USGS Banner

Project ID:2006VT26B

Title: An Adaptive Management System using Hierarchical Artificial Neural Networks and Remote Sensing for Fluvial Hazard Mitigation

Project Type: Research

Start Date: 03/01/2006

End Date: 02/28/2007

Congressional District: First

Focus Categories: Geomorphological Processes, Hydrology, Models

Keywords: artificial neural networks, fluvial geomorphology, remote sensing, eCognition, adaptive management, uncertainty assessment, stream assessment, hydrology

Principal Investigators: Rizzo, Donna;: Morrissey, Leslie

Federal Funds: $11,400

Non-Federal Matching Funds: $70,704

Abstract: Adaptive management of hydrologic systems requires modeling of dynamic, nonlinear relationships and assimilation of volumes of disparate data types over variable temporal and spatial scales. Artificial neural networks (ANNs) offering the capability to assimilate such complex data in real-time are thus promising tools for evaluating management alternatives. We propose to develop and test an hierarchical ANN system to more effectively integrate, model, and manage spatial and temporal hydrologic and fluvial geomorphic data. To demonstrate the efficient performance of ANN architectures in data assimilation, reduction, and classification at multiple scales, we will develop methods to enhance the GIS-based tools currently in use in Vermont watersheds to characterize the geomorphic condition and sensitivity of river reaches in response to historic and current watershed and corridor stressors. Input to the ANNs will include available GIS data layers, field data collected under (River Management Program's (RMP) geomorphic assessment protocol, and new data to be derived from high spatial resolution (0.16 - 2.4 m) remotely sensed aircraft and satellite data on land use and land cover, impervious surfaces, riparian buffers, and channel and valley slope. Recent advances in remote sensing technology make it possible to greatly improve the quantity and quality of input data in support of the proposed ANN. The proposed study will be conducted on three stormwater impaired watersheds in Chittenden County. These sites will be selected to take advantage of available multispectral remote sensing imagery (including LIDAR and QuickBird satellite data) and in cooperation with DEC RMP collaborators. No new remote sensing acquisitions are planned as part of this effort. Evaluation of the new data products will be conducted by ground surveys. Sensitivity analyses also will be conducted based on the results of the proposed ANN modeling system to address the relative importance of the various ground and remote sensing data sources to meet and improve upon RMP's current fluvial modeling capabilities.

The proposed modeling system is directly applicable to the fluvial hazard mitigation mission of the River Management Program (ANR/DEC), but will differ sharply from conventional hydrologic models currently in use by the volume, variety, and types of spatial and temporal data assimilated. Moreover, the architecture of the proposed hierarchical ANN system is sufficiently flexible to allow for its continual update and refinement in light of advances in our understanding of fluvial geomorphology. This research will evaluate not only a new and innovative data assimilation and analysis methodology, but also data products derived from remote sensing imagery that we believe will substantially improve hydrological modeling in Vermont. In addition, it will compliment the existing RMP state program, taking advantage of existing data, protocols, and personnel ? a modeling approach that could be adopted statewide. Our long-term goal is to build hydrologic information technology that provides watershed managers (regulators, regional planning organizations, municipalities, citizen groups, landowners, and other stakeholders) with an easy-to-use, graphical infrastructure for adaptive and effective decision-making at multiple spatial and temporal scales.

Progress/Completion Report, PDF

U.S. Department of the Interior, U.S. Geological Survey
Maintained by: John Schefter
Last Updated: Thursday, June 04, 2009
Privacy Statement || Disclaimer
|| Accessibility