Compared with the double-difference relative positioning method, the precise point positioning (PPP) algorithm can avoid the selection of a static reference station and directly measure the three-dimensional position changes at the observation site and exhibit superiority in a variety of deformation monitoring applications. However, because of the influence of various observing errors, the accuracy of PPP is generally at the cm-dm level, which cannot meet the requirements needed for high precision deformation monitoring. For most of the monitoring applications, the observation stations maintain stationary, which can be provided as a priori constraint information. In this paper, a new PPP algorithm based on a sliding window was proposed to improve the positioning accuracy. Firstly, data from IGS tracking station was processed using both traditional and new PPP algorithm; the results showed that the new algorithm can effectively improve positioning accuracy, especially for the elevation direction. Then, an earthquake simulation platform was used to simulate an earthquake event; the results illustrated that the new algorithm can effectively detect the vibrations change of a reference station during an earthquake. At last, the observed Wenchuan earthquake experimental results showed that the new algorithm was feasible to monitor the real earthquakes and provide early-warning alerts.

Accurate and efficient crop classification is fundamental to smart agriculture. Due to the limitations of optical remote sensing in cloudy and rainy regions, time-series Synthetic Aperture Radar (SAR) data has become the primary source for crop classification, owing to its all-weather capabilities and sensitivity to crop phenology. However, most existing classification algorithms are restricted to using either temporal or spatial features alone, hindering the full exploitation of the spatial structures and temporal scattering characteristics embedded in time-series SAR data. This study proposes a CNN-BiLSTM model that combines the strengths of Convolutional Neural Networks (CNN) in spatial feature extraction with the capability of Bidirectional Long Short-Term Memory (BiLSTM) networks to capture long-term temporal dependencies, achieving deep spatiotemporal coupling of time-series SAR data. The proposed approach overcomes the limitations of traditional models that rely on single-feature utilization, constructing a more comprehensive feature representation for crop classification. A case study was conducted using nine Sentinel-1A images from the main growing season (March–June 2023) in Leizhou City, Guangdong Province. Experimental results demonstrate that the CNN-BiLSTM model significantly outperforms Support Vector Machine (SVM), CNN, and BiLSTM models in crop classification, achieving an overall accuracy of 95.8% and a Kappa coefficient of 0.94. This approach provides a novel solution for crop classification in cloudy and rainy regions and offers important applications for precise monitoring and management in smart agriculture.