Radar Data Research Articles

Mapping Water Bodies and Wetlands from Multispectral and SAR Data for the Cross-Border River Basins of the Polish–Ukrainian Border

Using remote sensing data to accurately record water surface changes over large areas is crucial in monitoring water resources. However, mapping water surfaces from remote sensing data has its advantages and disadvantages. This study presents a method for mapping water surfaces and wetlands based on Sentinel-1/-2 data over a study area of more than 26,000 km2 in three river basins, the Bug, Dniester, and San, located along the Polish–Ukrainian border. To achieve this goal, an image processing algorithm with additional options was developed (special filters, type classification, and post-classification), which minimized the shortcomings and increased the accuracy of the method. As a result, by using optical and radar data, it was possible to create maps of water bodies in the study area in the driest month of the year from 2018 to 2021. The results were evaluated numerically and graphically. The accuracy of the method was assessed using the Kappa coefficient. For optical data, the lowest value was 76.28% and the highest was 88.65%; for radar data, these values were 87.61% and 97.18%, respectively. When assessing accuracy, the highest values were achieved for overall accuracy (OA), with a maximum of 0.95 (for SAR) and 0.91 (for optical data). The highest values were in user accuracy (UA), with a maximum value of 1 for both SAR and optical data.

Integrated Flood Impact and Vulnerability Assessment Using a Multi-Sensor Earth Observation Mission with the Perspective of an Operational Service in Lombardy, Italy

The frequency and the accumulation of medium–small flood events can cause severe impacts. In a climate change context, real-time monitoring and a fast risk assessment are needed to support the post-disaster phases. The present work presents a novel methodology that leverages the potential of earth observation data to produce a proof-of-concept for flood vulnerability assessment, serving as the basis for a Map Operational Service for the Lombardy region. The proof-of-concept is related to both flood hazard estimation and vulnerability assessment, considering the evaluation of the potentialities of the synthetic aperture radar data when used to feed a downstream service. Using the city of Pavia (Italy), which was affected by a flood event in November 2019, as a case study, we present an integrated flood impact approach that includes a combination of social and physical parameters. The results contribute to a processing chain designed as a pre-operational service where each data analytic retrieves thematic products to support the exposure and damage estimates based on earth observation-derived hazard products for emergency and recovery responses. Three different satellites covered more than 40 h of the flood’s evolution, supporting the great potential of the multi-sensor approach. Therefore, different sensor configurations in terms of spectral bands (X and C bands) and resolutions (from 10 to 1 m) provide a near real-time view of the event. Comparing the results obtained through the three hazard scenarios, a final social and physical Integrated Impact Index is obtained. The added value information leads to the determination of hotspots with which to prioritize effective interventions during emergency and recovery phases, crucial for capturing inherent conditions that allow communities to absorb impacts and cope with a damaging flood event.

Assessing Nowcast Models in the Central Mexico Region Using Radar and GOES-16 Satellite Data

In this study, the nowcast models provided by the Python pySTEPS library were evaluated using radar derived rain rate data and the satellite product Split-Window Difference (SWD) based on GOES-16 data, focusing on central Mexico. Initially, we obtained a characterization of the rainfall that occurred in the region using the radar rain rate and the SWD. Subsequently the nowcasts were evaluated using both variables. Two nowcast models were employed from pySTEPS: Extrapolation and S-PROG. The results indicate that average SWD is below 2.5 K, 90 min before the onset of rainfall events, and, on average, the SWD is 2 K during rainfall events. The results from both nowcast models were accurate and produced similar results. The nowcasts performed better when SWD data were used as input, having an average Probability of Detection (PoD) above 70% and a False Alarm Rate (FAR) reaching 40% for the 15-min prediction. The nowcasts were less accurate using the radar rain rate as input for the 15-min forecast, where the PoD was maximum 70% and FAR reaching 40%. However, these nowcasts were more reliable during well-organized precipitation events. In this work, it was determined that the nowcast models provided by pySTEPS can provide valuable rain forecasts using GOES-16 satellite and radar data for the central Mexico region.

Operational Assessment of High Resolution Weather Radar Based Precipitation Nowcasting System

North East Monsoon (NEM) is the major source of rainfall for the south-eastern parts of peninsular India. Short time rainfall prediction data (i.e., nowcasting) are based on the observations from Doppler weather radars which has a high spatial and temporal resolution. This study focuses on the short-term ensemble prediction system using weather radar data to predict precipitation during the NEM and is the first of its kind in the Indian region to make an assessment of the operational performance of the prediction system. Six rainfall events have been studied for the assessment of short-term prediction system where the precipitation systems are different and include a tropical storm observed over different days during the 2022 NEM season. To assess the performance of the system, Fractional Skill Scores (FSS) at a 1 km window have been computed for a lead time of 0–2 h for all the rainfall events with more than 750 samples using different optical flow methods and ensemble sizes. The best average skill score and maximum skill score obtained at a 2 h lead time is 0.65 and 0.78 for tropical storms, 0.5 and 0.78 for stratiform and 0.15 and 0.38 for convective precipitation. It has found that the performance of the model is best for precipitation systems that are widespread and have a longer life period.

Toward automated hail disaster weather recognition based on spatio-temporal sequence of radar images

Abstract Hail, an intense convective catastrophic weather, is seriously hazardous to people’s lives and properties. This article proposes a multi-step cyclone hail weather recognition model, called long short-term memory (LSTM)-C3D, based on radar images, integrating attention mechanism and network voting optimization characteristics to achieve intelligent recognition and accurate classification of hailstorm weather based on long short-term memory networks. Based on radar echo data in the strong-echo region, LSTM-C3D can selectively fuse the long short-term time feature information of hail meteorological images and effectively focus on the significant features to achieve intelligent recognition of hail disaster weather. The meteorological scans of 11 Doppler weather radars deployed in various regions of the Hunan Province of China are used as the specific experimental and application objects for extensive validation and comparison experiments. The results show that the proposed method can realize the automatic extraction of radar reflectivity image features, and the accuracy of hail identification in the strong-echo region reaches 91.3%. It can also effectively realize the prediction of convective storm movement trends, laying the theoretical foundation for reducing the misjudgment of extreme disaster weather.

Irrigation Detection Using Sentinel-1 and Sentinel-2 Time Series on Fruit Tree Orchards

In arid and semi-arid regions, irrigation is crucial to mitigate water stress and yield loss. However, the overexploitation of water resources by the agricultural sector together with the climate change effects can lead to water scarcity. Effective regional water management depends on estimating irrigation demand using maps of irrigable areas or national and regional statistics of irrigated areas. These statistical data are not always of reliable quality because they generally do not reflect the updated spatial distribution of irrigated and rainfed fields. In this context, remote sensing provides reliable methods for gathering useful agricultural information from derived records. The combined use of optical and radar Earth Observation data enhances the probability of detecting irrigation events, which can improve the accuracy of irrigation mapping. Hence, we aimed to utilize Sentinel-1 (VV and VH) and Sentinel-2 (NDVI) data to classify irrigated fruit trees and rainfed ones in a study area located in the Castilla La-Mancha region in Spain. To obtain these time-series data from Sentinel-1 and Sentinel-2, which constitute the input data for the classification algorithms, a tool has been developed for automating the download from the Sentinel Hub. This tool downloads products organized by tiles for the region of interest and for the entire required time-series, ensuring the spatial repeatability of each pixel across all products and dates. The classification of irrigated plots was carried out by SVM Support Vector Machine. The employed methodology displayed promising results, with an overall accuracy of 88.4%, indicating the methodology’s ability to detect irrigation over orchards that were declared as non-irrigated. These results were evaluated by applying the change detection method of the σp0 backscattering coefficient at plot scale.

Convective Properties and Lightning Activity in Different Categories of Thunderstorms over the Beijing Area during Five Warm Seasons

Based on comprehensive observations, including total lightning, Doppler radar, precipitation, and other meteorological data, the variations in thunderstorm properties and lightning activity of different categories for thunderstorms over the Beijing area during five warm seasons were investigated. According to the morphology of radar echo, thunderstorms were classified into five categories, including single convective cells, multi-cells, linear mesoscale convective system (MCS), nonlinear MCS, and weak convective precipitation system (WCPS). The diurnal variability of lightning, thunderstorm occurrence, and precipitation showed late-afternoon maxima, with the peak time of lightning frequency occurring before that of precipitation. Despite WCPS having the lowest lightning frequency, the percentage of +CG/CG was the highest with large peak currents. The convective available potential energy (CAPE) of linear MCS, multi-cells, nonlinear MCS, single cells, and WCPS categories followed a pattern from largest to smallest. Meanwhile, warm cloud depth (WCD) exhibited a smaller value in the well-organized thunderstorm categories and a larger value in the WCPS. The topographic forcing mechanism and large wind gradient along mountain slopes facilitated convection occurrence and enhancement, further promoting lightning production. Meanwhile, the nocturnal convection mechanism significantly impacted the activity of nonlinear MCS and WCPS.

High frequency Lunar Penetrating Radar quality control, editing and processing of Chang’E-4 lunar mission

Chinese lunar landing mission Chang’E-4 reached the far side of the Moon in January 2019 and has been providing unprecedented Lunar Penetrating Radar data able to explore the lunar subsurface down to more than 40 m (with its more resolutive high frequency band). Data are periodically released to the scientific community in raw PDS4 format. Here we provide different versions of the radar dataset after editing (i.e. pre-processing), partial, and full processing in order to provide a complete ready-to-use dataset to end-users (data collected since 4th January 2019 until 27th March 2023) which can be directly exploited for analysis, interpretation, inversion, as well as integration with imagery or other information. In particular, we implemented an efficient and objective way to remove duplicated traces representing more than 90% of original data, as well as a processing flow able to retain all the original data information, while avoiding redundancies. The provided datasets can be implemented with future data releases and straightforwardly exploited for any future analysis.

A Time-of-Flight and Radar Dataset of a neonatal Thorax Simulator with synchronized Reference Sensor Signals for respiratory Rate Detection

In this paper we present an open-source Time-of-Flight and radar dataset of a neonatal thorax simulator for the development of respiratory rate detection algorithms. As it is very difficult to gain recordings of (preterm) neonates and there is hardly any open-source data available, we built our own neonatal thorax simulator which simulates the movement of the thorax due to respiration. We recorded Time-of-Flight (ToF) and radar data at different respiratory rates in a range of 5 to 80 breaths per minute (BPM) and with varying upstroke heights. As gold standard a laser micrometer was used. The open-source data can be used to test new algorithms for non-contact respiratory rate detection.

Identification of Rainfall Events and Heavy Rainfall Events From Radar Measurements in Southeastern Australia

AbstractRadar data can be of significant utility in investigating characteristics of rainfall events that cannot be studied with rain gauges alone. The recent establishment of a long‐term, quality‐controlled data set covering most of the radars on the Australian continent enables a deeper characterization of rainfall, including heavy rainfall events. This study develops a methodology to identify and characterize rainfall events from radar data and tests its utility by applying it to the regions surrounding the major cities of Brisbane, Sydney, and Melbourne. The event characteristics studied include rainfall accumulation and intensity, event duration and spatial extent, and the contribution convective areas make to the overall event rainfall. Rainfall events in Brisbane and Sydney are found to be more intense, more convective, and smaller in extent whilst producing larger rainfall accumulations than Melbourne rainfall events. Rainfall event duration and total accumulated rainfall are strongly positively correlated, as are the overall event intensity and the intensity of convective rainfall. The events that produce the largest rainfall accumulations exhibit significant differences from the events that produce the highest rainfall intensities. Overall, the study demonstrates that long‐term radar data sets in Australia provide an invaluable and rich source to study rainfall characteristics in a variety of regions at a high spatial and temporal resolution.

Evidence of Ice‐Rich Layered Deposits in the Medusae Fossae Formation of Mars

AbstractSubsurface reflectors in radar sounder data from the Mars Advanced Radar for Subsurface and Ionospheric Sounding instrument aboard the Mars Express spacecraft indicate significant dielectric contrasts between layers in the Martian Medusae Fossae Formation (MFF). Large density changes that create dielectric contrasts are less likely in deposits of volcanic ash, eolian sediments, and dust, and compaction models show that homogeneous fine‐grained material cannot readily account for the inferred density and dielectric constant where the deposits are more than a kilometer thick. The presence of subsurface reflectors is consistent with a multi‐layer structure of an ice‐poor cap above an ice‐rich unit analogous to the Martian Polar Layered Deposits. The volume of an ice‐rich component across the entire MFF below a 300–600 m dry cover corresponds to a global equivalent layer of water of ∼1.5 to ∼2.7 m or ∼30%–50% of the total estimated in the North Polar cap.

Radar Echo Recognition of Gust Front Based on Deep Learning

Gust fronts (GFs) belong to the boundary layer convergence system. A strong GF can cause serious wind disasters, so its automatic monitoring and identification are very helpful but difficult in daily meteorological operations. By collecting convective weather processes in Hubei, Jiangsu, and other regions of China, 1422 GFs from 106 S-band new-generation weather radar (CINRAD/SA) volume scan data are labeled as positive samples by means of human–computer interaction, and the same number of negative samples are randomly tagged from no GF radar data. A deep learning dataset including 2844 labels with a positive and negative sample ratio of 1:1 is constructed, and 80%, 10%, and 10% of the dataset are separated as training, validation, and test sets, respectively. Then, the training dataset is expanded to 273,120 samples by data augmentation technology. Since the height of a GF is generally less than 1.5 km, three deep-learning-based models are trained for GF automatic recognition according to the distance from the radars. Three models (M1, M2, M3) are trained with the data at a 0.5° elevation angle from 65 to 180 km away from the radars, at 0.5° and 1.5° angles from 40 to 65 km, and at 0.5°, 1.5°, and 2.4° angles within 40 km, respectively. The precision, confusion matrix, and its derived indicators including receiver operating characteristic curve (ROC) and the area under ROC (AUC) are used to evaluate the three models by the test set. The results show that the identification precisions of the models are 97.66% (M1), 90% (M2), and 90.43% (M3), respectively. All the hit rates are over 89%, the false positive rates are less than 11%, and the critical success indexes (CSIs) surpass 82%. In addition, all the optimal critical points on the ROC curves are close to (0, 1), and the AUC values are above 0.93. These results suggest that the three models can effectively achieve the automatic discrimination of GFs. Finally, the models are demonstrated by three GF events detected with Qingpu, Nantong, and Cangzhou radars.

Deep learning–based inverse analysis of GPR data for landslide hazards

In mountainous landscapes, the diverse geotechnical conditions amplify landslide susceptibility. Factors such as precipitation and seismic activity can trigger landslides, while inherent hazards such as voids, fissures, and compaction deficits jeopardize long-term slope stability. Detecting and forecasting these susceptibilities accurately is crucial. In this paper, the time-domain finite-difference approach and the gprMax software are used to conduct forward modeling of landslide susceptibility. An electrical model of subsurface aqueous structures is created, including water-filled and air-filled cavities, fracture zones, and fault lines. The distinctive radar signal responses within these environments are examined, and a dataset of B-scan images associated with their electrical models is constructed. By employing deep learning algorithms and the robust nonlinear mapping ability of convolutional neural networks in the Pix2Pix generative adversarial network, we accelerate the intelligent inversion of the geological radar data on landslide susceptibility. This innovative approach effectively reconstructs hazard models, offering a reliable basis for interpretation of radar signals.

FMCW Radar on LiDAR map localization in structural urban environments

AbstractMultisensor fusion‐based localization technology has achieved high accuracy in autonomous systems. How to improve the robustness is the main challenge at present. The most commonly used LiDAR and camera are weather‐sensitive, while the frequency‐modulated continuous wave Radar has strong adaptability but suffers from noise and ghost effects. In this paper, we propose a heterogeneous localization method called Radar on LiDAR Map, which aims to enhance localization accuracy without relying on loop closures by mitigating the accumulated error in Radar odometry in real time. To accomplish this, we utilize LiDAR scans and ground truth paths as Teach paths and Radar scans as the trajectories to be estimated, referred to as Repeat paths. By establishing a correlation between the Radar and LiDAR scan data, we can enhance the accuracy of Radar odometry estimation. Our approach involves embedding the data from both Radar and LiDAR sensors into a density map. We calculate the spatial vector similarity with an offset to determine the corresponding place index within the candidate map and estimate the rotation and translation. To refine the alignment, we utilize the Iterative Closest Point algorithm to achieve optimal matching on the LiDAR submap. The estimated bias is subsequently incorporated into the Radar SLAM for optimizing the position map. We conducted extensive experiments on the Mulran Radar Data set, Oxford Radar RobotCar Dataset, and our data set to demonstrate the feasibility and effectiveness of our proposed approach. Our proposed scan projection descriptors achieves homogeneous and heterogeneous place recognition and works much better than existing methods. Its application to the Radar SLAM system also substantially improves the positioning accuracy. All sequences' root mean square error is 2.53 m for positioning and 1.83° for angle.

Information Extraction and Three-Dimensional Contour Reconstruction of Vehicle Target Based on Multiple Different Pitch-Angle Observation Circular Synthetic Aperture Radar Data

The circular synthetic aperture radar (CSAR) has the ability of all-round continuous observation and high-resolution imaging detection, and can obtain all-round scattering information and higher-resolution images of the observation scene, so as to realize the target information extraction and three-dimensional (3D) contour reconstruction of the observation targets. However, the existing methods are not accurate enough to extract the information of vehicle targets. Through the analysis of the vehicle target scattering model and CSAR image characteristics, this paper proposes a vehicle target information extraction and 3D contour reconstruction method based on multiple different pitch-angle observation CSAR data. The proposed method creatively utilizes the projection relationship of the vehicle in 2D CSAR imaging to reconstruct the 3D contour of the vehicle, without prior information. Firstly, the CSAR data obtained from multiple different pitch-angle observations are fully utilized, and the scattering points of odd-bounce reflection and even-bounce reflection echoes are extracted from the two-dimensional (2D) coherent CSAR images of the vehicle target. Secondly, the basic contour of the vehicle body is extracted from the scattering points of the even-bounce reflected echoes. Then, the geometric projection relationship of the “top–bottom shifting” effect of odd-bounce reflection is used to calculate the height and position information of the scattering points of odd-bounce reflection, so as to extract the multi-layer 3D contour of the vehicle target. Finally, the basic contour and the multi-layer 3D contour of the vehicle are fused to realize high-precision 3D contour reconstruction of the vehicle target. The correctness and effectiveness of the proposed method are verified by using the CVDomes simulation dataset of the American Air Force Research Laboratory (AFRL), and the experimental results show that the proposed method can achieve high-precision information extraction and realize distinct 3D contour reconstruction of the vehicle target.

Exploiting radar polarimetry for nowcasting thunderstorm hazards using deep learning

Abstract. This work presents the importance of polarimetric variables as an additional data source for nowcasting thunderstorm hazards using an existing neural network architecture with recurrent-convolutional layers. The model can be trained to predict different target variables, which enables nowcasting of hail, lightning, and heavy rainfall for lead times up to 60 min with a 5 min resolution, in particular. The exceedance probabilities of Swiss thunderstorm warning thresholds are predicted. This study is based on observations from the Swiss operational radar network, which consists of five operational polarimetric C-band radars. The study area of the Alpine region is topographically complex and has a comparatively very high thunderstorm activity. Different model runs using combinations of single- and dual-polarimetric radar observations and radar quality indices are compared to the reference run using only single-polarimetric observations. Two case studies illustrate the performance difference when using all predictors compared to the reference model. The importance of the predictors is quantified by investigating the final training loss of the model, with skill scores such as critical success index (CSI), precision, recall, precision–recall area under the curve, and the Shapley value. Results indicate that single-polarization radar data are the most important data source. Adding polarimetric observations improves the model performance compared to reference model in term of the training loss for all three target variables. Adding quality indices does so, too. Including both polarimetric variables and quality indices at the same time improves the accuracy of nowcasting heavy precipitation and lightning, with the largest improvement found for heavy precipitation. No improvement could be achieved for nowcasting of the probability of hail in this way.

Evaluation of hydrological models on small mountainous catchments: impact of the meteorological forcings

Abstract. Hydrological modelling of small mountainous catchments is particularly challenging because of the high spatio-temporal resolution required for the meteorological forcings. In situ measurements of precipitation are typically scarce in these remote areas, particularly at high elevations. Precipitation reanalyses propose different alternative forcings for the simulation of streamflow using hydrological models. In this paper, we evaluate the performances of two hydrological models representing some of the key processes for small mountainous catchments (< 300 km2), using different meteorological products with a fine spatial and temporal resolution. The evaluation is performed on 55 small catchments of the northern French Alps. While the simulated streamflows are adequately reproduced for most of the configurations, these evaluations emphasize the added value of radar measurements, in particular for the reproduction of flood events. However, these better performances are only obtained because the hydrological models correct the underestimations of accumulated amounts (e.g. annual) from the radar data in high-elevation areas.

Spatial and temporal variability analysis of snow cover parameters according to the urbanized area profile system

The purpose of the study is to analyze the spatial and temporal variability of snow cover parameters in the urbanized area of the city of Irkutsk and its adjacent areas according to the system of profiles reflecting the features of the area microrelief. Field surveys were conducted in accordance with standard recommendations for snow and geochemical surveys. Snow sampling was carried out at the entire snow depth, with the exception of the snow depth of 0.5-1 cm. Sampling points located at least 25 meters away from the roads. Snow sampling in the residential areas of the city was carried out in the places with undisturbed snow cover and free from extraneous snow banks and landfills. Global Mapper, Golden Software Surfer, Statistica software were used to carry out statistical analysis, build the models of profile sections and obtain the diagrams of snow cover parameter distribution. A 3D relief model was obtained on the basis of radar topographic survey data, which revealed a multiple formation relief of the studied area featuring mountains, high plains with undulating flat gently sloping watersheds, valleys, hollows and depressions. The height difference is up to 230 m. The diagrams of snow depth distribution built with regard to the profiles made it possible to identify the territories with the deepest snow cover. A geochemical analysis of the melt water filtrate was carried out over a three-year period. In 2021 a high correlation with electrical conductivity was found for the contents of tungsten, sodium, bromine, calcium, molybdenum, sulfur, barium, magnesium, antimony, tantalum, cesium, titanium, chromium, and silicon. The average correlation level with electrical conductivity was recorded for arsenic, copper, and lead. The analysis data obtained indicate a different contamination level of the area as a result of atmospheric precipitation in those years due to the activity of industrial facilities. The distribution patterns of the pH snow index have been revealed. It is shown that zones with pH<6 are mostly confined to the Irkutsk aluminum smelter. The zones with pH >6 are caused by the influence of the power plants using hydrocarbon fuels: gasoline, kerosene, fuel oil, diesel fuel, coal. The snow cover was subjected to field and laboratory studies using GIS technologies and physico-chemical methods. The distribution and migration of pollutants in various spatial and temporal aspects have been revealed taking into account the terrain relief. The conducted study opens up opportunities for modeling the landscape structure, taking into account meteorological parameters, phenological processes and snow cover state for the purposes of the national economy and the location of construction of various facilities.

The 3D Physics-Based Broadband Ground-Motion Simulation, including Topography Effects Causing Damage to Structures, for the 2022 Mw 6.6 Menyuan, China, Earthquake

ABSTRACT On 8 January 2022, Menyuan County, China, was struck by an Mw 6.6 earthquake that caused surface rupture in the epicentral area and severe damage to an important railway bridge. The earthquake was recorded by only one strong-motion station, which presents a challenge for quantitatively estimating the extent of the ground-motion distribution caused by this event. In this study, the spectral element method (SPECFEM3D code), which solves the elastodynamic equations and can capture the full physics of seismic-wave propagation, is employed for broadband (0–10 Hz) ground-motion simulations of this earthquake. A hybrid kinematic source is developed in which the final slip distribution combines a prescribed asperity model based on Interferometric Synthetic Aperture Radar data (as a source for low-frequency radiation) and a stochastic part (as a source for high-frequency radiation), which introduces spatial heterogeneities to the prescribed asperity model. The numerical approach is first validated by modeling the well-recorded 1994 Northridge earthquake before modeling the Menyuan earthquake. The simulated ground motion is compared with the only observed strong-motion station record, as well as with empirical Next Generation Attenuation-West2 ground-motion models. Then, topography effect in Menyuan earthquake is studied in detail. The simulated ground motions with and without surface topography indicate that the topography tends to focus and scatter the seismic wavefield, resulting in amplification of the ground shaking. The results show a significant correlation between the peak ground velocity (PGV) and topography. The PGV amplification caused by topography effects is period dependent, and its peak amplification reaches up to 50% within a typical resonance period (1–2 s). It could be inferred that the railway bridge probably vibrated in resonance and suffered severe damage owing to the amplified long-period ground motion caused by the topography.

An Long Short-Term Memory Model with Multi-Scale Context Fusion and Attention for Radar Echo Extrapolation

Precipitation nowcasting is critical for areas such as agriculture, water resource management, urban drainage systems, transport and disaster preparedness. In recent years, methods such as convolutional recurrent neural networks (ConvRNN) in deep learning techniques have been used to solve this task. Despite the effective improvement in forecasting quality, there are still problems with blurred and distorted prediction images, as well as difficulties in effectively forecasting high echo regions. To solve the above problems, this article presents a spatio-temporal long–short-term memory network model in view of multi-scale context fusion and attention mechanisms. This method fully extracts the short-term context information of different scales of radar image through the multi-scale context fusion module. The attention module broadens the time perception domain of the prediction unit so that the model perceives more historical time dynamics. Using the Hong Kong region weather radar data as a sample, the results of the experimental comparative analysis show that the spatio-temporal long and short-term memory network in view of multi-scale context fusion and attention mechanism achieves better prediction performance. Our model is effective in improving both image quality and meteorological assessment metrics with higher accuracy and more details.