Estimating the Area of Applicability of Remote Sensing-Based Machine Learning Models with Limited Training Data

Meyer, H; Pebesma, E

Abstract in Online-Sammlung (Konferenz)

Zusammenfassung

Machine learning has become state of the art to map environmental patterns based on remote sensing data. However, usually few training data points contrast to a large area to which the model is eventually applied. This raises the question if a model being trained on limited field samples can actually be reliably applied to the entire area of interest. At the case study of a land use/land cover classification, we present how the “Area of Applicability” (AOA) of a prediction model can be assessed to limit predictions only to the area where we enabled the model to learn about relationships based on the training data. We suggest that in remote sensing-based machine learning applciations, the AOA is reported alongside predictions and model performance estimates.

Details zur Publikation

Buchtitel: 2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS
Veröffentlichungsjahr: 2021
Sprache, in der die Publikation verfasst ist: Englisch
Link zum Volltext: https://doi.org/10.1109/IGARSS47720.2021.9553999