Gravity Falls Anomaly Detection

This repository provides an R-based framework for unsupervised anomaly detection in satellite image time series, implementing the method presented in our paper “Early Detection of Forest Disturbances in Homogeneous Stands”.

Structure

This Approach applies a LSTM based autoencoder for unsupervised anomaly detection in homogeneous forests. The Model is trained on Sentinel-2 L2A time series of healthy Norway Spruce stands.

The workflow is structured as followed:

• run the scripts: R/libraries.R and R/utilities.R
• run the script: R/data_preparation.R for creating ready to use time series from GEE data
• run the script: R/main.R for model initialization and training
• run the script: R/model_inference.R for model inference on test data
• run the script: R/PD_score.R for Applying the Pre-Defoliation-Score that we established in our Paper

Citation

paper at: https://ieeexplore.ieee.org/document/11141690.

@ARTICLE{11141690,
  author={Kirsch, Maximilian and Wernicke, Jakob and Datta, Pawanjeet Singh and Preisach, Christine},
  journal={IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing}, 
  title={Early Detection of Forest Disturbances in Homogeneous Stands – Deep Learning-Based Monitoring of Picea Abies}, 
  year={2025},
  volume={18},
  number={},
  pages={22300-22316},
  keywords={Climate change;Remote sensing;Insects;Autoencoders;Forestry;Climate change;Anomaly detection;Anomaly detection;autoencoder;bark beetle;forest;long short term memory (LSTM);remote sensing;spruce},
  doi={10.1109/JSTARS.2025.3603094}}

Abstract

Climate change has increased the vulnerability of forests to insect-related damage, resulting in widespread forest loss in Central Europe and highlighting the need for effective, continuous monitoring systems. Remote sensing-based approaches, often rely on supervised machine learning (ML) algorithms that require labeled training data. Monitoring temporal patterns through time series analysis offers a potential alternative for earlier detection of disturbances but requires substantial storage resources. This study investigates the potential of a Deep Learning algorithm based on a Long Short Term Memory (LSTM) Autoencoder for the detection of anomalies in forest health (e.g., bark beetle outbreaks), utilizing Sentinel-2 time series data. This approach is an alternative to supervised ML methods, avoiding the necessity for labeled training data. Furthermore, it is more memory-efficient than other time series analysis approaches, as a robust model can be created using only a 26-week-long time series as input. In this study, we monitored pure stands of spruce in Thuringia, Germany, over a 7-year period from 2018 to the end of 2024. Our best model achieved a detection accuracy of 87% on test data and was able to detect 65% of all anomalies at a very early stage (more than a month before visible signs of forest disturbance). Compared to the established time series break detection algorithm – breaks for additive season and trend and isolation forest anomaly detection, our approach consistently detected a higher percentage of anomalies at an earlier stage. These findings suggest that LSTM-based Autoencoders could provide a promising, resource-efficient approach to forest health monitoring, enabling more timely responses to emerging threats.

Model Structure

Workflow

Pre-Defoliation-Score