Qi Zhang

Accurate time series forecasting is fundamental to geohazard early warning, yet remains a major challenge. Conventional in situ geotechnical monitoring remains costly and spatially constrained, whereas deep learning applied to remote sensing data has become increasingly prevalent but often suffers from opacity of model decision-making. To address this issue, we propose a Transformer-based forecasting framework, namely PatchTST-Fusion, adapted for multivariate Multi-Temporal Interferometric Synthetic Aperture Radar (MT-InSAR) time series. The framework integrates model interpretability analysis through TimeSHAP, providing temporal and feature-level attributions across the input sequence. Landslide deformation time series are first derived from Copernicus Sentinel-1 SAR data. Variational Mode Decomposition is then applied to decompose the non-linear signals into trend, seasonal, and noise components. The denoised displacement series are modeled and forecast using the proposed PatchTST-Fusion, which incorporates rainfall and reservoir water level fluctuations as feature-level drivers. Application to the Daping landslide cluster in the Three Gorges Reservoir Area in China demonstrates that our method captures both the long-term and transient non-linear coupling between deformation and its triggers, surpassing state-of-the-art models including CNN-BiGRU-Attention, Informer and original PatchTST with 7–55% improvements in MAE and 10–52% improvements in RMSE. Beyond predictive gains, feature attribution of environmental triggers via TimeSHAP reveals that rainfall and reservoir regulation exert temporally distinct influences on slope kinematics, with high relative importance concentrated in specific periods and characteristic lagged responses. This interpretable framework provides both enhanced forecasting accuracy and process-based insights, offering a broadly applicable tool for landslide early warning in reservoir regions.

Infrared (IR) images inherently face the dual challenges of noise contamination and reduced contrast. However, existing image enhancement methods often overlook the intrinsic correlations between these factors—noise and low contrast—during multistage enhancement processes. Consequently, this oversight leads to a significant reduction in the fidelity of intricate details in IR images. In this article, we present a synergistic IR image enhancement network that simultaneously achieves denoising, contrast improvement, and detail preservation (DCDNet), which breaks down the overall enhancement process into more manageable steps. DCDNet is comprised of a detail awareness unit (DAU), a deep denoising prior (DDP), and a contrast improvement module (CIM). To maintain the details in the IR image, DAU is developed to extract the original detail feature information in DDP and integrate them into the CIM during contrast improvement to improve the final result. The detail information is derived from the encoder of the DDP, which focuses on denoising. The preserved detail features are subsequently incorporated into the decoder of the CIM, which is dedicated to enhancing contrast. Experimental results validate that our proposed approach surpasses other state-of-the-art methods for enhancing IR images in terms of the peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), visual information fidelity (VIF), and performance in downstream tasks. The code and dataset are publicly available at https://github.com/ChickenEating/IR-Enhancement .

The Stepped Frequency Waveform (SFW) is commonly employed in radar technology to synthesize wideband signals through the aggregation of narrow-band pulses, thereby achieving high-resolution range profiles without the need to increase the radar's instantaneous bandwidth. However, the inherent large time-bandwidth product of SFW introduces substantial ranging errors and energy dispersion, which significantly hinders its efficacy in detecting high-speed objects. This paper introduces a pioneering velocity estimation technique utilizing the fractional Fourier transform (FrFT) to address these limitations. Leveraging the characteristic of the Doppler signal from a moving target, which manifests as a chirp signal with a rate proportional to the target's velocity, the FrFT is utilized for precise velocity estimation. Subsequent to this, velocity compensation is applied using the deduced metrics, followed by the application of the inverse fast Fourier transform (iFFT) to pinpoint the target's exact location. To optimize the computational efficiency of determining the FrFT's optimal order, we propose an iterative algorithm founded on the golden section search method. The effectiveness of the proposed approach is verified by simulation data, and the results demonstrate that the proposed approach can accurately estimate the velocity and the range of the high-speed targets with a relatively low computational complexity.

Pre-disaster information storage is crucial for effective disaster response. The discussion regarding deep learning-based Light Detection and Ranging (Lidar) semantic segmentation technology for indoor small items has been ongoing in recent years. However, the methods applicable to large-scale outdoor Lidar datasets for pre-disaster information storage remain limited. This study aims to propose a novel deep learning-based network for city-scale Lidar semantic segmentation to support pre-disaster information storage, called channel attention and normal-based local feature aggregation network (CNLNet). This network is designed to segment common urban land cover objects, including buildings and vegetation. This network incorporates surface normal information and the channel attention mechanism into the RandLA-Net backbone. Ablation studies have been devised to assess the performance of these two features. During the pre-processing step, color information from optical images is fused with Lidar data. The findings demonstrate that CNLNet can enhance the accuracy of the RandLA-Net backbone by improving mIoU at least 1-2%. Including one of these two features also contributes to the backbone’s improved accuracy. Notably, CNLNet outperforms other well-known networks in terms of accuracy with the test of the public Sementic3D dataset. The study further reveals that the proposed network excels in building segmentation, a crucial facet of pre-disaster information storage. Moreover, the results show that spatial resolution, whether at 0.5m or 10m per pixel for optical images, has limited influence on outcomes. One theoretical contribution of this study is the demonstration of the advantages of integrating either surface normal information or a channel attention mechanism to enhance large-scale outdoor Lidar semantic segmentation. Labeled Lidar datasets have been created for training. The practical contribution is that it can optimize disaster response by efficiently facilitating pre-disaster information storage.

Polarimetric Synthetic Aperture Radar Interferometry (Pol-InSAR) based forest height estimation for ecosystem monitoring and management has been developing rapidly in recent years. Spaceborne Pol-InSAR systems with long temporal baselines of several days always lead to severe temporal decorrelation, which can cause a forest height overestimation error. However, most forest height estimation studies have not considered the change in dielectric property as a factor that may cause temporal decorrelation with a long temporal baseline. Therefore, it is necessary to propose a new method that considers dielectric fluctuations and random motions of scattering elements to compensate for the temporal decorrelation effect. The lack of ground truth for forest canopy also needs a solution. Unsupervised methods could be a solution because they do not require the use of true values of tree heights as the ground truth to calculate their estimation accuracies. This paper aims to present an unsupervised forest height estimation method called Dielectric Fluctuation and Random Motion over Ground (DF-RMoG) to improve accuracy by considering the dielectric fluctuations and random motions. Its performance is investigated using Advanced Land Observing Satellite (ALOS)-1 Pol-InSAR data acquired over a German forest site with temporal intervals of 46 and 92 days. The authors analyze the relationship between forest height and different parameters with DF-RMoG and conventional models. Compared with conventional models, the proposed DF-RMoG model significantly reduces the overestimation error due to temporal decorrelation in forest height estimation according to its lowest average forest height.

This paper proposes an improved model-based forest height inversion method for airborne L-band dual-baseline repeat-pass polarimetric synthetic aperture radar interferometry (PolInSAR) collections. A two-layer physical model with various volumetric scattering attenuation and dynamic motion properties is first designed based on the traditional Random Motion over Ground (RMoG) model. Related PolInSAR coherence functions with both volumetric and temporal decorrelations incorporated are derived, where the impacts of homogenous and heterogeneous attenuation and dynamic motion properties on the performance of forest height inversion were investigated by the Linear Volume Attenuation (LVA), Quadratic Volume Attenuation (QVA), Linear Volume Motion (LVM), and Quadratic Volume Motion (QVM) depictions in the volume layer. Dual-baseline PolInSAR data were acquired to increase the degree of freedom (DOF) of the coherence observations and thereby provide extra constraints on the forest parameters to address the underdetermined problem. The experiments were carried out on a boreal forest in Canada and a tropical one in Gabon, where physical models with LVA + QVM (RMSE: 3.56 m) and QVA + LVM (RMSE: 6.83 m) exhibited better performances on the forest height inversion over the boreal and tropical forest sites, respectively. To leverage the advantages of LVA, QVA, LVM, and QVM, a pixel-wise optimization strategy was used to obtain the best forest height inversion performance for the range of attenuation and motion profiles considered. This pixel-wise optimization surpasses the best-performing single model and achieves forest height inversion results with an RMSE of 3.21 m in the boreal forest site and an RMSE of 6.48 m in the tropical forest site.