The state of the earth: estimating physical parameters from noisy and incomplete earth observation data

Arp, L.R.

Physical parameters are scientific variables describing the state of a system governed by physical processes. The Earth system is of particular interest, containing many sub-systems with high real-world impact. Since we cannot rely on direct measurements, we turn to remotely sensed Earth observation data, typically in the form of satellite imagery, to infer the values of these parameters instead. Unfortunately, this approach comes with two core challenges. First, there are spatio-temporal gaps in both input- and validation data, from factors such as cloud cover and limited in-situ data collection sites. Second, multiple original conditions could have produced the same observations, resulting in ill-posedness on the inference problem. The contributions in this dissertation correspond to these two challenges. First, we propose novel spatial interpolation methods to fill in spatio-temporal data gaps in an effective and scalable manner. Second, we analyse the ill-posedness of the...Show morePhysical parameters are scientific variables describing the state of a system governed by physical processes. The Earth system is of particular interest, containing many sub-systems with high real-world impact. Since we cannot rely on direct measurements, we turn to remotely sensed Earth observation data, typically in the form of satellite imagery, to infer the values of these parameters instead. Unfortunately, this approach comes with two core challenges. First, there are spatio-temporal gaps in both input- and validation data, from factors such as cloud cover and limited in-situ data collection sites. Second, multiple original conditions could have produced the same observations, resulting in ill-posedness on the inference problem. The contributions in this dissertation correspond to these two challenges. First, we propose novel spatial interpolation methods to fill in spatio-temporal data gaps in an effective and scalable manner. Second, we analyse the ill-posedness of the parameter estimation problem, noting the central role of noise and high solution sensitivity to perturbations, and propose a method to estimate the full set of potential solutions to noisy inference problems. The work contained in this dissertation directly improves data consistency for parameter estimation pipelines, and positions the field to address ill-posedness in an informed, targeted manner.
Show less

Leiden University Scholarly Publications

View statistics

Documents

In Collections

The state of the earth: estimating physical parameters from noisy and incomplete earth observation data

Funding