Search

Precipitation nowcasting refers to the prediction of small-scale precipitation events at minute and kilometer scales within the upcoming 0 to 2 h, which significantly impacts both human activities and daily life. However, prevailing deep learning models have primarily focused on a single radar echo data source, limiting their ability to effectively capture intricate and rapidly evolving precipitation patterns. Thus, meteorological satellite is considered to supplement radar echo data. To achieve a comprehensive integration of multisource data with enhanced details, a two-stage fusion satellite and radar GAN-based prediction network (named FsrGAN) is proposed. In the first stage, we design a satellite-radar fusion prediction network known as FsrNet. This network employs an encoder-fusion-decoder architecture, where a novel spatial-channel attention (SCA) is proposed to enhance the filtering and fusion of multisource and multiscale features. In the second stage, we introduce a GAN-based network (FusionGAN) that also mines the complementary information of satellite images to sharpen the first-stage predicted radar maps with more details. Experiments are conducted on meteorological dataset in the Yangtze River Delta (YRD) region. The test results exhibit the notably superior performance of our model in terms of image quality and precipitation forecasting metrics in comparison to traditional optical flow-based methods and some well-known deep learning methods (ConvLSTM, ConvGRU, TrajGRU and PredRNN++). More importantly, our fusion model using satellite and radar data demonstrates the ability to predict convective initiation.

Abstract Intensification of short‐duration rainfall extremes contributes to increased urban flood risk. Yet, it remains unclear how upper‐tail rainfall statistics could change with regional warming. Here, we characterize the non‐stationarity of rainfall extremes over durations of 1–24 hr for the rapidly developing coastal megalopolis of the Greater Bay Area, China. Using high‐resolution, multi‐source, merged and gridded data we observe greater increases in rainfall intensities over the north‐central part of the region compared with the southern coastal region. Our results show, for the first time, that urbanization nonlinearly increases rainfall intensities at different durations and return periods. Over short durations (≤3‐hr) and short return periods (2‐yr), urban areas have the greatest scaling rates (≥19.9%/°C). However, over longer durations (≥9‐hr) rural areas have greater scaling rates, with a lower degree of dependency on both durations and return periods.

Precipitation nowcasting underlying various public services from rainstorm warning to flight safety is quite important and remains challenging due to the fast change in convective weather. Although some deep learning models have been proposed to make prediction automatically, most of them just deal with a single radar echo data source, making them hard to adapt to heterogeneous and diverse data in practice. In this article, a heterogeneous spatiotemporal attention fusion prediction (HST-AFP) network is proposed for radar echo extrapolation (deterministic output) and further precipitation nowcasting, which deals with mining and fusing knowledge from multiple heterogeneous spatiotemporal (ST) data sources, including history radar echo observations and numerical weather prediction data. With the help of the proposed attention-based ST diffusion module, the multiencoder is designed to extract information from both dense ST tensor and sparse ST tensor. On the other hand, the fusion decoder achieves very deep trainable residual fusion prediction by integrating scalewise attention fusion module and deep residual spatial and temporal attention mechanism. It can adaptively blend multisource ST features and rescale the multiscale temporalwise and spatialwise features for better prediction. Experiments in a real-world dataset of South China show that compared with the ingenious recurrent-neural-network-based methods and newly proposed UNet-based methods, our HST-AFP network can handle complex input with heterogeneity in both space and time domains, and performs better on the precipitation nowcasting metrics as well as requires remarkable shorter forecast time.

Abstract Turbulent dissipation rate (ɛ) is a crucial parameter in turbulence theory, and an essential component of higher‐order planetary boundary layer schemes for numerical weather prediction and climate models. It is most often modeled diagnostically based on the dissipation scaling ɛ ∝ e3/2/L, where e and L are the turbulence kinetic energy (TKE) and the size of the largest turbulent eddies, respectively. Utilizing three‐month‐long vertically‐extended observations accompanied by high resolution large‐eddy simulations, scaling‐based ɛ‐models are evaluated, focusing on their stability dependence under daytime convective conditions. The analysis uncovers biases in the parameterized ɛ profiles that cannot be corrected through tuning of model constants. The biases are attributed to the limited and even opposing stability dependence of the modeled dissipation length. Close examination reveals violation of the dissipation scaling by the inclusion of TKE associated with organized convection. A self‐similar dissipation length is obtained when only the isotropic component of TKE is considered.

Based on Doppler radar observation and reanalysis data, the statistical characteristics of mesovortices (MVs) during the first rainy season (April–June) in Guangdong, South China, from 2017 to 2019 are studied, including their spatiotemporal distributions, structural features and favorable environmental conditions. The results show that the MVs usually exhibit short lifetimes; about 70% last for less than 30 min. The intensity and horizontal scale of the MVs are proportional to their lifetime. Long-lived MVs have larger horizontal scales and stronger intensities than short-lived ones. The MVs are mainly observed over the Pearl River Delta region, followed by western Guangdong Province, but relatively fewer in both eastern and northern Guangdong Province. The uneven spatial distribution of the MVs is closely related to the differences in the topographical features and the environment conditions over South China. MVs are prone to form over flat regions. The Pearl River Delta and eastern Guangdong regions are relatively flat compared to the more mountainous western and northern Guangdong regions. Moreover, the monsoonal south-westerlies, water vapor flux, atmospheric instability and vertical wind shear over southwest Guangdong are significantly larger than those in other regions and are favorable for the formation of MVs. The occurrence frequencies of MVs in central and southern parts of Guangdong display similar diurnal variations, reaching the peak during the late afternoon and early evening while dropping to the minimum overnight. However, the situation is opposite in northern Guangdong, with the peak overnight and the minimum during the late afternoon and early evening. The regional differences in diurnal variation are likely related to the moving direction of mesoscale convective systems (MCSs) in Guangdong.

Abstract Dynamically updated and fine‐grained flooding maps are critical for situational awareness and decision support. However, traditional methods, such as eyewitness reports, remote sensing, and hydrology models, may fail to correspond with the rapidly changing urban hydrological environment. The presence of crowdsourced data (such as social media data) allows for timely and cost‐effective monitoring of flood hazards through collective observations; however, such data can be unreliable due to sample bias and low spatiotemporal resolution. Therefore, new measures to identify flood‐affected roads are desirable. In this study, we propose a methodology that leverages taxi GPS data to support post‐event flood mapping for the road network. This method can identify whether a significant reduction in the taxi passing rate for a road segment was related to precipitation, and automatically recognize the flood‐affected roads based on a logistic regression model. Using taxi GPS data in Shenzhen as an example, we derived the flood map of the road networks, and compared and validated the results. This study demonstrated the usefulness of taxi GPS data in generating high‐quality flood maps and the value of incorporating multiple data sources for sensing near real‐time flood in the city.

Short-term rainfall prediction by radar echo map extrapolation has been a very hot area of research in recent years, which is also an area worth studying owing to its importance for precipitation disaster prevention. Existing methods have some shortcomings. In terms of image indicators, the predicted images are not clear enough and lack small-scale details, while in terms of precipitation accuracy indicators, the prediction is not accurate enough. In this paper, we proposed a two-stage model (two-stage UA-GAN) to achieve more accurate prediction echo images with more details. For the first stage, we used the Trajectory Gated Recurrent Unit (TrajGRU) model to carry out a pre-prediction, which proved to have a good ability to capture spatiotemporal movement of rain field. In the second stage, we proposed a spatiotemporal attention enhanced Generative Adversarial Networks (GAN) model with a U-Net structure and a new deep residual attention module in order to carry out the refinement and improvement of the first-stage prediction. Experimental results showed that our model outperforms the optical-flow based method Real-Time Optical Flow by Variational Methods for Echoes of Radar (ROVER), and some well-known Recurrent Neural Network (RNN)-based models (TrajGRU, PredRNN++, ConvGRU, Convolutional Long Short-Term Memory (ConvLSTM)) in terms of both image detail indexes and precipitation accuracy indexes, and is visible to the naked eye to have better accuracy and more details.

Since strong convective weather is closely related to heavy precipitation, the nowcasting of convective weather, especially the nowcasting based on weather radar data, plays an essential role in meteorological operations for disaster prevention and mitigation. The traditional optical flow method and cross-correlation method have a low forecast accuracy and a short forecast leading time, while deep learning methods show remarkable advantages in nowcasting. However, most of the current forecasting methods based on deep learning suffer from the drawback that the forecast results become increasingly blurred as the forecast time increases. In this study, a weather radar nowcasting method based on the Temporal and Spatial Generative Adversarial Network (TSGAN) is proposed, which can obtain accurate forecast results, especially in terms of spatial details, by extracting spatial-temporal features, combining attention mechanisms and using a dual-scale generator and a multi-scale discriminator. The case studies on the forecast results of strong convective weather demonstrate that the GAN method performs well in terms of forecast accuracy and spatial detail representation compared with traditional optical flow methods and popular deep learning methods. Therefore, the GAN method proposed in this study can provide strong decision support for forecasting heavy precipitation processes. At present, the proposed method has been successfully applied to the actual weather forecasting business system.

Limited by the number of ground observation stations, PM2.5 retrieval from the remote sensing data is an effective complement to conventional ground observations and is a current research hotspot. The general principle behind the remote sensing retrieval of PM2.5 is to first retrieve the aerosol optical depth (AOD) and calculate the PM2.5 via the AOD-based statistical relationships. This method is likely to cause error propagation, which leads to instability in the retrieval model. In this paper, we propose a PM2.5 remote sensing retrieval method via an ensemble random forest machine learning method to directly establish the relationship between the moderate-resolution imaging spectroradiometer (MODIS) images and ground observational PM2.5 to avoid retrieval errors from the atmospheric aerosol optical depths and obtain PM2.5 retrieval results with higher precision and spatial resolution. The proposed method first uses a random forest to train and validate the MODIS images and ground observation station PM2.5 data; then, an optimal multi-model group, according to the determination coefficient R-square (R2) index, is selected. Finally, the optimal multi-model group is used on the whole MODIS image to obtain the PM2.5 retrieval result for the whole area. In an attempt to use machine learning technology to retrieve PM2.5, the experiments selected a substantial amount of MODIS image data during four seasons in Guangdong Province for validation and compared three performance indicators (R2, RMSE, and correlation coefficient (CC) to verify the superiority of the proposed algorithm.

Missing data in weather radar image sequences may cause bias in quantitative precipitation estimation (QPE) and quantitative precipitation forecast (QPF) studies, and also the obtainment of corresponding high-quality QPE and QPF products. The traditional approaches that are used to reconstruct missing weather radar images replace missing frames with the nearest image or with interpolated images. However, the performance of these approaches is defective, and their accuracy is quite limited due to neglecting the intensification and disappearance of radar echoes. In this study, we propose a deep neuron network (DNN), which combines convolutional neural networks (CNNs) and bi-directional convolutional long short-term memory networks (CNN-BiConvLSTMs), to address this problem and establish a deep-learning benchmark. The model is trained to be capable of dealing with arbitrary missing patterns by using the proposed training schedule. Then the performances of the model are evaluated and compared with baseline models for different missing patterns. These baseline models include the nearest neighbor approach, linear interpolation, optical flow methods, and two DNN models three-dimensional CNN (3DCNN) and CNN-ConvLSTM. Experimental results show that the CNN-BiConvLSTM model outperforms all other baseline models. The influence of data quality on interpolation methods is further investigated, and the CNN-BiConvLSTM model is found to be basically uninfluenced by less qualified input weather radar images, which reflects the robustness of the model. Our results suggest good prospects for applying the CNN-BiConvLSTM model to improve the quality of weather radar datasets.

The Three Gorges Dam (TGD) is one of the largest hydroelectric projects in the world. Monitoring the spatiotemporal distribution of extreme precipitation offers valuable information for adaptation and mitigation strategies and reservoir management schemes. This study examined variations in extreme precipitation over the Three Gorges Reservoir area (TGRA) in China to investigate the potential role of climate warming and Three Gorges Reservoir (TGR). The trends in extreme precipitation over the TGRA were investigated using the iterative-based Mann–Kendall (MK) test and Sen’s slope estimator, based on weather station daily data series and TRMM (Tropical Rainfall Measuring Mission) data series. The mean and density distribution of extreme precipitation indices between pre-dam and post-dam, pre-1985 and post-1985, and near and distant reservoir area were assessed by the Mann–Whitney test and the Kolmogorov–Smirnov test. The ratio of extreme precipitation to non-extreme precipitation became larger. The precipitation was characterized by increases in heavy precipitation as well as decreases in light and moderate rain. Comparing extreme precipitation indices between pre-1985 (cooling) and post-1985 (warming) indicated extreme precipitation has changed to become heavier. Under climate warming, the precipitation amount corresponding to more than the 95th percentile increased at the rate of 6.48%/°C. Results from comparing extreme precipitation for the pre- and post-dam, near reservoir area (NRA) and away from the reservoir area (ARA) imply an insignificant role of the TGR on rainfall extremes over the TGRA. Moreover, the impoundment of TGR did not exert detectable impacts on the surface relative humidity (RH) and water vapor pressure (WP).

This study examines the impact of three-dimensional variational data assimilation (3DVAR) on the prediction of two heavy rainfall events over Southern China by using a real-time storm-scale forecasting system. Initialized from the European Centre for Medium-Range Weather Forecasts (ECMWF) high-resolution data, the forecasting system is characterized by combining the Advanced Research Weather Research and Forecasting (WRF-ARW) model and the Advanced Regional Prediction System (ARPS) 3DVAR package. Observations from Doppler radars, surface Automatic Weather Station (AWS) network, and radiosondes are used in the experiments to evaluate the impact of data assimilation on short-term quantitative precipitation forecast (QPF) skill. Results suggest that extrasurface AWS data assimilation has slight but general positive impact on rainfall location forecasts. Surface AWS data also improve model results of near-surface variables. Radiosonde data assimilation improves the QPF skill by improving rainfall position accuracy and reducing rainfall overprediction. Compared with radar data, the overall impact of additional surface and radiosonde data is smaller and is reflected primarily in reducing rainfall overestimation. The assimilation of all radar, surface, and radiosonde data has a more positive impact on the forecast skill than the assimilation of either type of data only for the two rainfall events.

Chlorine radicals can enhance atmospheric oxidation which potentially increase tropospheric ozone concentration. However, few studies have been done to quantify the impact of chlorine emissions on ozone formation in China due to lack of chlorine emission inventory used in air quality model with sufficient resolution. In this study, Anthropogenic Chlorine Emissions Inventory for China (ACEIC) was developed for the first time, including emissions of hydrogen chloride (HCl) and molecular chlorine (Cl2) from coal combustion and prescribed waste incineration (waste incineration plant). The HCl and Cl2 emissions from coal combustion in China in 2012 were estimated to be 232.9 and 9.4 Gg, respectively, while HCl emission from prescribed waste incineration was estimated to be 2.9 Gg. Spatially high chlorine emissions were found in the North China Plain, the Yangtze River Delta and the Sichuan Basin. Air quality model simulations with the Community Multiscale Air Quality (CMAQ) modeling system were performed for November 2011 and the modeling results derived with and without chlorine emissions were compared. The magnitude of the simulated HCl, Cl2 and ClNO2 agreed reasonably with the observation when anthropogenic chlorine emissions were included in the model. The inclusion of the ACEIC increased the concentration of fine particulate Cl-, leading to enhanced heterogeneous reactions between Cl- and N2O5 which resulted in the higher production of ClNO2. Photolysis of ClNO2 and Cl2 in the morning and the reaction of HCl with OH in the afternoon produced chlorine radicals which accelerated tropospheric oxidation. When anthropogenic chlorine emissions were included in the model, the monthly mean concentrations of fine particulate Cl-, daily maximum 1-h ClNO2 and Cl radicals were estimated to increase by up to about 2.0 μg m-3, 773 pptv and 1.5 × 103 molecule cm-3 in China, respectively. Meanwhile, the monthly mean daily maximum 1-h O3 concentration was found to increase by up to 2.2 ppbv (3.8 %), while the monthly mean NOx concentration decreased by up to 0.5 ppbv (6.1 %). The anthropogenic chlorine emissions potentially increased the 1-h O3 concentration by up to 7.7 ppbv in China. This study highlights the need for the inclusion of anthropogenic chlorine emission on air quality modeling and demonstrated its importance on tropospheric ozone formation. [ABSTRACT FROM AUTHOR]