pytorch model weights

Description

This repository contains a collection of model weight files from extensive hyperparameter testing designed for climate prediction models. Each file is a result of testing various configurations to optimize the models’ performance for different simulation scenarios. The models are primarily aimed at forecasting global temperature changes using different sets of hyperparameters.

Hyperparameter Configurations

The hyperparameters tested include combinations of different models, layer depths, hidden dimensions, loss functions, batch sizes, training epochs, patience levels, learning rates, and sequence lengths. Here’s an overview of the parameters:

•	Models: Long Short-Term Memory (LSTM), Recurrent Neural Network (RNN), Gated Recurrent Unit (GRU), Multi-Layer Perceptron (MLP), Attention Model (Attn)
•	Number of Layers: 2, 4, 6
•	Hidden Dimensions: 100, 200
•	Loss Functions: Mean Squared Error (MSE), Mean Absolute Error (MAE), Huber Loss
•	Batch Sizes: 4096
•	Epochs: 40
•	Patience Levels: 10
•	Learning Rates: 0.01, 0.001
•	Sequence Lengths: 12, 24, 48

These parameters were methodically varied to understand their impact on the accuracy and efficiency of temperature projections under different climate scenarios. This extensive testing aids in identifying the most effective model configurations for accurate long-term climate predictions.

For more detailed information and to access the full suite of tools and resources, visit the Climate Prediction 2100 GitHub Repository.

global climate simulation results

Description

Result Visualziation Video

This repository hosts model weight files used in simulating global temperature projections until the year 2100. The simulations are based on eight different Shared Socioeconomic Pathways (SSPs), using the netCDF4 file format. These weights are crucial for generating accurate and reliable climate forecasts under various future global development scenarios.

SSP Scenarios

The model weights correspond to the following SSP scenarios, each representing different radiative forcing and socioeconomic trajectories:

•	SSP1-1.9 (“SSP119”): Sustainability – Taking the Green Road (low challenges to mitigation and adaptation)
•	SSP1-2.6 (“SSP126”): Middle of the Road (medium challenges to mitigation and adaptation)
•	SSP2-4.5 (“SSP245”): Regional Rivalry – A Rocky Road (high challenges to mitigation, medium challenges to adaptation)
•	SSP3-7.0 (“SSP370”): A Rocky Road (high challenges to mitigation and adaptation)
•	SSP4-3.4 (“SSP434”): Inequality – A Road Divided (low challenges to mitigation, high challenges to adaptation)
•	SSP4-6.0 (“SSP460”): Inequality – A Road Divided (medium challenges to mitigation and adaptation)
•	SSP5-3.4 (“SSP534”): Fossil-fueled Development – Taking the Highway (high challenges to mitigation, low challenges to adaptation)
•	SSP5-8.5 (“SSP585”): Fossil-fueled Development – Taking the Highway (high challenges to mitigation and adaptation)

Time Frame for Predictions

•	Recorded Data Period: 2015 - 2023
•	Predicted Data Period: 2024 - 2100

Temperature

Monthly Deviation based on monthly average temperature (1951-1980) as in Berkeley Earth(https://berkeleyearth.org/data/) Global Gridded Data

These model weights facilitate studies on the impact of different future scenarios on global temperatures, aiding policymakers, researchers, and the public in understanding potential climate outcomes. For more detailed information and access to the full project, visit the Climate Prediction 2100 GitHub Repository.