Mamba Weather Timeseries

Overview

This repository is for comparing a Mamba model to a LSTM model for weather prediction timeseries data.

Dataset

The model is trained on 8 varaibles from the GFS 0.25 degree dataset. The training dataset has a sample size of 36,160 and the test dataset has a sample size of 9,040.

Variables

• Temperature
• Surface pressure
• V component of wind
• U component of wind
• Specific humidity
• Convective precipitation
• Total precipitation
• Water equivalent of accumulated snow depth

At each time step, data is taken from 200 coordinated. The data is normalized to fit within the range of -1 to 1.

Models

I trained two models, a Mamba and a LSTM. Both models have the save parameters:

• Hidden dimensions: 512
• Number of layers: 3

Metrics

| Variable | Mamba (MSE) | LSTM (MSE) | |----------|-------------|------------| |Temperature|1.6630e-05|1.6136e-05| |Surface presure|5.1565e-05|7.3468e-05| |V component of wind|0.0008|0.0023| |U component of wind|0.0003|0.0020| |Specific humidity|0.0002|0.0009| |Convective precipitation|5.1313e-05|6.3685e-05| |Total precipitation|3.4177e-05|4.7444e-05| |Water equivalent of accumulated snow depth|1.4074e-06|1.1085e-12| |Average|0.00018|0.00068|

Lower MSE is better and shown in bold