Big Earth Data

Addressing incomplete reference data for leveraging Artificial Intelligence in Earth Observation applications

Artificial Intelligence (AI) has been, and will continue to be, widely applied in aerial and satellite imaging data processing for various Earth observation (EO) tasks, including environmental monitoring, climate change analysis, and agricultural assessment. However, these AI-driven applications rely heavily on the quality and quantity of available reference data to support model training and validation. Despite the critical importance of reference data in big EO data processing, several challenges often arise, such as data inconsistencies, limited quantities, incompleteness, and temporal degradation. These challenges collectively undermine the reliability of the outputs and the overall processing chain. Therefore, addressing and mitigating these reference data issues is essential for optimizing and enhancing EO-based products.

Recognizing the increasing reliance on AI in EO, this special issue focuses on advancing methodologies for addressing and managing reference data challenges in EO applications. It brings together pioneering research, theoretical insights, and innovative case studies to promote advancements in reference data enhancement and refinement. By tackling these challenges and proposing innovative solutions, this special issue aims to improve the precision and trustworthiness of AI-driven insights, providing valuable guidance for academics, industry professionals, and policymakers.

Potential topics include (but are not limited to) the following:

1.      Techniques for generating synthetic data to address limited training samples;

2.      Robust models for handling incomplete or unreliable reference data;

3.      Strategies for transferring knowledge from well-annotated datasets to others;

4.      Optimizing available reference data by selecting the most informative samples;

5.      Techniques for training models on imbalanced or incomplete datasets;

6.      Using physical constraints to guide learning in data-limited scenarios;

7.      Identifying and addressing biases introduced by imperfect reference data;

8.      Efficient workflow for spatially and/or temporally transferring available reference data;

9.      Developing benchmarks for evaluating models trained on incomplete reference data.