pressure_regression

In this experiment we implement a model to estimate the centre pressure value (hPa) of a typhoon given its satellite image.

Architecture

Results

Our model achieved a mean error of 8.56 hPa. Below you can find a scatter plot showing the estimation vs ground truth values on the validation data. We remark that the model is able to capture the overall trend.

Image format

• Images must be in range of [0, 255], where 0 and 255 correspond to 160 Kelvin and 310 Kelvin, correspondingly.
• The model accepts 128x128 images with resolution 1 pixel ≈ 10 Km. To this end we crop 128x128 regions from resized Digital Typhoon 256x256 images (original images come as 512x512).
• Images are assumed to have the typhoon eye in the image centre (i.e . at position [63, 63]).

Usage in code

You can also use the model in your code.

Load model

from pyphoon.models.tc_pressure_regression import tcRegNet
model = tcRegNet('weights.hdf5')

Preprocess data

from pyphoon.models.tc_pressure_regression import tcRegPreprocessor
X = ...  # Load (1, 256, 256) image or (N, 256, 256) array of images
X = tcRegPreprocessor().apply(X)

Prediction

Make sure to crop the images so as to take a centred square of dimension 128x128.

X = X[:, 64:64+128, 64:64+128, :]
y_pred = model.predict(X)