Search

This article demonstrates how 3-D vegetation structure can be approximated by interferometric synthetic aperture radar (InSAR) backscatter-height histograms. Single-look backscatter measurements are plotted against the InSAR phase height and are aggregated spatially over a forest patch to form a 3-D histogram, referred to as InSAR backscatter-height histogram or simply InSAR histogram. InSAR histograms resemble LiDAR waveforms, suggesting that existing algorithms used to retrieve canopy height and ground topography from radar tomograms or LiDAR waveforms can be applied to InSAR histograms. Three algorithms are evaluated to generate maps of digital terrain, surface, and canopy height models: Gaussian decomposition, quantile, and backscatter threshold. Full-polarimetric L-band uninhabited aerial vehicle synthetic aperture radar (UAVSAR) data collected over the Gabonese Lopé National Park during the 2016 AfriSAR campaign are used to illustrate and compare the performance of the algorithms for the HH, HV, VV, HH+VV, and HH−VV polarimetric channels. Results show that radar-derived maps using the InSAR histograms differ by 4 m (top-canopy), 5 m (terrain), and 6 m (forest height) in terms of average root-mean-square errors (RMSEs) from standard maps derived from full-waveform laser, vegetation, and ice sensor (LVIS) LiDAR measurements. [ABSTRACT FROM AUTHOR]

This paper presents a new damage-mapping algorithm based on coherence images estimated from multitemporal polarimetric-interferometric synthetic aperture radar (SAR) data. The interferometric coherence has been restricted in the conventional damage-mapping approaches because the decorrelation sources are too complicated to interpret accurately and temporal decorrelation effects caused by slowly occurring natural changes and disaster events are often coupled together. To overcome these limitations, we formulate a coherence model that accounts for temporal decorrelation in two simplified layers, ground and volume layers, for long-temporal repeat-pass scenarios with zero spatial baseline. The model parameters include: 1) ground-to-volume ratio, a factor to determine the relative scattering contribution of ground and volume layers; 2) temporally correlated change, which captures the exponentially decaying behavior of coherence with time; and 3) temporally uncorrelated change, which is associated with random temporal changes. We estimate the model parameters in three steps: coherence optimization, interferometric pair-invariant parameter estimation, and interferometric pair-variant parameter estimation. To isolate the effects of disaster events from background natural changes, we calculate the probability density functions of historical change pixel by pixel and produce a probability map of damage. We tested the algorithm with uninhabited aerial vehicle data acquired from 2009 to 2015 for mapping the area damaged by the 2015 Lake Fire in California. Based on performance evaluation using receiver operating characteristic curves for optimized coherences and averaged probability maps, the proposed method reduced the false alarm from 0.25 to 0.07 when the probability of detection was 0.85 compared to coherence products alone. [ABSTRACT FROM AUTHOR]

Detection of changes caused by major events—such as earthquakes, volcanic eruptions, and floods—from interferometric synthetic aperture radar (SAR) data is challenging because of the coupled effects with temporal decorrelation caused by natural phenomena, including rain, snow, wind, and seasonal changes. The coupled effect of major events and natural phenomena sometimes leads to misinterpretation of interferometric coherence maps and often degrades the performance of change detection algorithms. To differentiate decorrelation sources caused by natural changes from those caused by an event of interest, we formulated a temporal decorrelation model that accounts for the random motion of canopy elements, temporally correlated dielectric changes, and temporally uncorrelated dielectric changes of canopy and ground. The model parameters are extracted from the interferometric pairs associated with natural changes in canopy and ground using the proposed temporal decorrelation model. In addition, the cumulative distribution functions of the temporally uncorrelated model parameters, which are associated with natural changes in canopy and ground, are estimated from interferometric pairs acquired before the event. Model parameters are also extracted from interferometric SAR data acquired across the event and compared with the cumulative probabilities of natural changes in order to calculate the probability of a major event. Subsequently, pixels with cumulative probabilities greater than 75% are marked as changed due to the event. A case study for detecting volcanic ash during the eruption of the Shinmoedake volcano in January 2011 was carried out using L-band Advanced Land Observation Satellite PALSAR data. [ABSTRACT FROM AUTHOR]

This paper addresses the important yet unresolved problem of estimating forest properties from polarimetric-interferometric radar images affected by temporal decorrelation. We approach the problem by formulating a physical model of the polarimetric-interferometric coherence that incorporates both volumetric and temporal decorrelation effects. The model is termed random-motion-over-ground (RMoG) model, as it combines the random-volume-over-ground (RVoG) model with a Gaussian-statistic motion model of the canopy elements. Key features of the RMoG model are: 1) temporal decorrelation depends on the vertical structure of forests; 2) volumetric and temporal coherences are not separable as simple multiplicative factors; and 3) temporal decorrelation is complex-valued and changes with wave polarization. This third feature is particularly important as it allows compensating for unknown levels of temporal decorrelation using multiple polarimetric channels. To estimate model parameters such as tree height and canopy motion, we propose an algorithm that minimizes the least square distance between model predictions and complex coherence observations. The algorithm was applied to L-band NASA's Uninhabited Aerial Vehicle Synthetic Aperture Radar data acquired over the Harvard Forest (Massachussetts, USA). We found that the RMS difference at stand level between estimated RMoG-model tree height and NASA's lidar Laser Vegetation and Ice Sensor tree height was within 12% of the lidar-derived height, which improved significantly the RMS difference of 37% obtained using the RVoG model and ignoring temporal decorrelation. This result contributes to our ability of estimating forest biomass using in-orbit and forthcoming polarimetric-interferometric radar missions. [ABSTRACT FROM PUBLISHER]