Jeanine Engelbrecht

Parameters Affecting Interferometric Coherence—The Case of a Dynamic Agricultural Region

In South Africa, the need exists for a long-term monitoring system to detect deformation because of mining. Differential radar interferometry (dInSAR) is recognized for its ability to detect deformation without the need for extensive field observations. However, phase noise (assessed by interferometric coherence) is known to adversely affect interferometric measurements. In this paper, 12 polarimetric RADARSAT-2 SAR images are used to statistically analyze the sensitivity of coherence to perpendicular baseline, Doppler centroid (DC) difference, temporal baseline, land surface evolution, and polarization. The results suggest that the average scene coherence for HH and VV polarizations is most sensitive to the effects of temporal baseline if RADARSAT-2 data with small perpendicular baselines and DC differences are considered. However, an increase in the sensitivity to dc difference and perpendicular baseline is observed after analysis of ERS-2 data for which higher values of these parameters are available. The choice of optimal polarization is seasonally dependent, with VV polarization data being more suitable at the end of the growing season after harvesting, while HH polarization is more suitable during the peak of the growing season. Image pairs with at most 24-day temporal baselines are required during the peak of the growing season while successful interferogram generation is possible with temporal baselines of up to 120 days after the peak of the growing season. Subsidence basins are successfully detected, demonstrating that the dInSAR techniques would be suitable for the long-term monitoring of surface subsidence in this dynamic commercial agricultural environment.

A Simple Normalized Difference Approach to Burnt Area Mapping Using Multi-Polarisation C-Band SAR

In fire-prone ecosystems, periodic fires are vital for ecosystem functioning. Fire managers seek to promote the optimal fire regime by managing fire season and frequency requiring detailed information on the extent and date of previous burns. This paper investigates a Normalised Difference α-Angle (NDαI) approach to burn-scar mapping using C-band data. Polarimetric decompositions are used to derive α-angles from pre-burn and post-burn scenes and NDαI is calculated to identify decreases in vegetation between the scenes. The technique was tested in an area affected by a wildfire in January 2016 in the Western Cape, South Africa. The quad-pol H-A-α decomposition was applied to RADARSAT-2 data and the dual-pol H-α decomposition was applied to Sentinel-1A data. The NDαI results were compared to a burn scar extracted from Sentinel-2A data. High overall accuracies of 97.4% (Kappa = 0.72) and 94.8% (Kappa = 0.57) were obtained for RADARSAT-2 and Sentinel-1A, respectively. However, large omission errors were found and correlated strongly with areas of high local incidence angle for both datasets. The combined use of data from different orbits will likely reduce these errors. Furthermore, commission errors were observed, most notably on Sentinel-1A results. These errors may be due to the inability of the dual-pol H-α decomposition to effectively distinguish between scattering mechanisms. Despite these errors, the results revealed that burnt areas could be extracted and were in good agreement with the results from Sentinel-2A. Therefore, the approach can be considered in areas where persistent cloud cover or smoke prevents the extraction of burnt area information using conventional multispectral approaches.

Detection of Sinkhole Precursors Through SAR Interferometry: Radar and Geological Considerations

Sinkholes are an unpredictable geohazard that endanger life and property in dolomitic terrains. Sinkholes are a significant threat in Gauteng, South Africa’s most populated and urbanized province. Small-scale surface subsidence is frequently present prior to the collapse of a sinkhole. Therefore, the presence of precursory surface deformation can be exploited to develop early warning systems. Spaceborne synthetic aperture radar (SAR) differential interferometry (DInSAR) is able to monitor small-scale surface deformation over large areas and can be used to detect and measure precursors to sinkhole development. This letter investigates the use of repeat-pass DInSAR to detect sinkhole precursors in the Gauteng province. Twenty stripmap acquisitions from TerraSAR-X were acquired over a full year. DInSAR results revealed the presence of three previously unknown deformation features, one of which could be confirmed by subsequent field investigations. Furthermore, a water supply pipeline ruptured six months after the initial observation. The detection of the deformation, therefore, provided a viable early warning to landowners who were unaware of the subsidence. Detected deformation features were between 40 and 100 m in diameter. The maximum displacement measured was 50 mm over 55 days. Despite the successful detection, seven sinkhole events occurred in the observation period, for which no deformation could be detected. The results indicate that high-resolution X-band interferometry is able to monitor dolomite-induced instability in an urban environment. However, considerations related to SAR interferometry and physical sinkhole properties need to be addressed before DInSAR can be used in an operational early warning system.

Recommendations for long-term operational dinsar monitoring of mining-induced deformation in a dynamic agricultural region

Underground mining activities are associated with the potential for surface deformation. DInSAR techniques are known for its ability to successfully measure and monitor deforming areas. However, the limitations associated with this technology are mostly related to signal decorrelation effects which are introduced as a result of the surface and sensor characteristics. In this paper we examine the operational limitations of this technology in the presence of disturbances to the reflected signals, due to the dynamic agricultural nature of the sites under test. Three sources of SAR data captured at X-band, C-band and L-band respectively was considered. The results revealed that changes in the vegetative conditions affected C- and X-band data (as might be expected) more severely, although changes in surface conditions induced by tilling and harvesting also affected L-band data. C-band data could not measure deformation maxima whilst L-band data showed reduced sensitivity to deformation minima.

Coherence Optimization and Its Limitations for Deformation Monitoring in Dynamic Agricultural Environments

Differential interferometry techniques are well known for its ability to provide centimeter to millimeter scale deformation measurements. However, in natural and agricultural areas, the presence of vegetation and the evolution of the land surface introduce a phase noise component which limits successful interferometric measurement. This paper aims to address the known limitations of traditional dInSAR in the presence of disturbances to reflected signals due to agricultural activities by testing the polInSAR technique for its ability to increase interferometric coherence and to detect surface movement in the areas of interest. Both fully polarimetric RADARSAT-2 and ALOS PALSAR data were subject to coherence optimization using the multiple scattering mechanism (MSM) approach. For C-band RADARSAT-2 data, coherence optimization resulted in a statistically significant increase in interferometric coherence. However, the spatial heterogeneity of the scattering process and how it changes over time caused random phase changes associated with temporal baseline effects and the evolution of the land surface. These effects could not be removed from C-band interferograms using the MSM approach. Therefore, coherence optimization resulted in an increase in the random speckle in interferograms reducing the ability to derive high confidence interferometric measurements, indicating a drawback in the MSM approach to coherence optimization. On the other hand, coherence optimization on L-band data demonstrated an increase in the spatial homogeneity of the speckle noise suggesting that the MSM approach to coherence optimization on L-band data may be more successful in enhancing the ability to extract deformation measurements in dynamic agricultural regions. In general, a good agreement in deformation measurements derived from dInSAR and polInSAR techniques was observed.

The phenology of an agricultural region as expressed by polarimetric decomposition and vegetation indices

The monitoring of vegetation biophysical characteristics in agricultural regions has traditionally been achieved using visible and infrared data. However, limitations to long-term monitoring are associated with cloud-cover. In this paper we consider the use of fully polarimetric SAR data and H-A-α decomposition to characterize the link between the different phenological stages of crops and the change in their scattering behaviour over time. The results demonstrated that the analysis of scattering mechanisms observed on fully polarimetric data can provide information regarding the peak of the growing season, the time-frame associated with harvesting and the monitoring of regrowth of crops. Additionally, changes in the roughness conditions of soils associated with harvesting, tilling and planting was observed. This suggests that polarimetric data is complementary to traditional vegetation indices derived from visible and near infrared sensors.

A normalised difference alpha-angle approach to burn scar extraction on multiple-polarisation sar data

Wild fires are often disastrous events, damaging infrastructure and the environment. However, in fire-prone ecosystems, periodic fires are vital for ecosystem functioning. Fire managers in these regions seek to promote the optimal fire regime by managing fire season and fire frequency. For this purpose, detailed information on the extent and date of previous burns over large areas are required. This paper proposes a Normalised Difference α-Angle approach to burn scar mapping using α-angles derived from a pre-burn scene and a post-burn RADARSAT-2 scene. A threshold-based segmentation approach was used to extract the burnt-area extent while minimizing speckle effects. The results of the burn scar extraction were compared to the extent of a burn scar derived from Sentinel-2 data. The results suggest that, in general, a good agreement between the two approaches was obtained with an overall accuracy of 0.98 achieved. High errors of omission was associated with areas of high local incidence angle. In general, the result suggests that SAR data, analysed using the proposed approach, can be used in areas where persistent cloud cover or smoke from active fires prevents the use of conventional approaches.

Predicting modis EVI from SAR parameters using random forests algorithms

Vegetation indices derived from optical and multispectral data, such as Enhanced Vegetation Index (EVI), have been widely used to monitor the health and productivity of vegetation. However, their use is limited in areas with persistent cloud-cover and inclement weather. Due to the all-weather observations provided by Synthetic Aperture Radar (SAR) sensors, an alternative would be to derive vegetation descriptors from SAR data. This study investigates the use of C-band SAR parameters to simulate EVI derived from optical and multispectral data using Random Forest algorithms. The results suggest that the relationship between SAR observables and EVI was weak although the radar backscatter in the HV and HH polarisation as well as Van Zyl volume scattering are potential predictors for EVI.

Evidence of roof collapse detected on South African coal mines using sentinel-1 interferometry

Surface subsidence associated with underground coal mining is a known concern due to the associated environmental as well as health and safety risks. This paper evaluates the potential of Sentinel-1 repeat-pass differential interferometry for monitoring mining-induced deformation. The area under investigation is a coal mining region in South Africa where pillar-extraction is taking place. The pillar extraction causes a weakening of pillars leading to roof collapse and subsequent deformation at the surface. The results of Sentinel-1 dInSAR analysis suggest that several deformation features could be observed during almost 2-years of monitoring and the evolution and progression of deformation basins over time could be studied. The results further indicate that operational monitoring of mining-induced deformation by Sentinel-1 interferometry would be possible assuming data continuity can be ensured. Since many mines are considering pillar-extraction as the end-of-life of the mine is reached, continued Sentinel-1 dInSAR observations can be incorporated into operational monitoring programmes.

Coherence optimisation and its limitations for deformation monitoring in agricultural regions

Differential interferometry techniques are well known for their ability to provide cm to mm scale deformation measurements. However, in natural and agricultural areas, the presence of vegetation and the evolution of the land surface introduce a phase noise component which limits successful interferometric measurement. In underground mining environments, operational monitoring of surface deformation will be limited due to these temporal decorrelation effects. This paper aims to address the known limitations of traditional dInSAR in the presence of disturbances to reflected signals due to agricultural activities by testing the polInSAR technique for its ability to increase interferometric coherence and to detect surface movement in the areas of interest. Both fully polarimetric RADARSAT-2 and ALOS PALSAR data were subject to coherence optimisation using the Multiple Scattering Mechanism approach. For C-band RADARSAT-2 data, coherence optimisation resulted in a statistically significant increase in interferometric coherence. However, the spatial heterogeneity of the scattering process and how it changes over time caused random phase changes associated with temporal baseline effects and the evolution of the land surface. These effects could not be removed from C-band interferograms using the polInSAR approaches. Therefore, coherence optimization resulted in an increase in the random speckle in interferograms reducing the ability to derive high confidence interferometric measurements. On the other hand, coherence optimization on L-band data demonstrated an increase in the spatial homogeneity of the speckle noise indicating that coherence optimization on L-band data may be more successful in enhancing the ability to extract deformation measurements in dynamic agricultural regions.