Cloud platforms have given access to scientists in a vast database and processing power. This has allowed several breakthroughs in big data management, processing, and spatio-temporal scalability. However, the accuracy of these big data analytics and products is spatially restrained and it does not offer spatial information on the confidence of these analyses. Here, we provide a semi-automated workflow that estimates the spatial explicit uncertainty of a Belize-wide satellite derived bathymetry (SDB) procedure— roughly 7,017 km². The workflow optimizes the final bathymetry output through the training dataset by including additional reference points of lower uncertainty. We utilize the cloud-based by-products to acquire the probability of successful depth estimations and estimate uncertainty. The reference dataset comes from the ICEsat2 satellite resampled to the 10-m spatial resolution of the Sentinel-2 Level 2A products. For active learning and retraining reasons, the reference data are divided into six subsets. Thus, the uncertainty is estimated for each individual subset and then the retraining based on the rest of the subsets takes place.  Our results indicate that our demonstrated data-driven approach is able to achieve better accuracies compared to the initial SDB and map spatially-explicit nationwide uncertainty. Moreover, scientists could exploit our approach for the identification of biased reference data that mess with their models. Our novel tool can aid uncertainty-aware policy and decision making regarding the protection and conservation of coastal ecosystems.

Primary Presenter: Spyridon Christofilakos, German Aerospace Center (DLR) (spyridon.christofilakos@dlr.de)

Authors:

Avi Pertiwi, German Aerospace Center (DLR)  (Avi.Pertiwi@dlr.de)

Benjamin Lee, German Aerospace Center (DLR)  (Chengfa.Lee@dlr.de)

Dimosthenis Traganos, German Aerospace Center (DLR)  (Dimosthenis.Traganos@dlr.de)