Python Programs for Training and Implementing the ChangePointCNN-GNSS Model

This repository contains the Python programs and model for ChangePointCNN-GNSS, a two-stage hybrid framework for the automated detection of change points in GNSS daily displacement time series. The method integrates analytical algorithms with an image-driven Convolutional Neural Network (CNN) to reliably estimate long-term site velocities, which are crucial for studying tectonic motion and maintaining geodetic reference frames.

The archive includes all necessary software to replicate the study's methodology and the primary scientific output, comprising five primary files:

Repository Contents:

1. Train_ChangePointCNN-GNSS_VGG.py: The Python script used to train the ChangePointCNN-GNSS model from scratch using the provided dataset.

2. GNSS_CPD_VelocityEstimation_VGG.py: The main implementation script for applying the trained model to new GNSS data. This script performs automated change-point detection and calculates long-term site velocities.

3. ChangePointCNN_VGG_V7.keras: The pre-trained Convolutional Neural Network (CNN) model, ready for immediate use in the detection framework.

4. data.tgz: A compressed archive containing the training dataset of approximately 6,000 labeled time series plots used to train the CNN model.

5. IGS20_Velocities_at_Global_GNSS.txt: The resulting velocity dataset for approximately 14,600 global GNSS stations, estimated using this framework in the IGS20 reference frame.

This implementation produces high-precision site velocity estimates (95% CI < 1 mm/year), providing a valuable resource for researchers in geodesy, tectonophysics, and hazard mitigation