A Composite Transformer-Based Multi-Stage Defect Detection Architecture for Sewer Pipes

Urban sewer pipes are a vital infrastructure in modern cities,and their defects must be detected in time to prevent potential malfunctioning.In recent years,to relieve the manual efforts by human experts,models based on deep learning have been introduced to automatically identify potential defects.However,these models are insufficient in terms of dataset complexity,model versatility and performance.Our work addresses these issues with amulti-stage defect detection architecture using a composite backbone Swin Transformer.Themodel based on this architecture is trained using a more comprehensive dataset containingmore classes of defects.By ablation studies on the modules of combined backbone Swin Transformer,multi-stage detector,test-time data augmentation and model fusion,it is revealed that they all contribute to the improvement of detection accuracy from different aspects.The model incorporating all these modules achieves the mean Average Precision(mAP)of 78.6% at an Intersection over Union(IoU)threshold of 0.5.This represents an improvement of 14.1% over the ResNet50 Faster Region-based Convolutional Neural Network(R-CNN)model and a 6.7% improvement over You Only Look Once version 6(YOLOv6)-large,the highest in the YOLO methods.In addition,for other defect detection models for sewer pipes,although direct comparison with themis infeasible due to the unavailability of their private datasets,our results are obtained from a more comprehensive dataset and have superior generalization capabilities.

Tropical cyclone activities in the Western North Pacific in 2022

Based on the best-track dataset from the Shanghai Typhoon Institute/China Meteorological Administration,the paper provides a compre-hensive summary and analysis of tropical cyclone(TC)activities in the Western North Pacific(WNP)and the South China Sea(SCS)for 2022.Using the historical climatology from 1951 to 2020,the anomalous conditions during 2022 in TC frequency,origin locations,tracks,intensity,and duration for the entire ocean basin as well as landfall events in China are examined.Results show that the overall TC frequency is slightly lower than normal,but the multiple TC events have a very high frequency of occurrence.Origin locations of TCs,which mark the starting points of their paths,show a large westward and northward deviation from climatology.Around 40%of the named TCs exhibit a shift in their direction of movement from westerly to easterly.Additionally,comparisons of the means,medians,upper and lower quartiles all indicate that the intensity of TCs in 2022 is generally lower than the climatology,with the duration of TCs at tropical storm intensity or above being shorter than usual.A notable observation is the fewer incidence of TC landfalls in China,but with a geographical concentration in Guangdong Province.These anomalous annual TC activities are influenced by related atmospheric and oceanic environmental conditions modulated by multi-scale climate variability.Thefindings provide useful information for enhancing disaster mitigation strategies in the Asia-Pacific region.

Recent advances in landfalling tropical cyclone asymmetric rainfall mechanism and forecast verification over China

Held every four years,the International Workshop on Tropical Cyclone(IWTC)organized by the World Meteorological Organization has been a global leading conference in thefield of tropical cyclone.In preparation for the 10th IWTC(IWTC-10)in December 2022,a summary of research advances of landfalling tropical cyclone(LTC)rainfall during past four years of 2019–2022 has been prepared.Some of the latest research advances has been summarized in Lamers et al.(2023),which reviewed the latest forecast and disaster prevention methods related to TC precipitation.As a supplement,this article mainly focuses on the recent advances in LTC asymmetric rainfall evolution mechanisms and forecast verification results over China.Some newfindings have been made in the LTC inner-core size relationship with the asymmetric rainfall distri-bution.Some major advances focused on asymmetric microphysical characteristics in the TC rainbands.Current simulation and forecast per-formances of LTC precipitation have been analyzed,and different forecast error sources for rainfall during different landfall stages of TC were compared.To estimate the risk of TC rainfall hazards in China,a parameterized Tropical Cyclone Precipitation Model was reviewed as well in this article.

Forecasting tropical cyclone rainfall and flooding hazards and impacts

This review summarizes the rapporteur report on advances in monitoring and forecasting of rainfall associated with tropical cyclones(TCs)and its impact during 2014–18,as presented to the 10th International Workshop on TCs(IWTC-10)held in Bali,Indonesia during 5th–9th December 2022.Major physical processes that can modulate TC rainfall distribution,including topography,storm motion,vertical wind shear,and intensity,along with the fundamental physics of rain bands and clouds as simulated by numerical models,diurnal variation of rainfall,and various synoptic and mesoscale features controlling the rainfall distribution are briefly discussed.Improvements to the dynamic core and physical processes in global models are providing useable forecasts nearly up to 7 days.This report also summarizes,some tools that have been developed to predict TC rainfall.Lately there is a tendency for operational forecasting centers to utilize multi-model ensemble systems for rainfall forecasting that demonstrate superior performance than individual models,ensemble members,or even single model ensembles.Major impacts include pluvial and fluvial floods,and landslides.The techniques developed by various forecasting centers to assist in predicting and communicating the impacts associated with these events are also presented in this report.

RECENT ADVANCES IN RESEARCH AND FORECASTING OF TROPICAL CYCLONE TRACK, INTENSITY, AND STRUCTURE AT LANDFALL

This review prepared for the fourth International Workshop on Tropical Cyclone Landfall Processes(IWTCLP-4) summarizes the most recent(2015-2017) theoretical and practical knowledge in the field of tropical cyclone(TC) track, intensity, and structure rapid changes at or near landfall. Although the focus of IWTCLPIV was on landfall, this summary necessarily embraces the characteristics of storms during their course over the ocean prior to and leading up to landfall. In the past few years, extremely valuable observational datasets have been collected for TC forecasting guidance and research studies using both aircraft reconnaissance and new geostationary or low-earth orbiting satellites at high temporal and spatial resolution. Track deflections for systems near complex topography such as that of Taiwan and La Réunion have been further investigated, and advanced numerical models with high spatial resolution necessary to predict the interaction of the TC circulation with steep island topography have been developed. An analog technique has been designed to meet the need for longer range landfall intensity forecast guidance that will provide more time for emergency preparedness. Probabilistic track and intensity forecasts have also been developed to better communicate on forecast uncertainty. Operational practices of several TC forecast centers are described herein and some challenges regarding forecasts and warnings for TCs making landfall are identified. This review concludes with insights from both researchers and forecasters regarding future directions to improve predictions of TC track, intensity, and structure at landfall.

Rainfall forecast errors in different landfall stages of Super Typhoon Lekima (2019)

The rainfall forecast performance of the Tropical Cyclone(TC)version Model of Global and Regional Assimilation PrEdiction System(GRAPESTCM)of the China Meteorological Administration for landfalling Super Typhoon Lekima(2019)is studied by using the object-oriented verification method of contiguous rain area(CRA).The major error sources and possible reasons for the rainfall forecast uncertainties in different landfall stages(including near landfall and moving further inland)are compared.Results show that different performance and errors of rainfall forecast exist in the different TC stages.In the near landfall stage the asymmetric rainfall distribution is hard to be simulated,which might be related to the too strong forecasted TC intensity and too weak vertical wind shear accompanied.As Lekima moves further inland,the rain pattern and volume errors gradually increase.The Equitable Threat Score of the 24 h forecasted rainfall over 100 mm declines quickly with the time-length over land.The diagnostic analysis shows that there exists an interaction between the TC and the mid-latitude westerlies,but too weak frontogenesis is simulated.The results of this research indicate that for the current numerical model,the forecast ability of persistent heavy rainfall is very limited,especially when the weakened landing TC moves further inland.

OVERVIEW OF SEVERE TYPHOON FITOW AND ITS OPERATIONAL FORECASTS

Severe typhoon Fitow(1323)brought persistent and heavy rainfall to Zhejiang and the Shanghai area after it made landfall at Fujian Province of China in October 2013,breaking the rainfall records of several counties and districts in Zhejiang.In this paper,we provide an overview of the characteristics of Fitow’s landfall,including its track,intensity,structural evolution,heavy rainfall,and wind.We also describe some of the associated disastrous impacts.Finally,we provide verifications of operational forecasts of its track,intensity and rainfall.Though the track and intensity is well predicted,the rainfall persistence and enhancement in the second stage in Shanghai and north Zhejiang areas are not predicted out at all.The analysis presented in this paper provides forecasters and researchers with some valuable information on Fitow,which could form a useful basis for further studies.

Application of the frequency-matching method in the probability forecast of landfalling typhoon rainfall

In this paper,a revised method for typhoon precipitation probability forecast,based on the frequencymatching method,is developed by combining the screening and the neighborhood methods.The frequency of the high-resolution precipitation forecasts is used as the reference frequency,and the frequency of the lowresolution ensemble forecasts is used as the forecast frequency.Based on frequency–matching method,the frequency of rainfall above the rainstorm magnitude increases.The forecast members are then selected by using the typhoon tracks of the short-term predictions,and the precipitation probability is calculated for each member using a combination of the neighbor and the traditional probability statistical methods.Moreover,four landfalling typhoons(i.e.,STY Lekima and STS Bailu in 2019,and TY Hagupit and Higos in 2020)were chose to test the rainfall probability forecast.The results show that the method performs well with respect to the forecast rainfall area and magnitude for the four typhoons.The Brier and Brier skill scores are almost entirely positive for the probability forecast of 0.1–250 mm rainfall during Bailu,Hagupit and Higos(except for 0.1mm of Hagupit),and for<100 mm rainfall(except for 25 mm)during Lekima.

RECENT ADVANCES IN RESEARCH AND FORECASTING OF TROPICAL CYCLONE RAINFALL

In preparation for the Fourth International Workshop on Tropical Cyclone Landfall Processes(IWTCLP-IV), a summary of recent research studies and the forecasting challenges of tropical cyclone(TC) rainfall has been prepared. The extreme rainfall accumulations in Hurricane Harvey(2017) near Houston, Texas and Typhoon Damrey(2017) in southern Vietnam are examples of the TC rainfall forecasting challenges. Some progress is being made in understanding the internal rainfall dynamics via case studies. Environmental effects such as vertical wind shear and terrain-induced rainfall have been studied, as well as the rainfall relationships with TC intensity and structure. Numerical model predictions of TC-related rainfall have been improved via data assimilation, microphysics representation, improved resolution, and ensemble quantitative precipitation forecast techniques. Some attempts have been made to improve the verification techniques as well. A basic forecast challenge for TC-related rainfall is monitoring the existing rainfall distribution via satellite or coastal radars, or from over-land rain gauges. Forecasters also need assistance in understanding how seemingly similar landfall locations relative to the TC experience different rainfall distributions. In addition, forecasters must cope with anomalous TC activity and landfall distributions in response to various environmental effects.

RAPID INTENSIFICATION NEAR LANDFALL OF TYPHOON VICENTE(2012)

Typhoon Vicente(2012) underwent rapid intensification(RI) within 24 h before landfall in China's Mainland. Analysis of the large-scale environment and characteristics of Vicente identifies the aforementioned intensification as classic RI. The process occurred in an environmental flow with a deep-layer shear ranging from 5 ms-1 to 8 ms-1. Convection caused by persistent vertical shear forcing of the vortex was observed primarily in the downshear left quadrant of the storm. However, radar and satellite observations indicate that the northern convection of the inner core of Vicente quickly developed in the down-shear right three hours near landfall.