Sai Wang
Research Subjects
Research Interests
Sai Wang has a total of 21 co-authors. Here are the top co-authors by number of shared publications:
- Haiqiang Fu
School of Geosciences and Info-Physics, Central South University, Changsha 410083, China
2 shared publications - Chenglong Cao
School of Geosciences and Info-Physics and the Key Laboratory of Metallogenic Prediction of Nonferrous Metals and Geological Environment Monitoring, Ministry of Education, Central South University, Changsha, China
1 shared publication - Chuanjun Wu
State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan, China
1 shared publication - Cristina Gómez
IuFOR, EiFAB, Universidad de Valladolid
1 shared publication - Cui Zhou
School of Advanced Interdisciplinary Studies, Central South University of Forestry and Technology, Changsha 410004, China
1 shared publication Gang Chen
Xian Research Institute of Surveying and Mapping, Xian, China
1 shared publication
Latest Publications
- Open Access
- 11 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. 19 (19391404) , 10096-10108
A Novel Two-Stage Approach to Improve Forest Height Estimation From SAR Backscatter in Sloping Terrain
The use of Synthetic Aperture Radar (SAR) to monitor forest biophysical parameters is a significant focus of remote sensing research. However, in mountainous areas with complex terrain, topographic variations can distort the SAR backscatter signal, greatly constraining the accuracy and reliability of forest height retrieval. To address this challenge, this paper proposes a novel two-stage framework for estimating forest height from SAR backscatter intensity that minimizes terrain effects. First, an illuminated area correction method, which accounts for the geometric discrepancies between ground and SAR slant-range coordinates, is employed to retrieve the backscatter coefficient corresponding to the real sloping terrain area. Second, the angular dependence of the forest backscatter is incorporated into the vertical structure modeling, and subsequently, the forest height is extracted using a structured empirical model. To validate the effectiveness of the proposed method, experiments were conducted using two UAVSAR L band datasets from opposing imaging geometries, acquired over the Nantahala National Forest within the Appalachian Mountains. LVIS LiDAR data served as the reference height for accuracy validation. The experimental results demonstrate that the proposed two-stage method performs robustly on both the ascending and descending datasets, improving the forest height estimation accuracy by 33.9% and 21.3%, respectively, compared to the traditional method ignoring the terrain effects.
- 11 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. 19 (19391404) , 10096-10108
Open AccessA Novel Two-Stage Approach to Improve Forest Height Estimation From SAR Backscatter in Sloping Terrain
The use of Synthetic Aperture Radar (SAR) to monitor forest biophysical parameters is a significant focus of remote sensing research. However, in mountainous areas with complex terrain, topographic variations can distort the SAR backscatter signal, greatly constraining the accuracy and reliability of forest height retrieval. To address this challenge, this paper proposes a novel two-stage framework for estimating forest height from SAR backscatter intensity that minimizes terrain effects. First, an illuminated area correction method, which accounts for the geometric discrepancies between ground and SAR slant-range coordinates, is employed to retrieve the backscatter coefficient corresponding to the real sloping terrain area. Second, the angular dependence of the forest backscatter is incorporated into the vertical structure modeling, and subsequently, the forest height is extracted using a structured empirical model. To validate the effectiveness of the proposed method, experiments were conducted using two UAVSAR L band datasets from opposing imaging geometries, acquired over the Nantahala National Forest within the Appalachian Mountains. LVIS LiDAR data served as the reference height for accuracy validation. The experimental results demonstrate that the proposed two-stage method performs robustly on both the ascending and descending datasets, improving the forest height estimation accuracy by 33.9% and 21.3%, respectively, compared to the traditional method ignoring the terrain effects.
- Open Access
- 10 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64 (01962892) , 1-20
Enhanced Sub-Canopy Topography Retrieval Method Using Dual-Baseline TanDEM-X InSAR Data
Accurately measuring topography in forested areas using single-baseline InSAR data is challenging due to volume scattering effects and a limited number of observations. To address these limitations, we propose a new dual-baseline InSAR framework for retrieving sub-canopy topography utilizing two bistatic TanDEM-X InSAR observations. Our approach adopts the Random Volume over Ground (RVoG) model to estimate two spatially varying terms: ground-to-volume amplitude ratio and forest canopy height. Using these parameters we then estimate the phase center height (PCH), representing the elevation difference between the conventional InSAR-based digital elevation model (DEM, a numerical representation of terrain elevations) and the true ground surface. By subtracting the estimated PCH from the conventional InSAR DEM provided by each baseline, our method produces two individual sub-canopy topography products. Further refinement is achieved through a fusion process that integrates the sub-canopy topography maps derived from different baselines, enhancing the accuracy of the output. To validate the proposed framework, we conducted experiments in Sweden and Spain, across diverse forest types, climates, and terrain conditions. The accuracy of the estimated sub-canopy topography was assessed against airborne lidar data. At the Krycklan site in Sweden, our method achieved a root-mean-square error (RMSE) of 2.04 m, demonstrating a performance improvement of 60% and 70% over conventional InSAR, depending on the baselines used. In Spain, the Vegaquemada and Covaleda sites yielded RMSEs of 3.44 m and 4.70 m, respectively, surpassing conventional InSAR DEMs for forested regions as well as simple single-baseline approaches which leverage space-borne LiDAR to fit model parameters, though with some influence from terrain slope. This approach has the potential to support largescale sub-canopy topography with InSAR data.
- 10 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64 (01962892) , 1-20
Open AccessEnhanced Sub-Canopy Topography Retrieval Method Using Dual-Baseline TanDEM-X InSAR Data
Accurately measuring topography in forested areas using single-baseline InSAR data is challenging due to volume scattering effects and a limited number of observations. To address these limitations, we propose a new dual-baseline InSAR framework for retrieving sub-canopy topography utilizing two bistatic TanDEM-X InSAR observations. Our approach adopts the Random Volume over Ground (RVoG) model to estimate two spatially varying terms: ground-to-volume amplitude ratio and forest canopy height. Using these parameters we then estimate the phase center height (PCH), representing the elevation difference between the conventional InSAR-based digital elevation model (DEM, a numerical representation of terrain elevations) and the true ground surface. By subtracting the estimated PCH from the conventional InSAR DEM provided by each baseline, our method produces two individual sub-canopy topography products. Further refinement is achieved through a fusion process that integrates the sub-canopy topography maps derived from different baselines, enhancing the accuracy of the output. To validate the proposed framework, we conducted experiments in Sweden and Spain, across diverse forest types, climates, and terrain conditions. The accuracy of the estimated sub-canopy topography was assessed against airborne lidar data. At the Krycklan site in Sweden, our method achieved a root-mean-square error (RMSE) of 2.04 m, demonstrating a performance improvement of 60% and 70% over conventional InSAR, depending on the baselines used. In Spain, the Vegaquemada and Covaleda sites yielded RMSEs of 3.44 m and 4.70 m, respectively, surpassing conventional InSAR DEMs for forested regions as well as simple single-baseline approaches which leverage space-borne LiDAR to fit model parameters, though with some influence from terrain slope. This approach has the potential to support largescale sub-canopy topography with InSAR data.
- 3 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 63 (01962892) , 1-17
Height-Constrained Planimetric Geolocation Error Compensation for Elevation Refinement After InSAR-DSM Block Adjustment
Large-scale accurate digital surface model (DSM) generation is a key application of interferometric synthetic aperture radar (InSAR). Further refining the elevation by compensating the planimetric geolocation error after InSAR-DSM block adjustment remains challenging. To address this, a compensation method leveraging readily available elevation data was proposed. Relative offsets between the two DSMs were transformed into offsets relative to the reference DSM, followed by enhanced DSM mosaicking. For the first time, an optimization method was integrated with statistical adjustment (SA) for absolute offset determination. Experiments with TanDEM-X InSAR datasets near Guangzhou, China, demonstrate that (1) the Hybrid norm was identified as the optimal objective function, achieving the highest accuracy (total bias RMSE: 0.005 pixel) when combined with SA. (2) Incorporating InSAR quality parameters and border handling enhanced fusion accuracy, while sorting DSMs by ascending Euclidean distance significantly reduced memory usage. (3) The 90% relative elevation bias in along-track DSM overlaps was reduced to 1.51 m (an 82% improvement). The terrain-related elevation bias was corrected, yielding a mean and RMSE of 0.09 m and 3.18 m, respectively. The compensable offset range was approximately 0.001–123 pixels. These results demonstrate the effectiveness and applicability of the proposed method.
- 3 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 63 (01962892) , 1-17
Height-Constrained Planimetric Geolocation Error Compensation for Elevation Refinement After InSAR-DSM Block Adjustment
Large-scale accurate digital surface model (DSM) generation is a key application of interferometric synthetic aperture radar (InSAR). Further refining the elevation by compensating the planimetric geolocation error after InSAR-DSM block adjustment remains challenging. To address this, a compensation method leveraging readily available elevation data was proposed. Relative offsets between the two DSMs were transformed into offsets relative to the reference DSM, followed by enhanced DSM mosaicking. For the first time, an optimization method was integrated with statistical adjustment (SA) for absolute offset determination. Experiments with TanDEM-X InSAR datasets near Guangzhou, China, demonstrate that (1) the Hybrid norm was identified as the optimal objective function, achieving the highest accuracy (total bias RMSE: 0.005 pixel) when combined with SA. (2) Incorporating InSAR quality parameters and border handling enhanced fusion accuracy, while sorting DSMs by ascending Euclidean distance significantly reduced memory usage. (3) The 90% relative elevation bias in along-track DSM overlaps was reduced to 1.51 m (an 82% improvement). The terrain-related elevation bias was corrected, yielding a mean and RMSE of 0.09 m and 3.18 m, respectively. The compensable offset range was approximately 0.001–123 pixels. These results demonstrate the effectiveness and applicability of the proposed method.