Study On A Dynamical Statistical Analog Ensemble Forecast Model For Landfalling Typhoon Disasters Over Coastal China

This paper focuses on the dynamical-statistical analog ensemble model for landfalling typhoon disasters(LTD),specifically from the following three aspects:reconstructing a long-time county-level resolution disaster data set to provide a good data base for model construction;secondly,applying the theory of dynamical-statistical analog ensemble forecasting(DSAEF)to landfalling typhoon disasters,conducting simulation experiments for 30 landfalling typhoons from2004-2010 and pre-assessment experiments for 21 landfalling typhoons from2012-2015 to examine the model performance.Finally,the DSAEF＿LTD model improvement study was carried out to address the shortcomings in the pre-assessment results.The main results are as follows:(1)The typhoon disaster data set for 1980-2018 was reconstructed.An appropriate reconstruction methodology was developed,including three main stages:disaster data expansion for individual station,establishment of relationships between disaster and typhoon precipitation and wind,and disaster data set reconstruction.The reconstructed disaster data set data is characterised by a non-significant and slightly decreasing trend in cumulative annual damage.Typhoon disasters mainly occur from July to September each year.All of the particularly severe typhoons are landfall typhoons,and as typhoon damage increases,the maximum intensity of the typhoon during the impact period and the number of days in the impact period gradually increase.For a given damage category,the maximum intensity of a sidewiping typhoon during its impact period is significantly higher than that of a landfalling typhoon of that level.By analysing the tracks of typhoons of different damage categories,it can be seen that typhoons with a northward track may cause relatively more severe damage,regardless of whether they are sidewiping or landfalling.(2)The DSAEF＿LTD model was initially established.The model contained four components: obtaining the forecast track of a target landfalling typhoon;constructing the generalized initial value(GIV);identifying analogs based on the GIV,and assemble typhoon disasters of the analogs.In this early stage of model development,only typhoon track,landfall date,and intensity are considered in GIV.Both simulation and pre-assessment results showed that the model has good simulated and pre-assessed capability for the two important damage categories(medium damage and above,and severe damage and above)of landfalling typhoon disasters,specially for the medium damage and above.Also,TS＿sum of pre-assessment showed a significant positive correlation with the damage of a single typhoon disaster event.The model well captures the spatial distribution characteristics of the main damage,which helps the effective implementation of disaster prevention and mitigation in key areas.In addition,the BS results show that the errors of the experiment results are mainly derived from false alarms.(3)Conducting the DSAEF＿LTD model improvement study: five new ensemble schemes and five new similar boxes were introduced and three sets of improved comparison tests were carried out,showing that TS＿sum is substantially improved when both the similarity box and the ensemble scheme were improved and the corresponding best common scheme for pre-assessment experiments was used.In particular,the improvement is even more pronounced for station-based damage at the heavy category and above and also provides a more accurate description of the spatial distribution of main damage.The improvement is reflected in the reduction of false alarms.The model is only in its initial stages of development and will continue to develop towards maturity in the future as the GIV are enriched and a large number of experiments are carried out.