WindSR Dataset (Super-resolution dataset for satellite wind speed)

The WindSR Dataset contains more than 22,000 pairs of low-resolution/high-resolution wind speed images. The dataset is divided into two sets: a training set and a testing set. The training set encompasses 20,046 samples, each consisting of two wind speed measurements. One measurement comprises a high-resolution image with dimensions of (400, 400), representing wind speed data at a resolution of 7 km. The second measurement consists of a lower-resolution image with dimensions of (100, 100), representing the same wind speed data but at a resolution of 28 km. The testing set contains 2,102 samples and mirrors the structure of the training set, with each sample comprising a pair of wind speed measurements at different resolutions. WindSR dataset can be used for training super-resolution models focused on satellite data.

Here is a sample HR/LR pair from the WindSR dataset:

To understand how the dataset is prepared, you can read our paper published in the IEEE Access journal.

WindSR: Improving Spatial Resolution of Satellite Wind Speed Through Super-Resolution»

Download dataset

Training dataset (20,046 samples)

Validation dataset (2,102 samples)

Loading dataset

To load the datasets and check their shapes, you can use the following Python code:

import numpy as np

# Load the arrays
HR_TRAIN = np.load('hr_train.npy')
LR_TRAIN = np.load('lr_train.npy')

# Print the shapes
print('HR_TRAIN shape:', HR_TRAIN.shape)
print('LR_TRAIN shape:', LR_TRAIN.shape)

Citation

If you use this dataset in your research, please cite the following article. Thank you!