CanLaD - Canada Landsat Disturbance – Forest Insect Update

This repository contains the code used to generate the results presented in the paper:
Remote Sensing-Based Interannual Monitoring of Major Insect Disturbances in Canadian Forests (in review).
We used LandTrendr and the TempCNN model to detect forest disturbances across Canada from 1985 to 2024.

👉 Explore the interactive demo

---

Table of Contents

1. Data Availability
2. Methodology
3. Citation

---

Data Availability

Below is a summary of all the published datasets:

• Canada Landsat Disturbance (CanLaD) – 2025 Forest Pest Update
  Open.canada.ca

• Disturbance Detection Prior to 1984
  Open.canada.ca

• Canada Landsat Disturbance (CanLaD) – 2017
  Open.canada.ca

• Example Training Dataset
  GitHub

---

Methodology

1. Train the TempCNN model
   Use 01_Main_Train_Kfold.py
   👉 Example training dataset

2. Detect disturbance breaks with LandTrendr
   Use 02_LandTrendr_get_breaks.py

3. Apply TempCNN model to each break
   Run 03_Inference_TempCNN_LTD.py

4. Generate annual disturbance maps
   Convert outputs using 04_Transform_to_time_series.py

5. Apply cleaning with 12-pixel sieve filter
   Use 05_sieve_time_series.py

6. Extract the latest disturbance type and year
   Use 06_Latest_From_Time_Series.py

🖥️ Note:
The entire workflow was executed using the Government of Canada’s High Performance Computing (HPC) service.
The TempCNN model has a small memory footprint (<1 MB), making CPU-based inference efficient.
Canada was divided into ~2,000 tiles (10,000 km² each) for parallel processing.
Full processing time: ~2 days.

---

Citation

If you use this code or dataset, please cite the associated article (in review):

@article{perbet_canlad_2025,
  author    = {Pauline Perbet et al.},
  title     = {Remote Sensing-Based Interannual Monitoring of Major Insect Disturbances in Canadian Forests},
  journal   = {In Review},
  year      = {2025}
}