Importance of bistatic SAR features from TanDEM-X for forest mapping and monitoring

Highlights

• We investigated the information content of TanDEM-X data.

• We assessed the usefulness of mono- and bistatic SAR in a land cover mapping.

• Bistatic SAR features improve classifications of forested areas significantly.

Abstract

Deforestation and forest degradation are one of the important sources for human induced carbon dioxide emissions and their rates are highest in tropical forests. For man-kind, it is of great importance to track land-use conversions like deforestation, e.g. for sustainable forest management and land use planning, for carbon balancing and to support the implementation of international initiatives like REDD + (Reducing Emissions from Deforestation and Degradation). SAR (synthetic aperture radar) sensors are suitable to reliably and frequently monitor tropical forests due to their weather independence. The TanDEM-X mission (which is mainly aimed to create a unique global high resolution digital elevation model) currently operates two X-band SAR satellites, acquiring interferometric SAR data for the Earth's entire land surface multiple times. The operational mission provides interferometric data as well as mono- and bistatic scattering coefficients. These datasets are homogenous, globally consistent and are acquired in high spatial resolution. Hence, they may offer a unique basic dataset which could be useful in land cover monitoring.

Based on first datasets available from the TanDEM-X mission, the main goal of this research is to investigate the information content of TanDEM-X data for mapping forests and other land cover classes in a tropical peatland area. More specifically, the study explores the utility of bistatic features for distinguishing between open and closed forest canopies, which is of relevance in the context of deforestation and forest degradation monitoring. To assess the predominant information content of TanDEM-X data, the importance of information derived from the bistatic system is compared against the monostatic case, usually available from SAR systems. The usefulness of the TanDEM-X mission data, i.e. scattering coefficients, derived textural information and interferometric coherence is investigated via a feature selection process. The resulting optimal feature sets representing a monostatic and a bistatic SAR dataset were used in a subsequent classification to assess the added value of the bistatic TanDEM-X features in the separability of land cover classes. The results obtained indicated that especially the interferometric coherence significantly improved the separability of thematic classes compared to a dataset of monostatic acquisition. The bistatic coherence was mainly governed by volume decorrelation of forest canopy constituents and carries information about the canopy structure which is related to canopy cover. In contrast, the bistatic scattering coefficient had no significant contribution to class separability. The classification with coherence and textural information outperformed the classification with the monostatic scattering coefficient and texture by more than 10% and achieved an overall accuracy of 85%. These results indicate that TanDEM-X can serve as a valuable and consistent source for mapping and monitoring tropical forests.

Introduction

Human activities, summarized by the UNFCCC (United Nations Framework Convention on Climate Change) as land use, land-use change and forestry (LULUCF), affect changes in carbon stocks (Intergovernmental Panel on Climate Change (IPCC, 2003)). Their role in the mitigation of climate change has long been recognized. The knowledge of carbon storage at a certain point in time as well as its changes due to deforestation, afforestation, and other land-use transformations are therefore of great importance (Watson et al., 2000). Thus, the mapping and monitoring of tropical forests as a potential significant carbon store are relevant in climate change studies and in the implementation of REDD + (Reducing Emissions from Deforestation and Degradation; Gibbs, Brown, Niles, & Foley, 2007).

Synthetic aperture radar (SAR) systems are considered an important tool for mapping and monitoring tropical forests due to their weather independence (e.g. Kuntz, 2010). High accuracies in land cover and forest/non-forest classifications have been obtained in the temperate forests of Germany using very high-resolution, multiple polarization, multi-temporal X-band SAR data from the TerraSAR-X mission (Breidenbach, Ortiz, & Reich, 2010) and in Austria in combination with a LiDAR dataset (Perko, Raggam, Deutscher, Gutjahr, & Schardt, 2011). However, longer wavelengths, such as L- and P-band are considered more appropriate than X- and C-band in the separation of different forest types and in the detection of secondary or degraded forest due to their increased penetration into the forest canopy (Castro et al., 2003, Saatchi et al., 1997, Stolz and Mauser, 1995). A number of studies show the suitability of space-borne L-band SAR systems for the large scale classification of tropical forest areas. For example, the classification of multi-temporal JERS-1 mosaics (Sgrenzaroli et al., 2004, Simard et al., 2000, Simard et al., 2001) achieved high accuracies, which could even be increased by high resolution dual-polarization ALOS PALSAR data (Hoekman et al., 2010, Longépé et al., 2011, Walker et al., 2010). Even though Morel et al. (2011) also classified plantations and forests with a high accuracy using ALOS PALSAR in a study area in Borneo, they found a significant decrease in the accuracy by adding a logged forest class to the classification. Moreover, there is a significant gap in globally available high resolution SAR coverages, especially since ALOS PALSAR and ENVISAT ended their missions in 2011 and 2012, respectively and the continuation of climate change studies must be ensured without those systems (European Space Agency, 2012).

On the other hand, single polarization X-band data show limitations in the separability of forest classes (Castro et al., 2003). Case studies using TerraSAR-X mission data in the tropical forests of Brazil, Uganda, and Central Kalimantan (Otukei et al., 2011, Santos et al., 2010, Ullmann et al., 2012) suggested that the use of multiple polarizations can provide satisfactory results for land cover and forest classifications. For example, Santos et al. (2010) used dual polarimetric TerraSAR-X data in order to classify primary forest, degraded forest, pasture, and bare soil. He emphasized the significance of the entropy obtained from the Cloude decomposition (Cloude & Pottier, 1997). Although primary forest was separable from pasture and bare soil in X-band, confusions with degraded forest were found (Santos et al., 2010).

Previous investigations showed that automated mapping with single channel high frequency SAR data does not achieve a sufficient degree of detail and accuracy, especially in the mapping of different forest types or degraded forests (Del Frate et al., 2008, Santos et al., 2010, Van der Sanden and Hoekman, 1999). Therefore, very high-resolution, different wavelengths or synergetic use with optical data, multiple polarizations and/or multi-temporal analysis are necessary to achieve high accuracies with high frequency radar (Breidenbach et al., 2010, Bruzzone et al., 2004, Erasmi and Twele, 2009, Van der Sanden and Hoekman, 1999).

Nevertheless, the objective of this investigation is to show that accurate forest monitoring results can be achieved without the necessity to deploy multiple polarizations and multi-temporal data, which may be inconsistent in acquisition modes and timing due to acquisition constraints. Those constraints are caused by the fact that available radar systems providing high resolution and polarimetric or repeat-pass capabilities are commercially operated and thus, data acquisition is tasked on demand, making consistent coverages nearly impossible.

The TanDEM-X mission might be an option to overcome these constraints given by incomplete coverage or information content. Starting in 2010, the aim of the TanDEM-X mission is the acquisition of global coverage of bistatic, single-pass interferometric SAR images to produce an accurate digital elevation model (Krieger et al., 2007). For that, the two TanDEM-X sensors acquire Earth's entire land surface several times during the three-year mission duration in high resolution mode. Thus, the mission offers a suitable data source for up-to-date, homogeneous and globally consistent, high resolution land cover survey as baseline for LULUCF monitoring. An initial global coverage was achieved in 2011 and the second coverage will be completed in 2013. The TanDEM-X data are acquired simultaneously from two spatially separated sensors (TerraSAR-X and TanDEM-X). Together, they form a bistatic SAR system and a SAR interferometer in space, where both satellites receive the signal from a common illuminated footprint under different look angles (Krieger et al., 2007, Krieger et al., 2010). Due to the constellation of two closely flying SAR sensors, one sensor is used as transmitter and receiver (monostatic/active), whereas the other sensor is only a receiver (bistatic/passive). According to Willis (1991), this is called bistatic acquisition.

This acquisition results in a combination of two scattering coefficients. The simultaneous multi-angle view enables the detection of objects not visible in monostatic mode (Walterscheid, Ender, Brenner, & Loffeld, 2006). The angular difference between the two sensors leads to a modification of the received signals depending on the scattering mechanisms and thus provides additional information on the geometric properties of the objects. A number of studies (Dubois-Fernandez et al., 2006, Krieger and Moreira, 2006, Krieger et al., 2010) suggest the hypothesis that the availability of mono- and bistatic scattering coefficients will improve not only the segmentation and classification of land cover classes like urban areas which are mainly affected by dihedral scattering and thus are sensitive to aspect angle, but also natural surfaces like agricultural fields or forests. The difference in the received signal is predominant for dihedral scattering observed from man-made objects which is reduced at the passive (receive only) sensor. This in turn results in more homogenous image statistics which is more favorable for the classification of natural surfaces (Walterscheid et al., 2006). Dubois-Fernandez et al. (2006) detected a modification in backscatter even at a very small angle difference between the two sensors. However, they did not use this information in a land cover classification, but identified the need for further analysis.

For the interferometric generation of a high-quality 3-D surface model from the TanDEM-X mission data the most significant advantage of the bistatic acquisition is the reduction of temporal decorrelation and atmospheric disturbances (Krieger et al., 2007). Therefore, the bistatic interferometric coherence – a measure of the phase decorrelation and the quality of the interferometric derivate (Bamler & Hartl, 1998) – reveals special characteristics compared to the repeat-pass constellation normally used. The interferometric coherence is influenced by the baseline, the Doppler centroid frequency, the system noise, the SAR processing, and relevant scene properties like temporal and volume decorrelation (Hanssen, 2001, Wegmüller and Werner, 1997).

Even for the repeat-pass ERS-1 mission providing only moderate resolution Wegmüller and Werner, 1995, Wegmüller and Werner, 1997 demonstrated the benefit of interferometric coherence in land cover classifications, based on the information of the temporal and volume decorrelation. Following the interferometric coherence in combination with the radar backscatter has been used frequently for land-use classifications (e.g. in Bruzzone et al., 2004, Engdahl and Hyyppa, 2003, Strozzi et al., 2000, Wegmüller and Werner, 1995, Wegmüller and Werner, 1997). Moreover, the use of ERS interferometric coherence for land cover classifications in Sumatra, Indonesia resulted in an increased separability of forest, plantations and deforested areas, whereas certain vegetation classes could not be distinguished (Stussi, Liew, Lim, Nichol, & Goh, 1997). Perko et al. (2011) demonstrated confusions in the repeat-pass interferometric coherence of X-band due to temporal decorrelation of agriculture and forest classes. According to Ribbes et al. (1997), it could be argued that a minimization of the temporal baseline would be beneficial for the use of interferometric coherence in classifications (Ribbes et al., 1997, Stussi et al., 1997). In fact, the two satellites of TanDEM-X have been flying in close formation since October 2010, resulting in a temporal baseline of about a tenth of a second. Thus, temporal decorrelation effects can be neglected (Toraño Caicoya, Kugler, Papathanassiou, & Hajnsek, 2012), e.g. effects due to differences in moisture content or moving objects.

Therefore, the goal of this research is to investigate and exploit the information content of mono-temporal bistatic TanDEM-X data for land cover and forest mapping in a tropical forest. The TanDEM-X data are used to classify basic land cover classes which are in line with the IPCC good practice guidance for land use, land use change and forestry (IPCC, 2003). In addition, the study investigates if the bistatic interferometric coherence improves the separation of open and closed forest, which today is not feasible in deploying monostatic TerraSAR-X data (Santos et al., 2010). The temporal transition from closed to open canopy forest can be seen as an indication of forest degradation and thus, is relevant in the context of REDD +.