Speaker
Description
The coherence is a parameter that characterizes the noise of the interferometric phase, as well as the SNR improvement achieved with multi-image filtering or averaging of backscatter images. Our main interest relates to the fact that the coherence depends on specific characteristics of the imaged scatterers, so that it can support applications as land-use classification and bio- and geo-physical parameter retrieval.
During the tomographic phase the BIOMASS P-band SAR acquires tomographic stacks of seven images. All acquisitions are using quad polarization. The intervals between the subsequent acquisitions are 3 days, the perpendicular baselines around 700m. A tomographic set provides multi-temporal backscatter, as well as interferometric coherence values for 3, 6, 9, 12, 15, and 18 day intervals, all at hh, hv, vh, and vv polarization. Polarimetry (e.g. polarimetric decompositions) is therefore also supported and can be done independently for the seven acquisitions, supporting on one hand the investigation of temporal change and on the other hand the calculation of temporal averages and other statistical parameters.
We selected data sets over different biomes and investigated the dependences of the coherence on relevant factors. Pairs with different time intervals, baselines, polarizations and backscatter levels were used to characterize temporal decorrelation, volume decorrelation effects, polarization characteristics, and noise effects. Polarimetric decompositions, cross-channel coherence calculations, and noise equivalent sigma zero (NESZ) estimates were used to support the interpretation.
Over most surfaces several factors are relevant for the resulting coherence. The signal noise is very relevant for all areas with a low backscatter. For relatively smooth surfaces, e.g. water surfaces, agricultural fields, and flood plains, the backscatter is often at or even below the NESZ level, resulting in very low coherence values almost independently of the actual presence of temporal change, especially at cross-polarization. A useful test is to calculate the hv to vh polarization coherence for an individual date. Apart from the noise decorrelation a coherence of 1.0 is expected in this case for almost all scatter types. So, the observed coherence directly indicates the level of the signal noise related decorrelation. For a signal to noise ratio of 1 (same level of signal and noise, total scattering observed is twice the NESZ) the noise coherence factor becomes 0.5.
At co-polarization the backscatter over low vegetation (e.g. grassland) and agricultural fields the backscatter level is often sufficiently higher than the NESZ. Consequently, temporal change can be identified through lower coherence levels. Multi-temporal coherence composites can be used to show change occurring in different intervals in different colors.
Over forest the backscatter at co- and cross-polarization is much larger than the NESZ, noise decorrelation becomes irrelevant. Using the tomographic stacks, it is difficult to separate gradual temporal decorrelation and volume decorrelation effects because of the linear dependence of the available time intervals and perpendicular baselines. The coherence slightly decreases with increasing baseline and time interval. Observing very similar coherence levels for different 3-day intervals, we rather think that the volume effect is the more relevant cause for the decorrelation. This is observed at co- and cross-polarization. At cross-polarization dihedral effects are not the dominant scattering mechanism, so scattering from stronger branches is thought to explain the relatively high cross-polarization coherence values over forest. Based on these early results we expect that the baseline dependence of the P-band coherence over forest, determined for BIOMASS tomographic stacks, is of interest for the retrieval of forest parameters as the above ground biomass.