Thursday, 7 June 2018
Aspen Ballroom (Grand Hyatt Denver)
In many actual cases, the numerical model of wind velocity within the Tropical Cyclone (TC) scale is obviously lower than that of the center wind with subjectively determined intensity, and the TC’s subjective positioning shows deviation from its center position in the numerical model. So this paper, based on the statistical analysis of historical data, investigates using the Generalized Additive Model (GAM) with the TC’s maximum wind speed radius as the dependent variable. Moreover, and then substitutes the estimation result into the modified VBogus model to reconstruct the wind asymmetry distribution of TC by the asymmetry distribution of fitting size. Thus to improve the problem of positioning and intensity deviation in subjective and objective in the forecast editing for TC wind field.
Based on a total of 444 TCs including 12653 time-series of the Northwest Pacific best track TC historical data from 2001 to 2016 by Joint Typhoon Warning Center (JTWC), 8832 time-series above Tropical Depression (TD) were selected to analysis including Latitude (LAT), Longitude (LON), Maximum sustained wind speed (VMAX), Minimum sea level pressure (MSLP) , Pressure of the last closed isobar (RADP), Radius of the last closed isobar (RRP, referred to as TC scale), Radius of max winds (MRD, referred to as TC kernel).The results show:1) The divergence of RADP is the smallest while the TC scale tops among all the elements; the TC scale mainly ranges between 200 km and 500 km (hereinafter referred to as TC scale concentration area);t. The mean of TC kernel is 55.1km, but the maximum and the minimum differ dramatically and both emerge at TD and Tropical Storm (TS). Mostly the TC kernel of Typhoon (TY) or above is 20 km to 40 km. 2) All the TC kernel decreases with intensity in Longitude；the TC kernel of TS or above decreases with intensity in Latitude and the TCs with the same intensity augments has a larger kernel if it’s located further northern，but the TC kernel of TD decreases with intensity between 15°N to 21°N, Whereas other latitudes increase with intensity. All TC scales decreased with intensity enhancement and kernel reduction, but the TCs at the TY level and below have large scale and big kernel, and the TCs at the STY level and above have large scale and small kernel.3) In the TC scale concentration area, the TC kernel and TC scale of each intensity are close to parallel linear distribution. When TC scale is over 500km, the TC kernel increases with the TC scale, then decreases the trend and presents a nonlinear distribution. When the TC scale is within 200km, the TC kernel of TD decrease with the scale, while the TC kernel of other intensity TC are the opposite.
Through the above statistical analysis, it is found that there has a linear and non-linear relationship between the TC kernel and the intensity and the TC scale, so MRD is introduced as a dependent variable, LON, LAT, MSLP and VMAX as independent variables to establish the Generalized Additive Model (GAM) with the GUASSIAN distribution as the variable distribution, and the equation is: MRD~S(VMAX)+S(MSLP)+LAT+LON. S() is a copula function. After many operations and comparisons, the model (Deviance explained) with the explanatory ability of 57.8% is chosen as the optimal model, and the number of valid model samples is 5,734, with the residuals of the Normal distributed, which meet the modeling requirements. Substituting the MRD estimated by GAM and other TC’s element into the modified VBogus model, the tangential wind field distribution within the TC scale is yielded. Finally, by combining the decomposition of the wind field of the numerical model and the tangential wind obtained by the modified VBogus model, the wind field distribution after the reconstruction of TC scale is obtained.
According to the ship observation data, the reconstructed TC wind field within the TC scale from 2001 to 2016 in the Northwest Pacific are compared with the ERA-Interim 10m wind field. The verified results show that the reconstructed TC wind field can improve the small wind velocity of the numerical mode within the TC scale and effectively correct the deviation of the TC center and the subjective positioning. Therefore, the reconstructed TC wind field is more reasonable and it can improve the editing efficiency of the precision grid prediction wind field during the TC period. That have a significant positive effect in the operational platform.
Based on a total of 444 TCs including 12653 time-series of the Northwest Pacific best track TC historical data from 2001 to 2016 by Joint Typhoon Warning Center (JTWC), 8832 time-series above Tropical Depression (TD) were selected to analysis including Latitude (LAT), Longitude (LON), Maximum sustained wind speed (VMAX), Minimum sea level pressure (MSLP) , Pressure of the last closed isobar (RADP), Radius of the last closed isobar (RRP, referred to as TC scale), Radius of max winds (MRD, referred to as TC kernel).The results show:1) The divergence of RADP is the smallest while the TC scale tops among all the elements; the TC scale mainly ranges between 200 km and 500 km (hereinafter referred to as TC scale concentration area);t. The mean of TC kernel is 55.1km, but the maximum and the minimum differ dramatically and both emerge at TD and Tropical Storm (TS). Mostly the TC kernel of Typhoon (TY) or above is 20 km to 40 km. 2) All the TC kernel decreases with intensity in Longitude；the TC kernel of TS or above decreases with intensity in Latitude and the TCs with the same intensity augments has a larger kernel if it’s located further northern，but the TC kernel of TD decreases with intensity between 15°N to 21°N, Whereas other latitudes increase with intensity. All TC scales decreased with intensity enhancement and kernel reduction, but the TCs at the TY level and below have large scale and big kernel, and the TCs at the STY level and above have large scale and small kernel.3) In the TC scale concentration area, the TC kernel and TC scale of each intensity are close to parallel linear distribution. When TC scale is over 500km, the TC kernel increases with the TC scale, then decreases the trend and presents a nonlinear distribution. When the TC scale is within 200km, the TC kernel of TD decrease with the scale, while the TC kernel of other intensity TC are the opposite.
Through the above statistical analysis, it is found that there has a linear and non-linear relationship between the TC kernel and the intensity and the TC scale, so MRD is introduced as a dependent variable, LON, LAT, MSLP and VMAX as independent variables to establish the Generalized Additive Model (GAM) with the GUASSIAN distribution as the variable distribution, and the equation is: MRD~S(VMAX)+S(MSLP)+LAT+LON. S() is a copula function. After many operations and comparisons, the model (Deviance explained) with the explanatory ability of 57.8% is chosen as the optimal model, and the number of valid model samples is 5,734, with the residuals of the Normal distributed, which meet the modeling requirements. Substituting the MRD estimated by GAM and other TC’s element into the modified VBogus model, the tangential wind field distribution within the TC scale is yielded. Finally, by combining the decomposition of the wind field of the numerical model and the tangential wind obtained by the modified VBogus model, the wind field distribution after the reconstruction of TC scale is obtained.
According to the ship observation data, the reconstructed TC wind field within the TC scale from 2001 to 2016 in the Northwest Pacific are compared with the ERA-Interim 10m wind field. The verified results show that the reconstructed TC wind field can improve the small wind velocity of the numerical mode within the TC scale and effectively correct the deviation of the TC center and the subjective positioning. Therefore, the reconstructed TC wind field is more reasonable and it can improve the editing efficiency of the precision grid prediction wind field during the TC period. That have a significant positive effect in the operational platform.
Keywords: Tropical cyclone, Wind distribution, Generalized Additive Model(GAM),The Modified VBogus Model
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner