Forecast Weather Image Generation via Cascaded Gabor Filters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Cross-correlation image processing methods for generating short-term and mid-range weather forecasts often produce highly discontinuous track vector fields, leading to inaccurate forecast images due to unconstrained track vectors that can cross over or point in opposite directions.
Innovation Solution
A cascaded field alignment technique is applied to weather radar images using a series of Gabor filters with increasing wavenumbers, iteratively determining modes of motion and deforming the template image to minimize gradient errors, resulting in a displacement field that accurately represents the motion of weather features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If cross-correlation image processing is used to generate forecast weather images, then the method can predict storm motion, but the track vector fields become highly discontinuous and inaccurate
Solution Approach 1:
The patent segments the track vector field computation into multiple scales using a cascade of filters with increasing wavenumbers. Each filter level processes the gradient error at a specific scale, building up the displacement field progressively from coarse to fine scales, ensuring continuity at each level before moving to the next.
Solution Approach 2:
The patent changes the parameter of filter wavenumber progressively through the cascade, starting with low wavenumber filters that capture large-scale motions and gradually increasing to higher wavenumbers for finer details. This parameter progression ensures that the displacement field remains continuous and physically meaningful throughout the computation.
2Device complexity
If unconstrained track vectors are used in cross-correlation processing, then the computation is simple, but the track vectors can cross over or point in opposite directions causing discontinuities
Solution Approach 1:
The patent applies preliminary constraints through the spectral filter that prevent unphysical track vector behaviors before they occur. The filter design inherently prevents track vectors from crossing or pointing in opposite directions by enforcing continuity constraints in the frequency domain, eliminating the need for post-processing corrections.
Solution Approach 2:
The spectral filter acts as an intermediary between the gradient error and the displacement field computation. It mediates the transformation by enforcing physical constraints while maintaining computational efficiency, producing track vectors that are both physically realistic and computationally efficient to generate.
3Productivity
If a single-scale filter is used to determine displacement field, then the processing is fast, but it cannot capture multi-scale weather feature motions
Solution Approach 1:
The patent segments the frequency spectrum into multiple bands using a cascade of filters with different wavenumbers. Each filter level captures motions at a specific scale, allowing the system to efficiently process multi-scale weather features by dividing the complex multi-scale problem into simpler single-scale subproblems that can be solved independently and combined.
Solution Approach 2:
The patent adds the frequency/wavenumber dimension to the processing by applying filters at multiple wavenumber levels. This transforms the single-scale problem into a multi-scale solution by operating in the frequency domain, where each wavenumber level handles a specific range of spatial scales, thereby capturing diverse weather feature motions efficiently.
Data Source
AI summary
Described are a method and an apparatus for generating a forecast weather image such as a forecast weather radar image. The method uses filters to approximate viscous alignment and to thereby determine displacement fields having meaningful structure. In various embodiments a power-law energy spectrum is utilized for deformations in the displacement field through the application of a set of Gabor filters. The Gabor filters are applied in a sequential manner to gradient error images and values of modes of motions corresponding to the Gabor filters are determined. The values are used to generate the displacement field which can then be applied to an existing weather image to generate a forecast weather image.