Wind & Temperature Time Series Forecasting

This repository contains all the resources, data, and code related to my undergraduate project on wind and temperature time series forecasting using deep learning.

---

5th Semester

This directory contains a Plots subdirectory with the following visualizations:

• 2m air temperature.png: Plot of 2-meter air temperature.
• 2m dew point temperature.png: Plot of 2-meter dew point temperature.
• 10m u-component of wind.png: Plot of 10-meter u-component (zonal) wind.
• 10m v-component of wind.png: Plot of 10-meter v-component (meridional) wind.

---

6th Semester

This directory is organized into the following sections:

1. Datasets

Contains all the hourly datasets from 2013 to 2023 (11 Years).

---

2. Plots

Organized as follows:

• 24 Hours to 6 Hours

  • Multivariate (u10 + v10)
    • u10_v10_6hrs/
      • u10_v10_6hrs_loss.png: Loss curve for 6-hour forecasts of u10 and v10.
      • u10_v10_6hrs_plot.png: Predicted vs. actual plots for 6-hour forecasts of u10 and v10.
    • u10_v10_6th_hour/
      • u10_v10_6th_hour_loss.png: Loss curve specifically for the 6th forecast hour.
      • u10_v10_6th_hour_plot.png: Predicted vs. actual plots for the 6th forecast hour.
  • Univariate (u10)
    • u10_Loss_24h_6h.png: Loss curve for univariate 6-hour u10 forecasts.
    • u10_Plot_24h_6h.png: Predicted vs. actual plots for univariate 6-hour u10 forecasts.

• 30 Days to 7 Days / Univariate

  • t2m/
    • t2m_loss_30d_7d.png: Loss curve for 7-day t2m forecasts.
    • t2m_plot_30d_7d.png: Predicted vs. actual plots for 7-day t2m forecasts.
  • u10/
    • u10_Loss_30d_7d.png: Loss curve for 7-day u10 forecasts.
    • u10_Plot_30d_7d.png: Predicted vs. actual plots for 7-day u10 forecasts.

• Visualizing u10 Plots

  • Plot for u10 (Hourly).png: Hourly spatial mean time series for selected days.

---

3. Source Code

Organized as follows:

• 24 Hours to 6 Hours

  • Multivariate (u10 + v10)
    • u10_v10_6hrs/
      • u10_v10_6hrs.py: Script for forecasting the next 6 hours of u10 and v10 wind components using the past 24 hours of spatially averaged data.
    • u10_v10_6th_hour/
      • u10_v10_6th_hour.py: Script for generating visualizations and metrics for the 6th forecasted hour using the outputs from the multivariate model.
  • Univariate (u10)
    • u10_24h_6h.py: Script for forecasting the next 6 hours of the u10 wind component using the past 24 hours of spatially averaged data.

• 30 Days to 7 Days / Univariate

  • t2m/
    • t2m_30d_7d.py: Script for forecasting the next 7 days (168 hours) of spatially averaged 2-meter temperature (t2m) using the past 30 days (720 hours) of data.
  • u10/
    • u10_30d_7d.pth: Trained model weights for 7-day u10 forecasting.
    • u10_30d_7d.py: Script for forecasting the next 7 days (168 hours) of spatially averaged u10 wind component using the past 30 days (720 hours) of data.

• Visualizing u10 Plots

  • u10_Plots.py: Script for generating hourly spatial mean plots for a specific day of each year (2013–2023).

---