Data-Centric Machine Learning for Earth Observation: Necessary and Sufficient Features

TL;DR

This paper emphasizes a data-centric approach in earth observation machine learning, using model explanations to identify minimal yet effective feature sets, thereby improving data efficiency and model performance.

Contribution

It introduces a method leveraging model explanation techniques to determine the essential features and minimal data needed for optimal performance in geospatial datasets.

Findings

Some datasets achieve optimal accuracy with less than 20% of temporal instances.

In certain cases, a single modality's time series suffices for high performance.

Model explanation methods effectively identify crucial features for geospatial data.

Abstract

The availability of temporal geospatial data in multiple modalities has been extensively leveraged to enhance the performance of machine learning models. While efforts on the design of adequate model architectures are approaching a level of saturation, focusing on a data-centric perspective can complement these efforts to achieve further enhancements in data usage efficiency and model generalization capacities. This work contributes to this direction. We leverage model explanation methods to identify the features crucial for the model to reach optimal performance and the smallest set of features sufficient to achieve this performance. We evaluate our approach on three temporal multimodal geospatial datasets and compare multiple model explanation techniques. Our results reveal that some datasets can reach their optimal accuracy with less than 20% of the temporal instances, while in other datasets, the time series of a single band from a single modality is sufficient.