Validation of the Earth Observation Land Data Assimilation System by the Field Data of ESA SPARC Field Campaign
Chernetskiy, Maxim¹; Gómez-Dans, José²; Lewis, Philip^2
1Friedrich Schiller University - Institute of Geography, GERMANY; ²NCEO & UCL, UNITED KINGDOM

Large numbers of sensors orbit the Earth providing observations at a variety of spatial and temporal resolutions and spectral windows. These observations require an interpretation to extract useful information with which to understand and monitor the land surface. Typically, individual sensors and a wide variet of different interpretation approches have been used to produce information on land surface parameters. With the advent of the Sentinel era, this approach is counter-productive. Modern data assimilation techniques allow the use of physically sound radiative transfer models to get a robust estimate of the state of the land surface (as well as detailed uncertainty information) conditional on all available observations.

In this contribution, we explore the recently published EO-LDAS (Earth Observation Land Data Assimilation System) as a way to combine optical data with different spatial resolutions and spectral characteristics. The EO-LDAS framework provides a 4DVAR assimilation scheme with a weak constraint, resulting in an estimate of the state that is constrained by both the observations (through the use of a suitable radiative transfer model) and prior knowledge of the state values or their spatial/temporal evolution. While vegetation models could be used to predict the temporal trajectory of some states such as LAI, for other components of the state that are required to model the observed sigals, these models do not exist. Simiarly, no spatial models are available. We take a regularisation approach to this problem and use conditions of temporal and spatial smoothness to better constrain the problem, and use the EO-LDAS prototype to implement the DA system.

We demonstrate spatial smoothness constraints on data coming from a number of sensors (MERIS FR, MODIS, Landsat and CHRIS/Proba) acquired within the ESA SPARC 2004 field campaign over an agricultural area in Barrax (Spain). The proposed approach is used to estimate land surface parameters as well as their uncertainties from these heteorogeneous sensors. The parameter estimates are then used to forward model other observations derived from other sensors and compared to them as a further measure of the robustness of the proposed approach. We finally indicate some shortcomings of the current implementation and discuss some possible improvements.