Current systems for acquiring and accessing rangeland health data are inadequate to meet the needs of an increasingly complex management process that requires public accountability. Traditional field-based data acquisition, such as visual assessment, quadrat, and point frame methods are highly subjective, time consuming or too costly to practically apply over large land management units. While the availability of remotely sensed data has expanded substantially with the use of satellite and aircraft mounted hyperspectral scanners, these products are expensive, lack flexibility, can be misinterpreted, and often lack the resolution required by land managers. The purpose of this project is to reduce the costs and improve the efficiency and objectivity in monitoring for rangeland health conditions and wildlife habitat assessments. OBJECTIVES: 1. Identify a test site and develop a cooperative research agreement. 2. Establish the accuracy, resolution and level of detail in remotely sensed images acquired by the ARS system. 3. Compare the accuracy of VLSA imagery with traditional land cover assessment and classification techniques. 4. Determine the optimum image processing software and methodologies for interpretation of VSLA images. 5. Assess the density of digital photo samples needed to create accurate land cover maps. 6. Identify key features of the Geographic Information System component of RangeTech DSS. APPROACH: BRI will compare the accuracy and variability of data collected from four methods: VLSA imagery, satellite imagery, point sample data, and quadrat visual assessment. Thirty VLSA images per linear mile will be acquired from the research site using the HPGRS method. In conjunction with the image collection, each photo site will be assessed for vegetative and ground cover utilizing traditional point sampling and quadrat techniques. We will generate classifications from the VLSA images categorizing ground, shrub, and weed and canopy cover and density using the HPGRS method. Then we will develop common metrics for the point sample and quadrat data and compare the accuracy between the classification techniques. An accuracy assessment will be performed by ground-truthing 20% of the aerial photo frames. Data shall be collected by taking a one-square meter photo of the ground using a Canon E20 digital camera on a tripod. The ground cover and vegetation density results from the on-the-ground sampling, the Canon E20 on-the-ground sampling, the Canon E20 photos, the VLSA imagery, and the satellite imagery shall be compared to each other using 2-sample t-tests. Cross-validation techniques will yield a matrix illustrating errors of omission and errors of commission. An identical set of aerial photos will be classified with three different image processing softwares: VegMeasure, Feature Analyst, and ERDAS Imagine. The results of the classifications shall be compared with each other and with ground-truthed data. Integrating the classification method described above, we will interpolate the site using all available photo frames as sample points. Next, we will conduct a sensitivity analysis by systematically dropping 10% of the photo frames and interpolating land cover classes across the site once again. Finally, we will employ cross-validation techniques to compare the accuracy and error maps associated with the three interpolated land cover maps. The Kappa statistic will be computed to quantify the degree of coincidence between object class boundaries. Working with a cooperator group, we will develop goals and appropriate metrics for the Range Health DSS. Finally, we will develop a conceptual plan for types of data, metrics, GUI interface, etc. that will be included in the GIS-based range health information system
