Climate classification plays a fundamental role in understanding climatic patterns, particularly in the context of a changing climate. This study utilized hourly meteorological data from 36 major cities in China from 2011 to 2021, including 2 m temperature (T2), relative humidity (RH), and precipitation (PRE). Both original hourly sequences and daily value sequences were used as inputs, applying two non-hierarchical clustering methods (k-means and k-medoids) and four hierarchical clustering methods (ward, complete, average, and single) for clustering. The classification results were compared using two clustering evaluation indices: the silhouette coefficient and the Calinski–Harabasz index. Additionally, the clustering was compared with the Köppen–Geiger climate classification based on the maximum difference in intra-cluster variables. The results showed that the clustering method outperformed the Köppen–Geiger climate classification, with the k-medoids method achieving the best results. Our research also compared the effectiveness of climate classification using two variables (T2 and PRE) versus three variables, including the addition of hourly RH. Cluster evaluation confirmed that incorporating the original sequence of hourly T2, PRE, and RH yielded the best performance in climate classification. This suggests that considering more meteorological variables and using hourly observation data can significantly improve the accuracy and reliability of climate classification. In addition, by setting the class numbers to two, the clustering methods effectively identified climate boundaries between northern and southern China, aligning with China’s traditional geographical division along the Qinling–Huaihe River line.
Traffic flow prediction is a critical component of Intelligent Transportation Systems (ITS) and smart city infrastructures. This survey paper provides a comprehensive analysis of recent advancements in deep learning-based approaches for traffic flow prediction, focusing on spatiotemporal correlations and attention mechanisms. We systematically review five seminal papers that propose innovative neural network architectures including DHSTNet, Att-DHSTNet, and ASTMGCNet for citywide traffic prediction. Our survey examines their methodologies, key contributions, experimental results, and comparative performance. We organize the discussion around three main themes: (1) modeling dynamic spatiotemporal dependencies, (2) attention mechanisms for traffic prediction, and (3) hybrid neural network architectures. The paper includes detailed comparison tables and conceptual figures synthesized from the reviewed works. Our analysis shows that attention-based hybrid models outperform traditional techniques, with ASTMGCNet having the lowest RMSE (4.06) and MAPE (12.56%) on benchmark datasets. We end by outlining current issues and potential research directions in this rapidly changing subject.
8 July 2025 in ICCK Transactions on Advanced Computing and Systems
Traffic flow prediction is a critical component of Intelligent Transportation Systems (ITS) and smart city infrastructures. This survey paper provides a comprehensive analysis of recent advancements in deep learning-based approaches for traffic flow prediction, focusing on spatiotemporal correlations and attention mechanisms. We systematically review key representative works that propose innovative neural network architectures including DHSTNet, Att-DHSTNet, and ASTMGCNet for citywide traffic prediction. Our survey examines their methodologies, key contributions, experimental results, and comparative performance. We organize the discussion around three main themes: (1) modeling dynamic spatiotemporal dependencies, (2) attention mechanisms for traffic prediction, and (3) hybrid neural network architectures. The paper includes detailed comparison tables and conceptual figures synthesized from the reviewed works. Our analysis shows that attention-based hybrid models outperform traditional techniques, with ASTMGCNet having the lowest RMSE (4.06) and MAPE (12.56%) on benchmark datasets. We end by outlining current issues and potential research directions in this rapidly changing subject.
Accurate prediction of tropical cyclone (TC) induced rainfall is of great importance to disaster prevention but quite challenging to operational forecasters. This study quantifies the spatial characteristics of the hourly rainfall caused by TCs at several individual weather stations over the south China coast. We use the historical TC datasets from 2004 to 2018 and the hourly rainfall data from six weather stations to investigate the stations' rainfall distribution characteristics induced by TCs within a distance of 700 km to the weather stations. The hourly rainfall data during the study period from six coastal stations: Zhuzilin (ZZL), Shekou Ferry Terminal (SFT), and Wutong Village (WTV) in Shenzhen, and Hong Kong Observatory (HKO), Sha Tin (SHA), and Tai Mo Shan (TMS) in Hong Kong are used to investigate the statistical relationship. Results show that the most hazardous areas for TC‐induced rainfall are TCs within a distance of 200 km relative to the stations. High intensity TCs in the southwest quadrant may bring torrential rainfall to stations within 500 km distance, and low intensity TCs are likely to induce torrential rainfall to stations within 300 km distance when they are in the northeast and northwest quadrants. If TCs are over the region with the azimuth of 60–150° and distance out of 200 km to the six stations, there is almost no rainfall influence on those stations. The findings of this study are valuable to provide information for the local operational forecasters when predicting the potential rainfall along the south China coast due to future TCs.
The spatiotemporal characteristics of near-surface wind in Shenzhen were investigated in this study by using hourly observations at 92 automatic weather stations (AWSs) from 2009 to 2018. The results show that during the past 10 years, most of the stations showed a decreasing trend in the annual mean of the 10 min average wind speed (avg-wind) and the mean of the 3 s average wind speed (gust wind). Over half of the decreasing trends at the stations were statistically significant (p < 0.05). Seasonally, the decrease in wind speed was the most severe in spring, followed by autumn, winter, and summer. The distribution of wind speed tends to be greater in the east and coastal areas for both avg-wind and gust wind. From September to March of the following year, the prevailing wind direction in Shenzhen was northerly, and from April to August, the prevailing wind direction was southerly. The seasonal wind speed distribution exhibited two different types, spring–summer type and autumn–winter type, which may be induced by their different prevailing wind directions. The analysis by the empirical orthogonal function (EOF) method confirmed the previous findings that the mean wind speed was decreasing in Shenzhen and that two different seasonal wind speed spatial distribution patterns existed. Such a study could provide references for wind forecasting and risk assessment in the study area.