Method of predicting streamflow data
Inventors
Assignees
University of Alaska Fairbanks
Publication Number
US-11238356-B2
Publication Date
2022-02-01
Expiration Date
2038-06-22
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Abstract
A system and method for generating synthetic streamflow data for a malfunctioning streamgage is provided. The method uses both classification and regression techniques to accurately predict streamflow data for the malfunctioning streamgage based on measured streamflow data from other streamgages and based on correlations between the streamgages. The system and method may also provide a method of improved flood forecasting, by updating flood forecasts using synthetic streamflow data when measured streamflow data from one or more streamgages are unavailable. The system may generate flood forecast information and/or flood warning messages.
Core Innovation
The invention provides a system and method for generating synthetic streamflow data for a malfunctioning streamgage by using both classification and regression techniques. It processes measured streamflow data from a network of streamgages and leverages correlations between them to create accurate synthetic data when there is an interruption in the data feed from one or more gages. The method clusters streamgages into groups based on their correlated behaviors and then generates synthetic streamflow information for the affected streamgage using group information and predictive models.
The background problem addressed is the critical need for continuous, accurate streamflow data for hydrological prediction systems, particularly for flood forecasting. Flood events and malfunctioning streamgages can interrupt the flow of real-time data required by agencies for situational awareness and emergency operations, leading to reduced ability for effective decision-making and flood management.
By analyzing historical streamflow data to identify clusters and correlations among streamgages and using this as training for machine learning models (such as ensembles of decision trees), the system can produce near real-time synthetic data that matches the temporal resolution of the original measurements. This enables flood forecasting systems to update predictions and warnings even when real-time measurements from particular gages are not available.
Claims Coverage
There are four independent claims in this patent, each defining key inventive features involving the use of correlated streamflow data, clustering, predictive modeling, and integration with flood forecasting.
Generating synthetic streamflow data using streamgage clusters based on correlation
This feature involves: - Receiving measured streamflow data from multiple streamflow sources, including a target source. - Clustering these sources into two or more groups using the measured data and correlations between them. - Using information about the groups to generate synthetic streamflow data for the target source if its measured data is interrupted.
Creating a predictive model using clustering information to generate synthetic streamflow data
This feature covers: - Receiving measured streamflow data and clustering information about a plurality of sources (including a target). - Creating a predictive model trained using this data and clustering information. - Using the predictive model (which is an ensemble of decision trees) to generate synthetic streamflow data for the target source upon detecting a data interruption.
A method for improving the accuracy of flood forecasting with synthetic streamflow data
This feature includes: - Receiving measured streamflow data from multiple streamgages (including a target). - Clustering the streamgages based on correlations. - Building a predictive model using the clustered groups. - Generating flood forecast information and determining when there is an interruption at the target. - Using the predictive model to generate synthetic streamflow data, and updating the flood forecast with this synthetic data.
A system for integrating streamflow prediction and flood forecasting using predictive models
This feature claims: - A streamflow prediction system that receives measured data from multiple gages, clusters them, and builds a predictive model using groupings to predict synthetic streamflow data for any gage. - A flood forecasting system that generates forecast information using measured data, determines interruptions, retrieves the predictive model from the prediction system, uses it to generate synthetic data for the target gage, and updates the forecast information accordingly.
The inventive features center on methods and systems that generate synthetic streamflow data for malfunctioning gages using correlation-based clustering and machine learning models, and apply this data in real-time or near real-time for flood forecasting and warning systems.
Stated Advantages
Improves both the accuracy and timeliness of generating synthetic streamflow data when measured data from streamgages is unavailable.
Enables flood forecasting systems to continue operating and updating forecasts in real-time or near real-time even when some gages are malfunctioning or off-line.
Provides a more efficient and faster predictive modeling approach by relying on streamflow data for clustering without requiring extensive additional data collection or complex multivariate models.
Documented Applications
Generating synthetic streamflow data for malfunctioning streamgages to replace missing or corrupted real-time data.
Updating and improving the accuracy of flood forecasts and flood warning messages using synthetic streamflow data when measured data is interrupted.
Supporting flood mitigation activities and emergency response decisions by providing continuous streamflow data to water managers and authorities.
Integrating with flood forecasting systems such as WaVE for highly granular, real-time, predictive flood mapping and warnings.
Interested in licensing this patent?