AI-Enabled Prediction of Glacial Calving based on 4D Real-Time Multi-Sensor Monitoring

Glaciers play a critical role in the Earth's climate system and serve as important indicators of climate change. However, climatic influences and dynamic changes in glaciers are causing a loss of ice. One significant factor contributing to mass loss is glacier calving. Calving at the glacier front is a particularly challenging process to monitor, making it often a poorly understood part of the glacier dynamics.

Project Objective:

An accurate forecasting these events requires an in-depth understanding of the calving process, which will also lead to a better identification of factors that control its activity. This project aims to develop an approach for 4D multi-sensor monitoring by integrating methods from photogrammetry and artificial intelligence algorithms applied to multi-modal data.

Methodology

To capture spatial-temporal 3D data with very high detail accuracy, various sensors are used:

• Time-lapse cameras & thermal imaging cameras: Synchronized data collection allows for continuous monitoring of the glacier front, even at night.
• Multisensor systems: Acknowledging the profound influence that environmental factors and glacier dynamics have on the glacier calving, multi-sensor systems will be exploited to collect weather data and glacier velocities, among others, to enable a more holistic analysis of the calving phenomenon.

The integration of sensor technologies with AI methods will facilitate the automatic identification and quantification of calving events.

Expected Results

1. A comprehensive and high-resolution 3D inventory of glacier calving, enabled by the development of a multi-sensor monitoring system. The obtained data will enable the generation of time-series of 3D models for calving detection.

2. An accurate automatic identification of calving events and their volumes, based on the integration of sensor data (e.g., thermal cameras, weather stations, seismometers) with AI methods. This, along with sub-daily image acquisition, will allow for 4D data analysis, integrating the evolution of changes into the study of the behaviour of the glacier front.

3. Two prediction methods that contribute to the forecast of glacier calving and address the factors that control it:

• For identification of fracture activation by accurately observing calving ice blocks and monitoring their deformation until they reach a point of non-stability. Based on this, power-law functions will be defined to estimate the break-up time in advance.
• For training a specialized neural network for time series prediction that integrates the extensive calving inventory together with environmental and glacier dynamics data. This type of AI is particularly suited for predicting future states in non-linear processes. Using the trained network, the volume and location of future calving events at the glacier front will be predicted under given conditions.

Through these insights, the project will provide insights into the influential factors controlling glacier calving. This knowledge represents a significant advancement in understanding calving processes and glacier behavior, particularly in the context of climate change's challenges.

Project Team

• Laura Camila Duran Vergara
• Dr. Xabier Blanch Gorriz
• Prof. Dr. Anette Eltner
• Steffen Welsch

The project is funded by the German Research Foundation (DFG).