The main purpose of this task is focused on the combination of data from multiple sources and in multiple modes.
With TanDEM-X and Sentinel-1 data (with 11 and 6 days as their revisit time, respectively), good results were obtained in terms of crop-type mapping by time series of coherence data. However, a complete analysis of the added value of this type of data and the best procedure to combine it with backscatter information is necessary.
Pol-InSAR data acquired by TanDEM-X during its Science Phase in 2015, which were gathered using very large baselines (2-3 km), were exploited for crop parameter retrieval using the common RVoG model. Later experiments have confirmed the limitations of this model to completely describe the observed data.
Consequently, alternative approaches based on machine learning should be explored for the inversion of biophysical parameters. In addition, tests over forest areas in Spain, for which ground data are available, are being carried out. On the other hand, single-pass coherence at HH and VV channels has been recently used as input feature for crop-type classification.
Experiments with multi-frequency radar data are mostly limited to comparisons of performances of the different bands. Testing the effective combination of C and X-band data to produce improved EO products, based on the complementary sensitivities of these bands and on the different temporal and spatial resolutions they provide, is a key point of this project task.
The strong dependence of radar data on the observation geometry results in a clear potential for joint exploitation of data acquired by the same sensor with different angles of incidence and/or in ascending and descending orbits. In other words, the fusion of time series of different incidence angles and tracks is expected to provide good results. For this purpose, the used data is the available from Sentinel-1 and X-band satellites (TerraSAR-X, TanDEM-X and PAZ).
The fusion of radar and optical data has been proposed many times in the past as a way to improve information revisit time and to enhance the performance of EO products by combining the complementary sensitivities of both domains. However, with the launch of Sentinel-1 and Sentinel-2 constellations, both providing time series of images with 6 and 5 days of revisit time, this fusion can provide much more information than envisaged in the past.
UA is responsible for the coordination of this project task.
Contribution of repeat-pass interferometry to the estimation of biophysical parameters of the scene and in land cover classification
Development of enhanced techniques for the combination of time series of repeat-pass coherence with other data sources
Improvement evaluation of land cover and crop type classification provided by single-pass polarimetric SAR interferometric data from TanDEM-X
Fusion of time series of data acquired at different frequency bands (C and X) assessment in classification and scene parameter retrieval
Development and evaluation of procedures for combining data from multiple tracks (incidence angles and ascending/descending orbits) in classification, scene parameter and deformation time-series retrieval
Quantification of the added value of the radar and optical data fusion in crop growth monitoring, with both ground-based and satellite time series
Development of a mathematical tool to pre-assess the gain in retrieval applications provided by the combination of data from multiple sources
Extract recommendations on the observation scenarios (data type, revisit period, daily times, processing schemes...) for future GEOSAR missions