Minimize Featured Stories

Refining Land Surface Product Consistency and Quality from Sensor Constellations

09 August 2012

Web Content Image

There is no doubt that Earth Observation (EO) data has great potential to contribute to land surface monitoring, providing consistent global coverage with historical archives to document changes. The increasing number of EO missions is opening new opportunities to users to access a larger number of data products derived from different sensors providing estimates of the same geo-physical or climate variables.

Modeling and predicting future changes of the land surface and their interaction with the atmosphere requires quantifying and monitoring surface properties such as, vegetation and soil percentage coverage, area of leaf cover per unit surface area and concentration of photosynthetic active pigments within the canopy.

Inconsistencies in the methodologies used to generate these products make it difficult to integrate products from more than one sensor system into modeling assimilation schemes.

The Support to Science Element (STSE) study EO-LDAS (Earth Observation - Land Data Assimilation Scheme) aims to overcome these difficulties by developing a scheme to directly assimilate radiance fields, instead of higher level data products.

EO-LDAS approaches the estimation of state variables (quantitative properties of the land surface) from remote radiometric measurements as an optimisation problem with multiple constraints. It provides the most reliable estimate of state vector elements in space/time with associated uncertainties, given our current understanding of processes (encoded as models) and radiometric measurements, provided by satellite observations with different spatial, spectral and temporal resolution. By utilizing assimilation techniques the EO-LDAS software basically simulates a virtual super sensor from observations made by a constellation different sensors.

EO-LDAS Software & Validation

Validation of the EO-LDAS prototype was undertaken to ensure the correct operation of the software and to demonstrate the utility of the approach.

Philip Lewis of the University College of London (UCL) and a key scientist behind the EO-LDAS development highlights the encouraging results in the Remote Sensing of Environment journal. This article is freely available on the Science Direct web pages. His team is convinced that the scheme offers a promising future as an operational system to estimate land surface properties from EO data.

A number of future enhancements have been proposed as the next steps towards a more operational implementation, including, extensions to the system functionality and computational efficiency improvement. The proposed enhancements allow more extensive real-world validation to be undertaken.

Michael Berger of ESA, initiator of the EO-LDAS activities, stresses that such a system would not only support the generation of consistent data products from a range of different sensors but it would also support sensor cross calibration and provide an ideal assessment tool for balancing the benefits of innovative mission concepts versus continuity missions in a given sensor constellation ensuring maximum scientific return.

In order to further promote the concept of data assimilation within the land community an extensive tutorial including software developed in the frame of the EO-LDAS activity is available at the EO-LDAS Website.

Spatial DA experiment results
original data high resolution data low resolution data
A - Original synthetic B - High resolution input C - Low resolution input
Posterior Mean Posterior uncertainty truth result
D - Posterior Mean E - Posterior uncertainty F - "Truth" (x) vs


Technical Information
The original synthetic dataset (a) has noise of sigma 0.15 added. The high resolution dataset (b) has 2/3 of the samples removed and the low resolution version (c) 1/3 removed. The posterior mean after EOLDAS d) shows very few artefacts of "blockiness" from the low resolution data. Small increases in Uncertainty where both datasets are missing can be seen (e) but there is very little bias in the overall result (f).
Courtesy: EO-LDAS Study Team