ABSTRACT The advent of unmanned aerial vehicles (UAVs) has significantly advanced precision agriculture by providing scalable, high-resolution imagery for farmland monitoring. However, the generation of descriptive captions from UAV images for agricultural applications remains an underexplored area of research. This study addresses this gap by introducing a prioritized retrieval model designed to generate captions that characterize crop growth stages from aerial imagery acquired by UAVs. In the absence of an existing benchmark, a dedicated Crop Growth Image Captions Dataset was developed to facilitate experimentation. The quality of the captions produced by the proposed model, in comparison with contemporary methods, was assessed using both objective evaluation metrics and subjective criteria. By leveraging UAV-based imagery, the proposed approach contributes towards automation in crop assessment, thereby mitigating the limitations of conventional, labour-intensive techniques and enhancing efficiency and accuracy in agricultural monitoring.

ABSTRACT Currently, a variety of land surface temperature (LST) products generated from thermal infrared bands have been already accumulated. Compared to the thermal infrared band, the mid-infrared band exhibits higher transmittance and greater robustness under humid atmospheric conditions, offering potential for further improving LST retrieval accuracy. However, the mid-infrared band presents larger variability in emissivity and higher estimation difficulty, limiting the effectiveness of LST retrieval using mid-infrared remote sensing data sources. This study proposed a MODIS night-time mid-infrared LST retrieval algorithm that integrates reflectance spectral characteristics to estimate emissivity. A few-shot machine learning model was established to build the correlation between MODIS optical band reflectance and mid-infrared band emissivity within a simulated dataset accounting for mixed spectral components, then applied to real observational data. Validation results from SURFRAD ground stations indicate an overall RMSE of 2.5521 K for this new algorithm, with values of 2.5558 K under dry atmospheric conditions and 2.5021 K under humid atmospheric conditions. The new algorithm can accurately retrieve night-time LST without significant error increasing as atmospheric water vapor content rises. Future work will further study on fields including eliminating daytime solar radiance effects, conducting multi-surface-type validation, and reducing dependence on external parameters.

ABSTRACT Remote sensing increasingly targets environmental processes that are nonstationary, intermittent, and incompletely observable. Conventional microwave and LiDAR processing assumes uniform temporal sampling, stationary statistics, and fixed model structures, limiting sensitivity to abrupt transitions and evolving spatial-temporal patterns. Building on natural time analysis and ecoinformatics, this paper presents a radical idea with an information-centred methodological framework that combines event-driven temporal encoding, entropy measures, and evolving spatial statistics within established microwave radiative transfer and LiDAR geometric models. This reformulates the microwave brightness temperature series using event-based entropy to enhance the detection of regime shifts. Meanwhile, LiDAR point clouds are analysed by considering terrain models through spatial entropy and variogram evolution to characterize structural changes. Validations using representative datasets show that this proposed method is more robust than conventional analyses in the presence of noise, irregular sampling, and data gaps. Quantitatively, the detection lead time improved by 15–25%, and robustness was maintained under up to 20% data loss. The method is compatible with satellite, airborne, and terrestrial sensing systems and can be integrated into operational workflows, subject to dataset-specific calibration and parameter tuning.

ABSTRACT Precipitation influenced by altered cloud characteristics is primarily due to anthropogenic climate change. This study investigates Northern (NR) and Southern (SR) regions of India, to delineate Monsoonal heterogeneity through cloud properties using satellite data (2002-2024), statistical and machine learning models. Cloud parameters show rising trend. Rate of rise differs substantially for cloud liquid water path (CLW) during 2013–2022, with a relatively low rate in NR (0.62 ± 0.20 g/m2/decade) and significantly larger (1.58 ± 0.29 ) in SR indicating complex interactions between topography and atmospheric dynamics modulating cloud characteristics. Changes in cloud effective radius (CER) and cloud optical thickness (COT) across the latitudes could be due to spatial gradients in anthropogenic aerosol loading and associated meteorological influences in NR. Cloud clustering in NR has shown distinct structures (silhouette score ≈0.39), with one of clusters exhibiting high CER, COT and CLW, implying presence of active convective regimes and persistent congestus clouds. SR showed less distinct clustering, marking consistent transition towards higher values from 2002–2012 to 2013–2024. This distinction may be due to enhanced convective intensity driven by changes in atmospheric moisture and large-scale dynamics. This study advances comprehension of regional variation of monsoon dynamics impacted by climate change.

ABSTRACT The novelty of this letter lies in preliminarily analysis of the wave mapping mechanism using Gaofen-3 (GF-3) acquired in extended wide (EW) mode at low incidence angle of 10–20°. In total, 35 Chinese GF-3 images in vertical-vertical (VV) were collocated with hindcasted wave spectra from the numeric model WAVEWATCH-III (WW3). Validation of WW3-simulated significant wave heights (SWHs) against the Haiyang-2 (HY-2) altimeter wave products yields a root mean squared error (RMSE) of 0.44 m with a correlation coefficient (r) of 0.94 and a scatter index (SI) of 0.19, confirming the reliability of the WW3 simulations for this study. Subsequently, the synthetic aperture radar (SAR) intensity spectrum is simulated by multiplying the hindcasted wave spectrum by three modulation transfer functions (MTFs), i.e. tilt, hydrodynamic modulation, and velocity bunching. Based on experimental results, the spectrum correlation coefficient (Cor) and the spectrum squared error (E) of total energy between SAR intensity spectra and simulations using three MTFs are 0.94 and 7.79, respectively. These values represent significant large compared to other MTF combinations, which yield higher errors (approximately 6 E) and lower correlations (around 0.91 Cor). Therefore, it is concluded that velocity bunching has less influence on SAR mapping mechanism at low incidence angle, which is further supported by the Shapley Additive exPlanations (SHAP) analysis performed using eXtreme Gradient Boosting (XGBoost). This study establishes the foundation of wave retrieval from SAR image at low incidence angle.

ABSTRACT The requirement for creating a robust and efficient earthquake warning system is to estimate the parameters of the upcoming earthquake (location, time and magnitude) with low uncertainty. This goal is achieved by increasing the number of observational earthquake precursors. Today, satellite remote sensing has made it possible to observe the behaviour of a large number of LAI (Lithospheric Atmospheric Ionospheric) precursors. Next, it is necessary to estimate the final desired earthquake parameters by using different fusion methods and combining the earthquake parameters estimated by different precursors. In this study, using the Kalman filter fusion-based method and analysing different LAI precursors related to three powerful earthquakes including: Turkey (6 February 2023), Myanmar (28 March 2025) and Russia-Kamchatka (9 July 2025), the magnitude parameter of the mentioned earthquakes is estimated. Then, the obtained results are compared with the real observed magnitude value and also three fusion methods including Fuzzy Inference System (FIS), Dampsfer Schafer (DS) and Bayesian. The results obtained show that the proposed Kalman filter fusion method, along with other fusion methods, can be used effectively in predicting earthquake parameters.

ABSTRACT Recently, transfer learning, and especially domain adaptation (DA), has been introduced to the cross-scene remote sensing image scene classification task. The existing DA approaches for the remote sensing image built on the hypothesis that the source domain and target domain share identical category spaces. However, in a new region, novel categories may appear. Open-set domain adaptation (OSDA) methods is proposed to solve such problems. Existing OSDA methods need to determine the thresholds or other hyper-parameters (such as number of unknown sample) to detect novel classes. In this paper, an end-to-end open-set domain adaptation for cross scene classification (OSDA_CSC) framework is proposed to solve these problems. In this framework, to overcome the different feature distributions caused by the different class spaces and the feature discriminability decrease in the remote sensing OSDA scenario, we propose an unknown-class detector using a one-vs-all network considering class distances to find the unknown classes in the target domain. We also introduce target domain neighbourhood clustering loss to improve the target domain feature discriminability. The proposed method can improve cross-scene classification and automatically label the unknown class samples in the target domain.

ABSTRACT Urgent remote sensing events such as floods, earthquakes, and target detection require real-time information from on-board synthetic aperture radar (SAR) data processing. In SAR image preprocessing, radiometric correction and coarse geometric correction are time-consuming. Although graphics processing unit (GPU) is commonly used for acceleration, most existing GPU-based methods still execute these corrections sequentially, failing to fully utilize central processing unit (CPU) resources and multi-thread scheduling, leading to redundant time consumption. To address this inefficiency, we propose an efficient on-board SAR image preprocessing method based on a heterogeneous parallel architecture with a collaborative CPU/GPU computing strategy. Unlike conventional sequential implementations, our method integrates both antenna range pattern correction and coarse geometric correction into a single CUDA kernel through optimized design of the heterogeneous parallel architecture, significantly improving preprocessing efficiency. Experimental results on measured SAR data from the Taijing-4 satellite using edge-computing devices demonstrate that our method achieves better performance than existing state-of-the-art methods. The proposed method provides 13% higher acceleration than a general GPU approach, while maintaining image quality and positioning accuracy.

ABSTRACT Short-lived extreme rainfall events pose major challenges for flood forecasting and disaster response because of their high intensity and strong spatial localization. This study examines how rainfall datasets with different spatial and temporal resolutions represent such events over India. A total of 100 extreme rainfall cases are analysed using two datasets: a high-resolution satellite-based precipitation dataset (8 km, hourly) that combines multi-sensor satellite observations calibrated with rain gauge measurements, and the daily gridded rainfall dataset from the India Meteorological Department (IMD) at 25 km resolution derived from a dense gauge network. To isolate the effects of data aggregation, the high-resolution dataset is also converted to daily totals and spatially coarsened to 25 km resolution. The original hourly dataset is treated as the reference. Statistical comparisons based on peak intensity, time-series behaviour, and intensity-weighted visualization show that IMD daily data underestimate peak rainfall by up to 55.5%, while temporally aggregated and spatially coarsened datasets underestimate peaks by about 22.7% and 39.5%, respectively. These biases are most pronounced in convectively active and urban regions where rainfall can exceed 80 mm h⁻¹ within 1–2 hours. The results highlight the importance of high-resolution hourly data for capturing short-duration rainfall extremes and improving real-time flood forecasting.

ABSTRACT Semantic change detection (SCD) is a critical task in the field of remote sensing, which aims to provide semantic understanding of remote sensing imagery by simultaneously detecting change regions and classifying their semantic categories. It plays a significant role in applications such as land use monitoring and ecological environment assessment. With the increasing availability of bi-temporal remote sensing datasets, a variety of deep learning-based models have been widely applied to the task of SCD detection. However, existing methods still suffer from limitations in modelling the interaction of cross-temporal semantic features and in the adaptive fusion of differential information, which constrains detection accuracy. Thus, to address these limitations, we propose ESTFNet, an efficient spatiotemporal fusion network for SCD detection. The network is built around two core modules: the Efficient Spatiotemporal Cross-Attention (ESTCA) and the (BDAF), which address feature interaction and feature fusion, respectively. Extensive experimental results demonstrate that ESTFNet outperforms several recent representative SCD models on remote sensing datasets, showing superior performance in complex scenarios. The code of our ESTFNet is available at https://github.com/loudongang/ESTFNet.