Neural Climate Forecasting With Re-Gridded Data Augmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing climate forecasting systems face challenges in achieving accurate, fast, and robust long-lead climate predictions due to computational complexity and limited observational data, with traditional dynamical models requiring extensive computational resources and machine learning methods being constrained by short data records.
Innovation Solution
A neural network-based climate forecasting model trained on pre-processed multi-model ensemble data from global climate simulation models, utilizing spatial and temporal homogenization, augmentation, and transfer learning to leverage both simulation and observational data, reducing computational requirements while enhancing predictive accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional dynamical models are used for climate forecasting, then forecasting accuracy is improved, but computational complexity and resource requirements increase
Solution Approach 1:
The patent creates synthetic climate data copies through data augmentation techniques, generating additional training samples from limited observational data. This allows machine learning models to be trained on expanded datasets without requiring proportionally more computational resources, resolving the contradiction between improving forecast accuracy and reducing computational complexity
Solution Approach 2:
The patent replaces traditional dynamical models (mechanical/physical systems based on fluid dynamics equations) with machine learning models that learn patterns directly from data. This substitution reduces computational complexity while maintaining forecasting accuracy, as ML models can capture climate patterns without solving complex physical equations in real-time
2Measurement precision
If traditional dynamical models incorporate more climate processes and finer spatial grids, then forecasting accuracy is improved, but computational power requirements increase
Solution Approach 1:
The patent performs preliminary data processing and augmentation steps before model training, including spatial re-gridding, temporal homogenization, and synthetic data generation. By preparing enhanced datasets in advance, the system enables ML models to achieve high accuracy without requiring the continuous computational power needed by dynamical models to resolve fine spatial grids and multiple climate processes
3Productivity
If machine learning methods are used for climate forecasting, then computational efficiency is improved, but data availability is limited by short observational records
Solution Approach 1:
The patent extensively applies data copying through synthetic data generation, creating multiple augmented versions of limited observational records. Techniques include adding noise, applying transformations, and generating synthetic samples that expand the effective size of the training dataset, thereby providing sufficient data for ML training while maintaining computational efficiency
Solution Approach 2:
The patent transforms limited observational data into expanded training sets by applying parameter changes such as spatial re-gridding to different resolutions, temporal homogenization across different time scales, and various statistical transformations. These parameter changes create diverse training samples from limited source data, resolving the contradiction between computational efficiency and data availability
Data Source
AI summary
Methods and systems for training a neural network (NN)-based climate forecasting model on a pre-processed multi-model ensemble of global climate simulation data from a plurality of global climate simulation models (GCMs), are disclosed. The methods and systems perform steps of determining a common spatial scale and a common temporal scale for the multi-model ensemble of global climate simulation data; spatially re-gridding the multi-model ensemble to the common spatial scale; temporally homogenizing the multi-model ensemble to the common temporal scale; augmenting the spatially re-gridded, temporally homogenized multi-model ensemble with synthetic simulation data generated from the spatially re-gridded, temporally homogenized multi-model ensemble; and training the NN-based climate forecasting model using the spatially re-gridded, temporally homogenized, and augmented multi-model ensemble of global climate simulation data. Embodiments of the present invention enable accurate climate forecasting without the need to run new dynamical global climate simulations on supercomputers.