as height above the stream channel and slope for each point, are fitted to the logit function. The logit function is a special case of the generalized linear model used in logistic regression.
The valley bottom tool yields a continuous raster layer depicting the probability that an area is in the valley bottom (Figure 2). By applying different thresholds to this probability layer, land managers can produce valley bottom layers with different extents, depending on management needs.
Figure 2. Example of the valley bottom probability output.
The RSAC Riparian Mapping Toolbar provides the use of object-oriented vegetation mapping capabilities using a random forests classifier. This is a streamlined version of the classification process described in the Object-based Classification: Classification and Regression Tree (CART) method. To use these capabilities, the user provides image segments for the study area and representative training data for the different classes of vegetation. The image segments must be generated independent of the toolbar using eCognition or another segmentation software package. Forest Service users have access to eCognition through their regional remote sensing coordinators. Training data can be field-verified or photo-interpreted. Based on the training data, random forests creates a model that maps the vegetation throughout the study area. The streamlined vegetation methods have proven effective at adapting to a variety of study areas featuring different challenges to modeling riparian extent and composition.
Figure 3. Example of final riparian extent and composition output.
The accuracy of the final riparian delineation depends on the strength of the relationship between the vegetation classes and the spectral and topographic variables. The primary driver of the random forests model is often the valley bottom probability surface and several spectral variables. If an area is concave and spectrally appears to support highly photosynthetic vegetation, it is likely this area will be classified as riparian. This assumption fails over large concave basins that are dominated by dense upland vegetation. Since these areas appear to be riparian both topographically and spectrally, it is very difficult to omit these areas without omitting many areas that are truly riparian. This is the primary limitation to this tool. Additional limitations result from the consistency and accuracy of the training data provided to run the valley bottom probability and random forests models.
Contact: Ian Housman (email@example.com) for information on downloading this tool.
Hosmer, S.W.; Lemeshow, S. 1989. Applied logistic regression. New York: John Wiley and Sons.
You must have an account and be logged in to post or reply to the discussion topics below. Click here to login or register for the site.