Neural Network Climate Forecasting via Multi-Model Ensemble Selection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current climate forecasting methods, relying on computationally intensive dynamical models, face challenges in generating fast and robust long-lead forecasts with high accuracy due to computational complexity and limited observational data, especially constrained by the short record of climate data.

Innovation Solution

An artificial neural network-based climate forecasting model is trained on global climate simulation data and fine-tuned with observational historical data, using a multi-model ensemble approach to select and combine validated GCM datasets, reducing computational power requirements while achieving forecasting skills comparable to operational dynamical models.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional dynamical models are used for climate forecasting, then forecast accuracy is improved, but computational complexity and computing power requirements increase

Engineering Contradiction:
Improveforecast accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent replaces traditional mechanical dynamical climate models with an artificial neural network-based forecasting system. The neural network learns patterns from historical climate data and GCM simulations, substituting the need for complex physical equation solving while maintaining forecasting accuracy. This substitution reduces computational complexity by using data-driven pattern recognition instead of solving differential equations in real-time.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent performs preliminary actions by pre-training the neural network on extensive GCM simulation data and historical observations before actual forecasting. The model learns from pre-processed multi-model ensembles and hindcast data, so that during operational forecasting, it can quickly generate predictions without re-running complex dynamical models. This preliminary training phase captures the complexity upfront, enabling fast subsequent predictions.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If traditional dynamical models are used for climate forecasting, then forecast accuracy is improved, but forecasting speed decreases

Engineering Contradiction:
Improveforecast accuracyVSAvoidforecasting speed
Core Design Contradiction:
Measurement precisionVSSpeed

Solution Approach 1:

The neural network system replaces slow dynamical model computations with fast pattern recognition. Once trained, the neural network can generate forecasts in minutes or seconds compared to days or weeks required by traditional dynamical models, while maintaining comparable accuracy through its learned representations of climate behavior.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent creates a computational copy of climate system behavior through the neural network. Instead of running the actual complex dynamical models during forecasting, the neural network creates a simplified copy that replicates the essential climate patterns and relationships, enabling fast predictions that mirror the behavior of full dynamical models without their computational burden.

Inventive Principle:
Principle #26Copying

3Measurement precision

If more climate processes are incorporated into GCMs on finer spatial grids, then forecast accuracy is improved, but computing power requirements increase

Engineering Contradiction:
Improveforecast accuracyVSAvoidcomputing power
Core Design Contradiction:
Measurement precisionVSPower

Solution Approach 1:

The patent extracts the essential climate patterns and relationships from extensive GCM simulation data during the training phase. The neural network learns to capture the key dynamics of complex climate processes without needing to explicitly represent all physical processes during forecasting. This extraction allows the model to achieve high accuracy using simplified computations during operational deployment.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The neural network substitutes the need for running high-resolution dynamical models with a data-driven approach that has already learned from such models. The computing power intensive process of solving fluid dynamics equations is replaced by matrix operations in the neural network that have already captured the essential physics during training, enabling accurate forecasts with minimal computing power.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

4Reliability

If ensemble modeling is used to reduce forecast uncertainty, then forecast reliability is improved, but computational power requirements increase

Engineering Contradiction:
Improveforecast reliabilityVSAvoidcomputational power
Core Design Contradiction:
ReliabilityVSPower

Solution Approach 1:

The patent merges multiple GCM datasets into a unified neural network training process. Instead of running separate ensemble members and combining their outputs, the neural network is trained on data from multiple models simultaneously, learning the consensus and divergences across models. This merging approach achieves ensemble benefits with a single model structure, reducing computational overhead while maintaining reliability.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The neural network replaces the computational mechanism of running multiple separate dynamical model simulations with a single trained network that has learned from multi-model data. The uncertainty reduction achieved through ensemble modeling is replicated through the neural network's ability to capture inter-model variability and identify robust patterns that persist across different models, eliminating the need for repeated expensive simulations.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS20240094435A1Systems and methods for selecting global climate simulation models for training neural network climate forecasting models
Publication Date: 2024.03.21 CLIMATEAI INC
  • US20240094435A1 patent drawing
  • US20240094435A1 patent drawing
  • US20240094435A1 patent drawing

AI summary

Methods and systems for generating a multi-model ensemble of global climate simulation data from a plurality of pre-existing global climate simulation model (GCM) datasets, are disclosed. The methods and systems perform steps of computing a GCM dataset validation measure based on at least one sample statistic for at least one climate variable from the pre-existing GCM dataset; selecting a validated subset of the plurality of pre-existing GCM datasets; selecting a subset of GCM datasets; generating candidate ensembles of GCM datasets; computing an ensemble forecast skill score for each candidate ensemble based on a model analog; generating the multi-model ensemble of GCM datasets by selecting a candidate ensemble with a best ensemble forecast skill score; and training the NN-based climate forecasting model using the multi-model ensemble of GCM datasets. Embodiments of the present invention enable accurate climate forecasting without the need to run new dynamical global climate simulations on supercomputers.