Weather forecasting systems and methods
Inventors
Assignees
University of Alabama in Huntsville • University of Alabama at Birmingham UAB
Publication Number
US-11249221-B2
Publication Date
2022-02-15
Expiration Date
2036-03-17
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Abstract
A weather forecasting system has a data processing system that receives weather data from one or more sources and processes such data in conjunction with a weather forecasting algorithm in order to forecast weather for one or more geographic regions. In this regard, the weather data is input into a machine learning algorithm, which applies learned weights and relationships to the inputs in order to calculate at least one score indicating a probability that precipitation or other weather event will occur in the future within a certain time period (e.g., within the next 1 hour or some other unit of time) in one or more geographic regions. For each such geographic region, the weather forecasting logic may also predict the extent to which rain or other precipitation, lightning, or other weather event will occur during the time period.
Core Innovation
The invention provides a weather forecasting system and method that utilizes a data processing system to receive and process weather data from multiple sources. This data is analyzed alongside a weather forecasting algorithm, particularly employing a machine learning algorithm that applies learned weights and input relationships to predict the probability of future weather events such as precipitation, lightning, or other atmospheric conditions for specific geographic regions and timeframes. The system outputs at least one score for each sub-region, indicating the probability and/or expected extent of a weather event occurring.
To enhance prediction accuracy, the system incorporates data from numerical weather prediction models, satellite imagery, other weather prediction sources, and topographical information. The machine learning approach is trained with historical data (training and 'truth' data), learning regression equations specific to sub-regions, which can then dynamically adjust as new observations are processed. The prediction scores can be visualized on probabilistic weather maps, providing users with a graphical representation of at-risk areas.
The problem addressed by the invention is the inherent difficulty in producing accurate and timely weather forecasts, especially for short-term (0-6 hour, or 'nowcasting') events like severe storms, tornadoes, and related hazards. Conventional forecasting struggles with uncertainties in these short timeframes, so the development of improved forecasting techniques—as disclosed—aims to deliver more accurate and earlier warnings of significant weather events.
Claims Coverage
There are two independent claims that define the main inventive features of the system and method for forecasting weather using machine learning and multi-source weather and terrain data.
Machine learning-based weather forecasting system integrating satellite, weather, and terrain data
The system comprises: - Memory for storing satellite image data, weather data (including measurements and forecasts), and terrain data for a geographic region. - At least one processor programmed to: - Identify and track cumulus cloud objects within satellite images over time, with each cloud object representing a cumulus cloud. - Determine a plurality of interest fields for a cloud object based on weather measurements for a first sub-region corresponding to the cloud path. - Calculate a convective initiation score indicating the probability the tracked cumulus cloud will produce precipitation in the future, based on the interest fields. - Analyze satellite image data, terrain data, and weather data using a machine learning algorithm, applying learned weights to input variables (including the convective initiation score) to generate a forecast score for a second sub-region within the cloud's path. The forecast score indicates the probability or extent to which a weather event is predicted for that sub-region and time period. - An output interface configured to provide output based on the forecast data.
Method for weather forecasting using tracked cumulus clouds, interest fields, and machine learning
The method comprises the steps of: 1. Storing, in memory, satellite image data, weather data (measurements and forecasts), and terrain data for a geographic region. 2. Identifying and tracking cumulus cloud objects within the satellite image data over time. 3. Determining, by processor, interest fields for a tracked cumulus cloud, with each interest field based on weather measurements for a first sub-region corresponding to the cloud's path. 4. Calculating, by processor, a convective initiation score for the cumulus cloud indicating the future probability of precipitation. 5. Analyzing satellite image data, terrain data, and weather data to derive input variables for a machine learning algorithm, with at least one variable including the convective initiation score. 6. For at least a second sub-region in the predicted cloud path, applying learned weights to the input variables and mathematically combining them via the machine learning algorithm to yield a forecast score for that sub-region and time period. 7. Providing a weather forecast for the geographic region based on the forecast score.
The independent claims establish a weather forecasting architecture integrating identification and tracking of cumulus clouds, derivation of interest fields, calculation of a convective initiation score, and the use of data-driven machine learning algorithms (via weighted input variables) to forecast local weather events, outputting results as weather data or maps.
Stated Advantages
Provides a probabilistic weather map for graphical illustration of areas likely to be affected by precipitation or other weather events within a specified timeframe.
Enhances forecast accuracy and enables earlier warning for significant weather events, particularly in the 0-6 hour nowcasting window.
Optimizes weather prediction by dynamically adjusting the influence zone (footprint) based on meteorological factors for improved forecast effectiveness.
Documented Applications
Providing weather forecasts for specific geographic regions, including prediction maps showing the probability and/or extent of weather events such as precipitation and thunderstorms.
Generating graphical weather maps to display predicted precipitation, severity, or threat levels for various weather events (e.g., heavy rainfall, lightning, hail, tornadoes) in user-specified time periods.
Interested in licensing this patent?