Abstract Title
Creating Validation Dataset for Remote Sensing of Water Resources
Additional Funding Sources
This project was made possible by the NSF Idaho EPSCoR Program and by the National Science Foundation under Award No. OIA-1757324.
Abstract
Water is essential to all life and there is an urgent need to quantify current extents of water resources and how they are changing over time. Current survey methods are either costly, time consuming by hand delineations of aerial imagery or are satellite image classifications limited to identifying relatively large bodies of water. Many water resource areas, particularly in dryland systems, are smaller than current inventories identify and/ or are covered by vegetation. By mapping mesic vegetation indicative of access to water as well as water bodies, we gain insight into water availability in these water limited systems. As we advance the classification of satellite images, we need robust validation datasets. In this work, we produced high resolution validation images through supervised classifications in Google Earth Engine then manually refined the classifications. The complexities of landscapes make supervised classification difficult due to the fact that many features are spectrally similar and classifiers have trouble distinguishing among them. With further refinement and expert knowledge, these errors in classification can be resolved. These validation images are part of the first steps to testing and creating reliable methods to accurately and quickly classify remotely sensed images. The validation datasets we produce in this work will allow us to test the accuracy of these methods during development.
Creating Validation Dataset for Remote Sensing of Water Resources
Water is essential to all life and there is an urgent need to quantify current extents of water resources and how they are changing over time. Current survey methods are either costly, time consuming by hand delineations of aerial imagery or are satellite image classifications limited to identifying relatively large bodies of water. Many water resource areas, particularly in dryland systems, are smaller than current inventories identify and/ or are covered by vegetation. By mapping mesic vegetation indicative of access to water as well as water bodies, we gain insight into water availability in these water limited systems. As we advance the classification of satellite images, we need robust validation datasets. In this work, we produced high resolution validation images through supervised classifications in Google Earth Engine then manually refined the classifications. The complexities of landscapes make supervised classification difficult due to the fact that many features are spectrally similar and classifiers have trouble distinguishing among them. With further refinement and expert knowledge, these errors in classification can be resolved. These validation images are part of the first steps to testing and creating reliable methods to accurately and quickly classify remotely sensed images. The validation datasets we produce in this work will allow us to test the accuracy of these methods during development.