Radar Nowcasting with Dual Machine Learning Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current radar nowcasting methods face challenges in accurately predicting short-term weather patterns due to low spatial and temporal resolution, and the decoupling time of predictions is limited, often resulting in inaccurate precipitation forecasts and poor wind velocity predictions.
Innovation Solution
The use of a combination of three machine learning models to enhance radar image nowcasting by translating Numeric Weather Prediction (NWP) data into realistic radar reflectivity images, incorporating dynamic and thermodynamic conditions, and increasing spatial and temporal resolution, while extending the decoupling time of predictions to 3-6 hours.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Duration of action of moving object
If traditional NWP models are used for weather forecasting, then predictions can be made for long time frames (10 days or more), but the spatial and temporal resolution is coarse and computational complexity is high
Solution Approach 1:
The patent segments the forecasting task into two distinct components: a long-term NWP model for general weather patterns and a specialized machine learning model for short-term nowcasting (0-6 hours). This segmentation allows each model to optimize for its specific time frame and resolution requirements, with the ML model processing radar image sequences at high temporal resolution (minutes) and spatial detail that would be computationally prohibitive for traditional NWP models to maintain over 10-day periods.
2Productivity
If radar-only based nowcasting methods like TREC are used, then processing speed is fast and temporal resolution is high (minutes), but spatial resolution and accuracy of internal dynamics are insufficient
Solution Approach 1:
The patent merges the strengths of traditional radar-based nowcasting methods (fast processing, high temporal resolution) with machine learning techniques by training a neural network on sequences of radar images. This combination maintains the rapid processing capability of radar-only methods while significantly improving spatial resolution and the representation of internal weather dynamics through the ML model's ability to learn complex patterns from historical radar data sequences.
3Manufacturing precision
If machine learning models are trained on radar image sequences, then spatial and temporal resolution of predictions is improved, but the decoupling time is limited to only 1-2 hours
Solution Approach 1:
The patent introduces NWP model outputs as an intermediary that bridges the gap between radar image sequences and future predictions. The system uses radar images to train a machine learning model that incorporates NWP forecast data, allowing the model to extend its predictive capability beyond the 1-2 hour decoupling limitation. The NWP data serves as a mediator that provides broader atmospheric context and extends the effective prediction horizon while maintaining the high resolution benefits of radar-based approaches.
4Ease of operation
If TREC calculates correlation coefficients between successive radar images, then motion vectors can be determined, but contradictory vectors may occur and wind velocity prediction is poor
Solution Approach 1:
The patent applies feedback mechanisms by training the machine learning model on sequences of radar images where the model learns from historical patterns of motion vector evolution. The system incorporates feedback loops that allow the model to correct contradictory vectors by comparing predicted motion patterns against actual observed sequences, progressively improving wind velocity prediction accuracy through iterative learning from training data that includes numerous examples of vector field evolution.
Data Source
AI summary
Predicting weather radar images by building a first machine learning model to generate first predictive radar images based upon input weather forecast data, and a second machine learning model to generate second predictive radar images based upon historical radar images and the first predictive radar images. Further by generating enhanced predictive radar images by providing the first machine learning model weather forecast data for a location and time and providing the second machine learning model with historical radar images for the location and an output of the first machine learning model.