| 21 Oct 2026
An Optimized Method for Slope Deformation Monitoring via Parallel-Track InSAR Data Fusion
Abstract. This study addresses geometric distortion in SAR imagery caused by complex topography in reservoir areas, which adversely affects the accuracy of InSAR monitoring. A multi-orbit data fusion technique is proposed, integrating the R-Index(im) with an improved Geometric Distortion (GD) Index. Using the study landslide mass at Danjiangkou Reservoir as a case study, the research leverages Sentinel-1A satellite data. The GD index model is optimized by incorporating a shadow effect coefficient (Sh). Combined with the R-Index(im) 's capability for distortion-region identification, an adaptive weighted fusion model is established. Empirical analysis demonstrates that this approach effectively corrects deformation monitoring biases caused by geometric distortion. Compared to single-orbit data, the fusion results reduce deformation rate errors by 0.02 mm/year and 0.36 mm/year, while expanding the monitoring coverage by 32 %. Verification against GNSS ground-truth data confirms that this method significantly enhances the accuracy and reliability of InSAR monitoring in complex terrain, offering an effective technical solution for reservoir landslide early warning.
The authors combine terrain visibility and geometric distortion indices to merge velocities from two adjacent ascending Sentinel-1 tracks over a reservoir landslide. Incomplete coverage is a common difficulty in steep terrain, and the GNSS observations provide a useful basis for evaluating the results. My main questions concern the LOS conversion, the fusion equations, and the evidence for improved accuracy.
Q1. Section 3.3 states that measurements from both tracks are converted to the primary track's LOS direction, but the conversion equation is not provided. A single LOS measurement does not determine the displacement vector. Please explain how the direction of movement was determined or assumed for this conversion. Please also identify the GNSS component reported in Table 4 and explain how it was projected for comparison with InSAR.
Q2. The eastern dataset ends in 2023, while the western dataset and GNSS observations extend to December 2024. Were the compared rates calculated over the same period? The SBAS processing also needs a brief description.
Q3. The equations and their definitions need checking. The coefficients in Eq. (1) do not match the accompanying text, and the equation does not clearly implement the zero values assigned to shadow and layover. Below Eq. (2), the weighting terms are described as R-index values, although the preceding explanation identifies GD as the basis for weighting. Please also explain the negative GD values in Figure 9 and how they are handled in the calculation. The statement that layover observations are excluded conflicts with the later description of fusion in layover regions.
Q4. At TP01, TP02, and TP11 in Table 4, the fused value falls outside the range of the two reported track values. A weighted average with non-negative weights summing to one should lie between its inputs. Please clarify whether the table reports the values before or after LOS conversion and interpolation, and show the inputs used in Eq. (2).
Q5. Please describe the validation procedure more clearly. The reported errors match the mean absolute errors calculated from the six stations covered by both tracks, but this subset is not identified. Reporting RMSE alongside MAE would give a fuller picture: the improvement over the better individual track is very small in MAE but more apparent in RMSE. Stations covered by only one track should be discussed separately, particularly the large discrepancies at TP07 and TP12. Please also clarify whether GNSS was used only for validation or contributed to correcting the InSAR results.
Q6. The accuracy assessment should rely on comparisons with independent observations. A smoother map, a mean velocity closer to zero, or a smaller spatial standard deviation does not by itself demonstrate improved accuracy. The velocity distribution also reflects real spatial differences in movement. Likewise, agreement with a surface interpolated between a small number of GNSS stations is less informative than direct station comparisons.
Q7. For Figure 14, please explain what evidence supports treating the highlighted uplift as an error rather than actual movement. Please also show how the 32% coverage increase was calculated, distinguishing pixels with valid observations from either track from those filled by Kriging interpolation.
Q8. Wang et al. (2024) already describe GD-based fusion of parallel tracks. Please explain what the present combination adds. A comparison with equal-weight fusion and the original GD weighting would help establish the benefit of the proposed weighting.
Q9. Please identify the exact 12.5 m DEM product, its source elevation data, and its native resolution. The manuscript attributes its terrain detail to L-band penetration, but this needs to be supported by the origin of the elevation product rather than the characteristics of the ALOS sensor alone.
Minor comments: