Ground Truth Sampling to Support Remote Sensing Research and Development: Submersed Aquatic Vegetation Species Discrimination Using an Airborne Hypers pectral/L dar System

Reif, Molly; Piercy, Candice; Jarvis, Jessie; Sabol, Bruce; Macon, Chris; Loyd, Richard; Colarusso, Phil; Dierssen, Heidi; Aitken, Jen

Inferring conditions about the earth’s surface using remotely sensed electrooptical measurements almost always requires the use of reference, or “ground truth” data. Ground-based measurements typically involve collecting measurements of the phenomena or target being remotely sensed and can range from employing physical field checks to aerial photography (Lillesand et al. 2004). More commonly, they include physical and chemical measurements with geographic positions and observations for comparison with remotely sensed imagery. Generally, ground truth is used to assist with 1) image analysis and interpretation (e.g. image classification) of remotely sensed imagery, 2) remote sensor calibration, and 3) accuracy assessment of image analysis results (Lillesand et al. 2004). Much of the emphasis on collecting ground truth is for verification and assessment of imagery analysis; however, there are no universally accepted standards for assessing accuracy (Congalton and Green 2009). Considerations for assessing accuracy in the collection of ground truth should include such topics as the distribution of the phenomena being mapped, sample size, number, type, and frequency of collection, and consistency and objectivity in measurement and collection (Congalton and Green 2009). Researchers in the 1970s began to introduce simple techniques for accuracy assessment (Ginevan 1979), followed by more detailed efforts described in Congalton et al. (1983). Furthermore, some guidance and examples for statistically sound approaches in determining sample size are available (Hord and Brooner 1976, van Genderen and Lock 1977, Hay 1979, Rosenfield et al. 1982, Congalton 1988). Also, consideration for choosing the appropriate sampling strategy is described in Ginevan (1979), Fitzpatrick-Lins (1981), and Stehman (1992). When a remote sensing or image processing technique is under development, it is mandatory to have sufficient ground truth data to test not only the accuracy of the final image analysis output, but also the intermediate steps in that process. General guidance for ground truth collection in remote sensing research and development (R&D) is extremely limited, as it is dependent on specifics of the technology being developed. Although much work has been done to develop procedures for accuracy assessment, there is a need to better understand and develop the role of ground truth and collection methods for evolving remote sensing technology.