1. Introduction
China is adjacent to the northwest Pacific and experiences an average of seven to eight landfalling tropical cyclones (LTCs) every year. The disasters caused by LTC rainstorms bring serious casualties and economic losses to China (Chen et al. 2010). For example, Supertyphoon Lekima (2019) caused the daily precipitation in Zhejiang and Shandong Province to exceed historical extremes, causing 14 million people to be affected, with 56 deaths, 14 people missing, and direct economic losses exceeding 50 billion yuan. Therefore, improving the forecasting skill of TC rainstorms is of great significance to the mitigation of typhoon-induced disasters and the protection of human lives and property.
At present, the main approach to improving landfalling typhoon precipitation (LTP) forecast is to further develop and enhance the operational numerical weather prediction (NWP) models. In this regard, some researchers have focused on the development of data assimilation technologies (Zhu et al. 2016; Singh and Tyagi 2019; Wang et al. 2020), while others have sought to improve the parameterization schemes for different physical processes (Yu et al. 2013; Xu et al. 2014) or advance the technology around the method of ensemble forecasting (Hsiao et al. 2013; Hong et al. 2015). Another important way to improve LTP forecasts is to develop a combination of dynamical models with statistical methods (referred to as the dynamical–statistical approach), which can be divided into three categories (Ren and Xiang 2017): The first category involves using TC tracks predicted by dynamical models and historical rainfall observations, and the LTP forecast is obtained from the perspective of the climatic average (Marks et al. 2002; Lee et al. 2006; Lonfat et al. 2007); the second type obtains TC rainfall by adopting TC track forecasts and the rainfall integration from the initial rainfall rates (Kidder et al. 2005; Liu 2009; Ebert et al. 2011); and the third type works by constructing a dynamical–statistical scheme that consists of various internal TC variables and its environmental fields (Li and Zhao 2009; Zhong et al. 2009).
In a series of papers, a research team at the Chinese Academy of Meteorological Sciences has developed the Dynamical–Statistical–Analog Ensemble Forecast for Landfalling Typhoon Precipitation (DSAEF_LTP) model (Ren et al. 2020; Ding et al. 2020; Jia et al. 2020), which was applied to forecast the accumulated land-based precipitation during the typhoon’s lifetime. The core concept is to use the TC track as forecast by the operational NWP model, to compare this track with those of past years’ cyclones over a specified similarity region, to choose closest analog typhoons. Each closest analog typhoon corresponds to a total accumulated precipitation filed over land, which can be assembled to obtain the forecast result for each meteorological rainfall station derived from the rainfall that occurred in the closest analogs. The analogs are chosen through the use of generalized initialized values (GIVs) which include variables that may influence landfalling typhoon precipitation available in both the forecast model output and in the analogs or historic typhoons. In the original model (Ren et al. 2020), the GIV included the typhoon track and landfall season, which shows the forecast ability of the DSAEF_LTP model is comparable to that of the operational NWP models (ECMWF, GFS, T639) for TC accumulated precipitation of ≥250 and ≥100 mm. Subsequently, Ding et al. (2020) introduced TC intensity, and Jia et al. added an expanded suite of similarity regions (Jia et al. 2020) and ensemble schemes (Jia et al. 2022) to further improve the forecast skill.
To date, the DSAEF_LTP model has been applied to forecasting the accumulated land-based precipitation during the typhoon’s lifetime. It has thus been a multiday forecast, beginning from a time before landfall, and extending over the several days post-landfall while the cyclone is still tracked. To ameliorate or prevent the impact of typhoon disasters, the Chinese Meteorological Agency is also motivated for the improvement of daily precipitation forecasts. The current paper is the first investigation of the application of the DSAEF_LTP to daily precipitation, hereafter referred to as the DSAEF_LTP_D model (“_D” for daily).
The second purpose for the current study is to investigate the influence of incorporating TC translation speed into the GIV to determine whether that will improve forecasts. It has long been known that slow moving weather systems are conducive to heavy rainfall at particular locations, due simply to the precipitation bands having a longer duration over the location (For example, Maddox et al. 1979). In recent decades it has been documented that this principle also applies to tropical cyclones. Studies have shown that the slower the TC movement, the longer is their influence time, and the greater is the impact of heavy rain (Emanuel 2017; van Oldenborgh et al. 2017; Yamaguchi et al. 2020; Titley et al. 2021). Ankur et al. (2020) showed that the heavy rainfall shifts from the TC rear sector to the forward sector relative to the direction of the TC movement as the translation speed increase [from the slow (translation speed ≤ 10.8 km h−1) to fast movers (translation speed > 14.4 km h−1)]. In China, in 1987 Supertyphoon Lynn, Typhoon No. 8710, recorded a daily maximum precipitation of 1151.9 mm, and in 2009 Typhoon Mora (No. 0908) recorded a maximum of 1623.5 mm. Both systems recorded these heavy precipitation amounts when the typhoons suddenly decelerated and stagnated (Chen and Xu 2017). In the United States in 2017 Hurricane Harvey stalled over southeastern Texas for several days, bringing about a rainfall over 1500 mm in Nederland, Texas, the highest TC-related rainfall total ever recorded in the United States (van Oldenborgh et al. 2017; Bosma et al. 2020). These and other recent findings are the motivation for the experiments reported here on the influence of cyclone speed of movement on the DSAEF forecast system.
This paper reports and discusses the results from daily precipitation simulation experiments for the heavy rainfall associated with Lekima, which produced serious casualties and property losses in many provinces in China. Following this introduction, section 2 introduces the DSAEF_LTP_D model. Section 3 introduces the experimental design. The results are analyzed in section 4, and then summarized and discussed in section 5.
2. The DSAEF_LTP_D model
The DSAEF_LTP_D model is a combination of dynamical and statistical methods. The DSAEF_LTP_D model is a combination of dynamical and statistical methods, whose goal is to find the closest analog TCs from the past years’ cyclones, and produce an ensemble forecast for each meteorological rainfall station derived from the rainfall that occurred in the closest analogs.
Figure 1 is a schematic diagram of the forecasting of daily precipitation in the DSAEF_LTP_D model. The black box is a given similarity region. Points A and B are the two diagonal points of the similarity region. The red line represents the complete track of the target TC (observed track + forecast track), on which point A1 is the location of the target TC at which the forecasting of daily precipitation begins, that is, the 24-h accumulated precipitation after A1 is to be predicted. The brown line indicates a historical TC track similar to the target TC, on which the blue open circles denote historical TC locations at 6-h intervals. When forecasting daily precipitation on a specified day, the point A1 is determined, and the distances between the historical TC locations and the point A1 are calculated, in which the shortest distance is denoted as d0 and the corresponding TC location on the historical track is recorded as point C1.To compare the similarity in the translation speed, the translation speed of the target TC adopts the average translation speed within 24 h after point A1, and the historical TC’s translation speed uses the average translation speed within 24 h after point C1.
Schematic diagram of the forecasting of daily precipitation by the DSAEF_LTP_D model.
Citation: Weather and Forecasting 37, 11; 10.1175/WAF-D-21-0135.1
The black box is a given similarity region with two diagonal points A and B. The red line represents the complete track of the target TC, on which A1 on it is the point beginning to predict the daily precipitation. The brown line is a historical TC track similar to the target TC, on which the blue open circles denote historical TC locations at 6-h intervals, with point C1 being the point closest to point A1. The length of the straight line between point C1 and A1 is the shortest distance between the historical TC and the target TC, denoted as d0.
In this model, the input variables are the parameters shown in Table 1. The DSAEF_LTP_D model involves four steps: obtaining TC forecast track, construction of generalized initial values (GIVs), recognition of similarity and ensemble forecast of TC precipitation. The step 3 recognition of similarity includes the recognition of TC track similarity and other similarity characters, where recognition of TC track similarity considers the influence of the terrain and underlying surface on TC precipitation. The more similar the track between the two TCs, the more similar the underlying surface and the influence of the terrain on the TCs. The specific steps are as follows:
List of parameters and corresponding significances of the DSAEF_LTP_D model.
-
The forecast track after the point A1, (i.e., the point beginning to forecast the daily precipitation) obtained from the NWP model and the observed track (prior to the point A1) are merged into a complete track as the target TC track (the red line in Fig. 1).
-
Construction of generalized initial values (GIVs). GIV includes the variables that may influence landfalling TC precipitation. These are physical factors including both TC internal variables (e.g., translation speed, intensity) and environmental variables (e.g., vertical wind shear, subtropical high, monsoon). In this paper, the GIV of the DSAEF_LTP_D model are TC track and translation speed.
-
Recognition of track similarity between historical TCs and target TC. In a given similarity region (determined by P1 in Table 1), the TC Track similarity area index (TSAI) (F. M. Ren et al. 2018) (determined by P1, P2, and P3) is calculated between the target TC track and all historical TC tracks, and organized from low to high, where the lower the TSAI, the higher the track similarity should be. The distance between the historical TC locations, which are at 6-h intervals, and the point A1 are calculated. The shortest distance is denoted d0, and the corresponding historical TC location is denoted as point C1. The 24-h accumulated precipitation field after point C1 is selected as the historical precipitation field similar to that of the target TC. Even if the historical TC track is closely similar to the target TC track (i.e., the TSAI is small), the distance d0 between points A1 and C1 may be large. In this case, the 24-h accumulated precipitation field after point C1 will have large differences with the corresponding 24-h precipitation field after point A1. Therefore, the DSAEF_LTP_D model will eliminate those historical TCs whose d0 values are greater than a certain value, determined by P4. At this stage, TC track similarity judgment (determined by P1, P2, P3, and P4) has been completed.
-
Recognition of other similarity characters. In this study, the TC translation speed is included as a GIV through parameter P5 in Table 1. There are seven parameter values to distinguish the similarity in translation speed, where 1 represents that the historical TC’s translation speed is unlimited, 2 represents that the historical TC’s translation speed is less than or equal to the target TC’s translation speed, and 3–7, respectively, represent the absolute value of the translation speed difference between the historical TCs and the target TC is within 17, 14, 11, 8, and 5 km h−1. For a specific parameter value, the historical TCs that meets the parameter conditions will be retained to make ensemble forecast.
Eventually, N (determined by P6) top-ranked historical TCs closest to the target TC are selected.
-
Ensemble forecast of TC precipitation. Finally, an optimized ensemble forecast scheme (determined by P7) is adopted to assemble the daily precipitation fields corresponding to the N top-ranked historical TCs, and the predicted precipitation for the target TC is obtained. In this step, the objective synoptic analysis technique (OSAT) is used to partition precipitation generated by the TCs (Ren et al. 2001, 2007; Wang et al. 2006).
To make the forecasting procedure clearer, a simple example is given. Supposing the parameters (P1–P7) in the Table 1 are 2, 3, 2, 8, 7, 4, and 1, respectively. This means that a rectangle with diagonal points A and B, where A is the TC location at 12 h prior to point A1 and B is the TC location at 18 h after point A1, is constructed (P1 = 2). In this rectangle region, the TSAI (determined by P1, P2, and P3) is calculated between the target TC track and all historical TC tracks, and the historical TCs with similar tracks are identified. Then, among the identified historical TCs with similar tracks, the historical TCs with d0 > 110 km (P4 = 8) and the translation speed difference of >5 km h−1 (P5 = 7) between the target TC and historical TCs should be eliminated. Eventually, four (P6 = 4) top-ranked historical TCs closest to the target TC are selected, and their daily precipitation field will be assembled by max value of each rain gauge station (P7 = 1) to obtain the predicted precipitation field of target TC. This procedure produces the forecast for one scheme.
3. Experimental design
a. Typhoon Lekima precipitation and track
Figure 2 shows the track of Typhoon Lekima and the distribution of Lekima-produced daily precipitation, as identified by OSAT, over each of five days: produced by Lekima from 8 to 12 August 2019. Lekima made landfall in Zhejiang at 1745 UTC 9 August 2019 (Fig. 2b) with winds of 52 m s−1 (supertyphoon). It then moved northwestward, passing through Zhejiang and Jiangsu Province (Fig. 2c) before landfalling for a second time in Shandong at 1250 UTC 11 August (Fig. 2d). Finally, it passed through the western part of the Shandong Peninsula and circled in Laizhou Bay on 12 August (Fig. 2e). Over the five days, from 8 to 12 August, several provinces, such as Zhejiang, Shanghai, Jiangsu, Shandong, and Liaoning, experienced rainstorms (24-h accumulated precipitation between 50 and 100 mm), heavy rainstorms (24-h accumulated precipitation between 100 and 250 mm) and torrential rainstorms (24-h accumulated precipitation ≥ 250 mm). Several stations in Zhejiang and Shandong broke their historical extreme records. The maximum daily precipitation of a single station on these five days was 80.7, 379.5, 367.3, 232.1, and 128.9 mm, respectively.
Observed 24-h accumulated rainfall induced by Supertyphoon Lekima (mm) for (a) 0000 UTC 8 Aug–0000 UTC 9 Aug 2019, (b) 0000 UTC 9 Aug–0000 UTC 10 Aug 2019, (c) 0000 UTC 10 Aug–0000 UTC 11 Aug 2019, (d) 0000 UTC 11 Aug–0000 UTC 12 Aug 2019, and (e) 0000 UTC 12 Aug–0000 UTC 13 Aug 2019. The track of Lekima is shown by the black line, and its locations at 6-h intervals each day are shown by the red marks.
Citation: Weather and Forecasting 37, 11; 10.1175/WAF-D-21-0135.1
Due to the wide distribution and high intensity of the heavy rainfall (24-h accumulated precipitation of ≥50 and ≥100 mm) from 9 to 11 August (Figs. 1b–d), these 3 days were selected as the research period to conduct daily precipitation simulation experiments for Lekima, in which the 24-h accumulated precipitation from 0000 to 0000 UTC the next day was forecast.
b. Experimental design
To examine the impact of TC translation speed in the DSAEF_LTP_D model, two groups of experiments were designed: experiment 1 involved the DSAEF_LTP_D model containing only one GIV, i.e., TC track (viz., trajectory of the TC center, not including the forward speed), referred to as DSAEF_LTP_D-1; and experiment 2 contained the TC track and TC translation speed, referred to as DSAEF_LTP_D-2.
To compare the forecast performance of the DSAEF_LTP model with that of NWP models, corresponding TC rainfall forecast data were obtained from three global forecast systems—namely, the European Centre for Medium-Range Weather Forecasts (ECMWF) model, the National Centers for Environmental Prediction Global Forecast System (GFS) model, and the Global/Regional Assimilation Prediction System (GRAPES) model run by the CMA; as well as one regional model—the Shanghai Meteorological Service Weather Research and Forecasting (WRF) Model ADAS Real-Time Modeling System (SMS-WARMS). The ECMWF, GFS, GRAPES, and SMS-WARMS models have the horizontal resolution of 0.125° × 0.125°, 0.25° × 0.25°, 0.25° × 0.25°, and 0.09° × 0.09°, respectively. To process the data uniformly, the horizontal resolutions of the four models have been converted to 0.1° × 0.1° by bilinear interpolation.
In addition, the tracks of Lekima—both the observed track and operational NWP model forecast track—were obtained from the China Meteorological Administration (CMA)/National Meteorological Information Center (NMIC). The historical best tracks during 1960–2019, including TC center positions at 6-h intervals, were obtained from the Shanghai Typhoon Institute (Ying et al. 2014). The observed precipitation data during 1960–2019 were also obtained from the CMA/NMIC. The rain gauge stations with missing data for more than 5 years and the rain gauge stations in Xinjiang province that were not affected by the typhoon have been eliminated. Finally, 2004 station data over mainland China reserved, and the quality of these observational station data has been controlled (Feng et al. 2004; Han and Li 2012; Z. H. Ren et al. 2018).
4. Results
a. Comparison of two experiments
As described above, two sets of simulation experiments were conducted on Lekima for the 3 days (9–11 August 2019).The threat scores (i.e., TS50 and TS100) from the 28 171, 17 318, and 4299 forecast schemes on the 3 days from DSAEF_LTP-1 are shown by black dots in Figs. 3a–c, and the 176 698, 87 137, and 15 672 forecast schemes in DSAEF_LTP-2 are shown by blue dots in Figs. 3d–f. Each dot in the figures represents a forecast scheme, and the red dot is the best forecast scheme determined by the maximum TS100 + TS50, whose corresponding parameter values are listed in Table 2. Figures 3a,d and Figs. 3b,e show that the forecast performance of the DSAEF_LTP_D model on 9 and 10 August has been improved by the addition of TC translation speed into the model. Quantitatively, TS100 + TS50 values increase from 0.8225 (0.4 + 0.4225), 0.8016 (0.2538 + 0.5478) to 0.9076 (0.5417 + 0.3659), and 0.8581 (0.2804 + 0.5777), respectively. However, as shown in Figs. 3c and 3f for 11 August, the model has the same forecast performance for both experiments with an identical value of TS100 + TS50 equal to 0.3021 (0.0889 + 0.2132).
Scatterplots of threat scores (TS50–TS100) from (a) 28 171 forecast schemes on 9 Aug, (b) 17 318 forecast schemes on 10 Aug, and (c) 4299 forecast schemes on 11 Aug of DSAEF_LTP_D-1 in black dots; and (d) 176 698 forecast schemes on 9 Aug, (e) 87 137 forecast schemes on 10 Aug, and (f) 15 672 forecast schemes on 11 Aug of DSAEF_LTP_D-2 in blue dots. TS50 and TS100 represent the threat scores for predicting accumulated rainfall of ≥50 and ≥100 mm, respectively. The red dots are the best forecast scheme of the DSAEF_LTP_D model.
Citation: Weather and Forecasting 37, 11; 10.1175/WAF-D-21-0135.1
Optimized parameters of the best schemes of the DSAEF_LTP_D models.
To evaluate the forecast performance of the DSAEF_LTP_D model for the 24-h accumulated precipitation of ≥100 and ≥50 mm, Fig. 4 compares the results from the two groups of experiments with the NWP models. On 9 August, for the 24-h accumulated precipitation ≥ 50 mm, DSAEF_LTP_D-1, has the highest TS50 (0.4225), ranking the first, with ECMWF being best of the NWP models [ECMWF (TS50 = 0.4000)]. For the 24-h accumulated precipitation ≥ 100 mm, TS100 of DSAEF_LTP_D-2 has significantly improved over DSAEF_LTP_D-1, increasing from 0.4 to 0.5417 and exceeding the four NWP models.
Threat scores (TS50, TS100, and TS100 + TS50) produced by the best scheme of the DSAEF_LTP_D models, compared with those of the four dynamical models (i.e., ECMWF, GRAPES, GFS, and SMS-WARMS) from 9 to 11 Aug.
Citation: Weather and Forecasting 37, 11; 10.1175/WAF-D-21-0135.1
On 10 August, the TS values of DSAEF_LTP_D-2 increase over version D-1 both in TS50 and TS100, of which TS50 has increased from 0.5478 to 0.5777 and ranks second behind ECMWF, and TS100 increases from 0.2538 to 0.2804 and is better than SMS-WARMS (0.1687), which performs the best among the NWP models.
For the heavy rainfall on 11 August, the best forecast performer is SMS-WARMS, with TS50 and TS100 values of 0.6028 and 0.5238, respectively. The DSAEF_LTP_D model has the same forecast skill after adding the translation speed and is inferior to all the NWP models, whose TS50 and TS100 values are 0.2132 and 0.0889, respectively. Generally, for the heavy rainfall associated with Lekima on 9 and 10 August, the DSAEF_LTP_D model with translation speed added is comparable to the NWP models, but for the rainfall on 11 August, the DSAEF_LTP_D model forecast performance is poor.
b. Analysis of DSAEF_LTP_D model forecast results
To further understand the forecast performance of the DSAEF_LTP_D model, Figs. 5b and 5c show the forecast precipitation on 9 August of the DSAEF_LTP_D-1 and DSAEF_LTP_D-2, respectively. Figures 5d–k show the precipitation distribution of the selected similar historical TCs, where Figs. 5d–g are the four ensemble members of DSAEF_LTP_D-1, and Figs. 5f–k are the six ensemble members of DSAEF_LTP_D-2 (Figs. 5f,g are the common ensemble members of DSAEF_LTP_D-1 and DSAEF_LTP_D-2). The historical TCs (TC200515 selected by the best scheme of DSAEF_LTP_D-1, and TC201212 and TC201509 selected by the DSAEF_LTP_D-2) are more consistent with the observations, which is the key for the DSAEF_LTP_D model successfully predicting the heavy rainfall center in eastern Zhejiang. Comparing Figs. 5b and 5c, which show the false-alarm stations (i.e., where the forecasted precipitation reaches a magnitude of ≥100 mm but the observed precipitation does not.) in blue dots, by introducing the translation speed, the number of false-alarm stations decreases from seven (Fig. 5b, five are concentrated in the blue box and two are outside it) to one (Fig. 5c), resulting in TS100 being significantly improved.
24-h accumulated precipitation (mm) from 0000 UTC 9 Aug to 0000 UTC 10 Aug 2019 associated with Lekima from (a) rain gauge observations, and the best scheme of (b) DSAEF_LTP_D-1 and (c) DSAEF_LTP_D-2. (d)–(g) The 24-h accumulated precipitation (mm) of four ensemble members of DSAEF_LTP_D-1 (i.e., TC200515, TC201419, TC196405, and TC199507). (f)–(k) The 24-h accumulated precipitation (mm) of six ensemble members of DSAEF_LTP_D-2 (i.e., TC196405, TC199507, TC198511, TC198911, TC201212, and TC201509). The black lines denote the actual track of Lekima in (a)–(c) and ensemble members in (d)–(k); the red lines represent the predicted track of Lekima at 0000 UTC 9 Aug 2019; and the orange rectangles depict the similarity region chosen by the best scheme.
Citation: Weather and Forecasting 37, 11; 10.1175/WAF-D-21-0135.1
Figures 6b and 6c show the distribution of the forecast precipitation on 10 August of DSAEF_LTP_D-1 and DSAEF_LTP_D-2, and Figs. 6d,e show their difference between the predicted and observed values. Figures 6f–n show the precipitation distribution of selected similar historical TCs, where Figs. 6f–l are the seven ensemble members of DSAEF_LTP_D-1, and Figs. 6j–n are the five ensemble members of DSAEF_LTP_D-2 (Figs. 6h–j are the common ensemble members of the DSAEF_LTP_D-1 and DSAEF_LTP_D-2). A successful TC track prediction (Fig. 6b or 6c) is the basis of the good performance of the DSAEF_LTP_D model for predicting the heavy rainfall associated with Lekima. Comparing Figs. 6d and 6e, from the introduction of TC translation speed the forecast performance of the DSAEF_LTP_D model for heavy rainfall improves. This is mainly manifested in eastern Shandong and Zhejiang (i.e., the area in the blue box in Fig. 6d), where for the 24-h accumulated precipitation ≥ 100 mm, the number of false-alarm stations decreases, the difference being because the historical TC199714 (Fig. 6i), selected by the best scheme of DSAEF_LTP_D-1, leads to overestimated precipitation amounts in these regions. Also in central Shandong for 24-h precipitation ≥ 50mm, as we can see from Figs. 6g and 6m, the daily precipitation distribution produced by TC200425(2) [TC200425 and TC200425(2) refer to the 24-h accumulated precipitation after 0600 and 1200 UTC 13 September, respectively] selected by DSAEF_LTP_D-2 is farther north than that of TC200425 selected by DSAEF_LTP_D-1, which is more accurate for predicting the center of heavy rainfall in central Shandong, although the predicted precipitation amounts are lower than the observations. It is worth noting that the best scheme of DSAEF_LTP_D-2 focuses on the similarity of TC track with a wider range (from the southeast coast to the north), while the best scheme of DSAEF_LTP_D-1 mainly focuses on the TC track similarity around the southeast coast, which results in the ensemble members of DSAEF_LTP_D-1 tending to forecast the heavy rainfall closer to the southeast coast compared with DSAEF_LTP_D-2.
24-h accumulated precipitation (mm) from 0000 UTC 10 Aug to 0000 UTC 11 Aug 2019 associated with Lekima from (a) rain gauge observations, and the best scheme of (b) DSAEF_LTP_D-1 and (c) DSAEF_LTP_D-2, the difference between the predicted and observed precipitation values (mm) of (d) DSAEF_LTP_D-1 and (e) DSAEF_LTP_D-2. (f)–(l) The 24-h accumulated precipitation (mm) of seven ensemble members of DSAEF_LTP_D-1 (i.e., TC198511, TC200425, TC200509, TC199714, TC199022, TC200515, and TC201817). (j)–(n) The 24-h accumulated precipitation (mm) of five ensemble members of DSAEF_LTP_D-2 [i.e., TC199022, TC200515, TC201817, TC200425(2), and TC200509(2)]. The black lines denote the actual track of Lekima in (a)–(c) and ensemble members in (f)–(n); the red lines represent the predicted track of Lekima at 0000 UTC 10 Aug 2019; and the orange and purple rectangles depict the similarity regions chosen by the best scheme of DSAEF_LTP_D-1 and DSAEF_LTP_D-2, respectively.
Citation: Weather and Forecasting 37, 11; 10.1175/WAF-D-21-0135.1
Figure 7b shows the forecast result of the best scheme of the DSAEF_LTP_D-1 or DSAEF_LTP_D-2 model on 11 August, which has a large deviation from the observations (Fig. 7a). Not only does it fail to predict the heavy rainfall center in central Shandong, but there are also overestimations of large-scale heavy rainfall in southern Liaoning. Furthermore, the model does not gain any improvement with the inclusion of TC translation speed. This is probably related to the similarity region selected by the best scheme. The similarity regions (P1) are identified by point A1 (the point at which the forecasting of daily precipitation begins—that is, the TC location at 0000 UTC 11 August) and the farthest forecast point of the track (i.e., the end of the TC track predicted by NWP). The southeast corner of the similarity region selected by the best scheme is at point A1 and the northwest corner is the TC location at 36 h after point A1. As seen in Fig. 7b the track of Lekima is northerly, which leads to a narrower similarity region (the orange rectangle). In this narrow similarity region, there are fewer TCs passing through the region, which limits the selection of similar historical TCs, leading to poor performance of the model.
24-h accumulated precipitation (mm) from 0000 UTC 11 Aug to 0000 UTC 12 Aug 2019 associated with Lekima from (a) rain gauge observations and (b) the best scheme of DSAEF_LTP_D-1 or DSAEF_LTP_D-2. The black lines denote the actual track of Lekima; the red line is the predicted track of Lekima at 0000 UTC 11 Aug 2019, and the orange rectangles depict the similarity regions chosen by the best scheme of DSAEF_LTP_D-1 or DSAEF_LTP_D-2.
Citation: Weather and Forecasting 37, 11; 10.1175/WAF-D-21-0135.1
c. Improvement in the DSAEF_LTP_D model similarity regions
According to the analysis in the prior sub section, the forecast performance of the DSAEF_LTP_D model for heavy rainfall at the magnitudes of ≥100 and ≥50 mm on 11 August is poor, which may be due to the limitation caused by narrow similarity regions. Therefore, further improvement in the similarity regions may be an effective way to improve the forecast skill of the DSAEF_LTP_D model.
The similarity region is a rectangle region where the extent of TC track similarity and TC translation speed similarity between target TC and historical TCs are identified. When forecasting the daily precipitation for a specified day, a similarity region of daily scale, which can as much as possible contain the track TC moves in this day, needs to be constructed. To address this, the daily translation distances of 1059 TCs affecting China from 1960 to 2019 were counted over each of the multiple days of the TC lifetime amounting to a total sample of 7407 days. The results are shown in Table 3, where the average value of the TC daily distance is 471.11 km and 96% of the TCs’ daily translation distances are within 1000 km. Therefore, the average and 96% quantile values of the TC daily translation distance, which can contain most of TC tracks in a day, were referenced to design the size of the similarity region with side length of 500 and 1000 km. And considering the newly added similarity regions need to make up for the deficiency that the similarity regions are too narrow due to the straight TC tracks when TC moving toward the north or west, the similarity regions are designed as squares. Finally, a 500-km2 area was selected as the medium similarity region, and a 1000-km2 area as the largest similarity region. Figure 8 is a schematic diagram of five newly added similarity regions. There are 21 similarity region values in the initial DSAEF_LTP_D model (P1 in Table 1). The 22nd scheme of P1 is designed to make a 1000-km2 frame (A1B1C1D1) with point A1 (the point at which the forecasting of 24-h accumulated precipitation begins) as the southeast corner of the similarity region, and a medium-sized 500-km2 frame (A2B2C2D2) is regarded as the 23rd scheme. Due to the complexity of TC track, the similar regions were moved around the TC track on the day. Then, the 24th, 25th, and 26th schemes are obtained by moving A1B1C1D1 to the right, down, and down to the right to A3B3C3D3, A4B4C4D4, and A5B5C5D5, respectively.
Schematic diagram of five newly added similarity regions.
Citation: Weather and Forecasting 37, 11; 10.1175/WAF-D-21-0135.1
Statistics of the daily translation distance of tropical cyclones.
After adding five new similarity regions, the forecast performance of DSAEF_LTP_D-2 further improves for the heavy rainfall associated with Lekima on 9 and 11 August. The improvement on 9 August is small, with TS100 + TS50 increasing by only 0.0091. However, the forecast skill for heavy rainfall on 11 August is significantly improved, with TS100 + TS50 increasing from 0.3021 to 0.4286 (an increase of 41.87%), in which TS100 and TS50 increase from 0.0889 and 0.2132 to 0.0930 and 0.3356, respectively (Fig. 9). However, there is still a deviation between the forecast results of the DSAEF_LTP_D model and the NWP models, and the TS100 + TS50 of ECMWF, GRAPES, GFS, and SMS-WARMS are 0.8504, 0.4511, 0.7258, and 1.1266, respectively. In the DASEF_LTP_D-2, only two physical factors, TC track and TC translation speed, have been considered, without considering the similarity of environmental factors. While the NWP models contains various complex physical processes, considering the influence of circulation pattern. The circulation pattern may have a great influence on the precipitation on 11 August, resulting in a deviation between the prediction results of the DASEF_LTP_D-2 and the NWP models.
The sum of threat scores (TS100 + TS50) of Lekima produced by the best scheme of the DSAEF_LTP_D model, in which the blue bars are the TS100 + TS50 values of DSAEF_LTP_D-2 and the orange bars indicate the growth in TS100 + TS50 with the five newly added similarity regions.
Citation: Weather and Forecasting 37, 11; 10.1175/WAF-D-21-0135.1
For DSAEF_LTP_D-2, the improvement on 11 August indicates that the newly added similarity regions mainly make up for the deficiency of the narrow similarity regions due to the northerly TC track. Comparison of Figs. 7b and 10a, indicates that for 11 August we can see that introduction of new similarity regions for DSAEF_LTP_D-2 alleviates the phenomenon of overestimating the 24-h accumulated precipitation amounts in southern Liaoning. Specifically, the values of P1–P7 of the best scheme are 23, 1, 1, 7, 1, 3, and 1, (the corresponding meanings are shown in Table 1), from which two main changes can be found. First, the number of historical TCs that have similar tracks to Lekima reaches nine, while the original narrow similarity regions can only screen out three TCs. Second, the translation speed similarity P5 changes from 1 (unlimited) to 7 (the difference between the target TC and the historical TC is within 5 km h−1), making the translation speed of screened historical TCs and the target TC (Lekima) closer, so that the forecast performance can be improved. However, it is worth noting that compared with the forecast results of the four NWP models (Figs. 10b–e), the DSAEF_LTP_D model (Fig. 10a) still has a certain deviation from the observation (Fig. 7a) in forecasting the heavy rainfall.
24-h accumulated rainfall (mm) from 0000 UTC 11 Aug to 0000 UTC 12 Aug in 2019 associated with Lekima from (a) the best scheme of DSAEF_LTP_D-2 with newly added similarity regions, (b) ECMWF, (c) GRAPES, (d) GFS, and (e) SMS-WARMS. The black lines denote the actual track of Lekima, and the orange rectangle depicts the similarity regions chosen by the best scheme of DSAEF_LTP_D-2 with newly added similarity regions.
Citation: Weather and Forecasting 37, 11; 10.1175/WAF-D-21-0135.1
On 9 August 2019 (Fig. 11a), the structure of Lekima center was complete, where the center of 500-hPa height coincided with the center of 850-hPa wind, and the TC intensity was strong.On August 10 (Fig. 11b), the dry and cold air behind the westerly trough began to invade from the west side of Lekima, and the 850-hPa wind speed near the TC center area decreased, especially the wind speed on the west side of Lekima, resulting in the prominent asymmetric structure of Lekima. On August 11 (Fig. 11c), Lekima moved to the Shandong Peninsula. The North China westerly trough was broken, and its southern section was merged into Lekima’s circulation. The cold air began to invade the typhoon body, and Lekima experienced extratropical transition on this day. The high value area of its low-level wind speed is in the north of Lekima, and the southeast jet with warm air from the Bohai Sea and the northerly cold air behind the westerly trough intersected in Shandong Province, causing the heavy rainfall. Among them, the invasion of cold air plays an important role in the generation of heavy rainfall. For the DSAEF_LTP_D model, it can successfully capture the similar case of extratropical transition—TC0509 (Di et al. 2008), so the precipitation forecast result on 11 August includes the impact of this factor by the ensemble of precipitation fields of similar historical TCs. However, the forecast performance of the model needs to be further improved by introducing environmental factors to directly reflect the impacts of baroclinic processes or extratropical transition.
500-hPa geopotential height (contours; gpm) and 850-hPa wind (the arrow denotes velocity; shading denotes wind speeds more than 20 m s−1) at (a) 1200 UTC 9 Aug, (b) 1200 UTC 10 Aug, and (c) 1200 UTC 11 Aug 2019.
Citation: Weather and Forecasting 37, 11; 10.1175/WAF-D-21-0135.1
d. Simulation forecast experiment of 10 samples in 2018
To examine the simulation effect of the DSAEF_LTP_D model on multiple samples, 10 TCs in 2018 were selected. The three daily precipitation forecast intervals include the day TC making landfall, and the days before and after it, from 0000 to 0000 UTC the next day. The best scheme of the model is the one with the largest average TS100 + TS50 of the common scheme of 10 TCs.
Figure 12 compares the average TS of 10 TCs from the best schemes of the DSAEF_LTP_D-1 and DSAEF_LTP_D-2 with the NWP models. Among all models, the regional model SMS-WARMS performs much better than other models. After introduction of TC translation speed, the DSAEF_LTP_D model has significantly improved on the day before TC landfall, where the TS100 + TS50 increases from 0.1070 (0.0333 + 0.0737) to 0.2958 (0.1000 + 0.1958), ranking second only to the regional model SMS-WARMS. On the day of landfall, the TS100 + TS50 increases from 0.2994 (0.0944 + 0.2050) to 0.3141 (0.0890 + 0.2251), ranking fourth, which is slightly inferior to the global models ECMWF and GFS. However, on the day after landfall, the forecast ability of the DSAEF_LTP_D-2 model is the same as DSAEF_LTP_D-1 and its forecast performance is poor, which is far from that of the dynamical models, except GRAPES.
Average threat scores (TS50, TS100, and TS100 + TS50) of 10 TCs produced by the best scheme of the DSAEF_LTP_D models, compared with those of the four dynamical models (i.e., ECMWF, GRAPES, GFS, and SMS-WARMS) on the day TC making landfall and the days before and after it.
Citation: Weather and Forecasting 37, 11; 10.1175/WAF-D-21-0135.1
5. Conclusions and discussion
This study has introduced a version of the Dynamical–Statistical–Analog Ensemble Forecast for Landfalling Typhoon Precipitation model applicable to daily or 24-h precipitation accumulations, the DSAEF_LTP_D model. Two sets of experiments have been conducted for versions of the model with and without TC translation speed in the GIVs, DSAEF_LTP_D-1 and DSAEF_LTP_D-2. The experiments are for the simulation of forecasts for 24-h accumulated precipitation exceeding the thresholds of 100 and 50 mm associated with Lekima (2019) over each of three days from 9 to 11 August 2019. The major conclusions are as follows:
-
After the introduction of TC translation speed into the DSAEF_LTP_D model, the forecasting performance of the best schemes for heavy rainfall on 9 and 10 August improved, with TS100 + TS50 increasing from 0.8225 (0.4 + 0.4225) and 0.8016 (0.2538 + 0.5478) to 0.9076 (0.5417 + 0.3659) and 0.8581 (0.2804 + 0.5777). On these two days the DSAEF_LTP_D-2, incorporating translation speed, can capture the heavy rainfall center more accurately. However, the forecasting skill is the same as before when forecasting the heavy rainfall on 11 August.
-
Compared with four NWP models (ECMWF, GFS, GRAPES, and SMS-WARMS), for the heavy rainfall on 9 and 10 August, the TS100 + TS50 value of DSAEF_LTP_D-2 is comparable to the best performer of the NWP models (ECMWF), while the performance of the DSAEF_LTP_D model for predicting the heavy rainfall on 11 August is poor.
-
The northward track of Lekima on 11 August resulted in a narrow similarity region, which restricted the selection of similar historical TCs. Five newly added similarity regions made up for this deficiency, and the forecast performance of DSAEF_LTP_D-2 for the heavy rainfall on 11 August was further improved, with TS100 + TS50 increasing from 0.3021 to 0.4286, an increase of 41.87%.
-
The DSAEF_LTP_D model is applied in 10 TCs in 2018. After introduction of TC translation speed, the forecast performance of the model has been improved on the days of TC landfall and before it, but not on the day after landfall.
Prior to the current study, the development of the DSAEF model has been directed toward total (multiday) accumulated precipitation after typhoon landfall. This study has shown promising results for the evolution of the precipitation over individual days. The improvement in the DSAEF_LTP_D model after adding TC translation speed and new similarity regions suggests that there are two main ways to improve the model—namely, introducing as many relevant variables as possible to the GIV in this model [including internal characteristics of TCs (e.g., TC size and intensity) and environmental conditions (e.g., the westerly trough, vertical wind shear, subtropical high, low-level jet)] and enriching the values of the model’s parameters. The analysis of the precipitation fields for 11 August, the third day of forecasts, showed that the environmental conditions, such as baroclinic processes and extratropical transition, may be important factors affecting the model’s forecast result; for example, the combination of the westerly trough and cold air influenced the precipitation produced by Lekima. To directly reflect the impacts of baroclinic processes or extratropical transition, the DSAEF_LTP_D model needs to construct the similarity in the environmental field to further improve the forecast performance.
As the first step to apply DSAEF theory (Ren et al. 2020) to daily precipitation forecast, this study focuses on one case, i.e., Lekima, to build the framework of the DSAEF_LTP_D model. In addition, this study helps to discover details that need to be improved, such as the similarity regions. The experiment of 10 TCs in 2018 shows the forecast performance of the DSAEF_LTP_D model is comparable to NWP models on the day before TC landfall, slightly inferior to the global models on the day of landfall, and performs poorly on the day after landfall. Nevertheless, the model is promising by incorporating more variables that influence TC rainfall. In this study, only the translation speed was introduced, and only preliminary experiments were carried out on multiple TC samples. In the future, large-sample experiments are needed to examine which daily precipitation forecast interval begins to lose value (i.e., which daily precipitation forecast interval the model is suitable for), and the forecasting accuracy after introducing other variables, especially for real-time application.
Acknowledgments.
The authors would like to express thanks to the three anonymous reviewers for their constructive comments for the improvement of the paper. This work was supported by the National Key R&D Program of China (Grant 2019YFC1510205), the Hainan Provincial Key R&D Program of China (SQ2019KJHZ0028), the National Natural Science Foundation of China (Grant 41675042), and the Jiangsu Collaborative Innovation Center for Climate Change.
REFERENCES
Ankur, K., N. K. R. Busireddy, K. K. Osuri, and D. Niyogi, 2020: On the relationship between intensity changes and rainfall distribution in tropical cyclones over the North Indian Ocean. Int. J. Climatol., 40, 2015–2025, https://doi.org/10.1002/joc.6315.
Bosma, C. D., D. B. Wright, P. Nguyen, J. P. Kossin, D. C. Herndon, and J. Marshall Shepherd, 2020: An intuitive metric to quantify and communicate tropical cyclone rainfall hazard. Bull. Amer. Meteor. Soc., 101, E206–E220, https://doi.org/10.1175/BAMS-D-19-0075.1.
Chen, L., and Y. Xu, 2017: Review of typhoon very heavy rainfall in China (in Chinese). Meteor. Environ. Sci., 40, 3–10, https://doi.org/10.16765/j.cnki.1673-7148.2017.01.001.
Chen, L., Y. Li, and Z. Cheng, 2010: An overview of research and forecasting on rainfall associated with landfalling tropical cyclones. Adv. Atmos. Sci., 27, 967–976, https://doi.org/10.1007/s00376-010-8171-y.
Di, L., and Coauthors, 2008: Impacts of cold air intrusion on extratropical transition of Typhoon Masta. Daqi Kexue Xuebao, 1, 18–25.
Ding, C., F. Ren, Y. Liu, J. L. Mcbride, and T. Feng, 2020: Improvement in the forecasting of heavy rainfall over South China in the DSAEF_LTP model by introducing the intensity of the tropical cyclone. Wea. Forecasting, 35, 1967–1980, https://doi.org/10.1175/WAF-D-19-0247.1.
Ebert, E. E., M. Turk, S. J. Kussion, J. B. Yang, M. Seybold, P. R. Keehn, and R. Kuligowski, 2011: Ensemble tropical rainfall potential (eTRaP) forecasts. Wea. Forecasting, 26, 213–224, https://doi.org/10.1175/2010WAF2222443.1.
Emanuel, K., 2017: Assessing the present and future probability of Hurricane Harvey’s rainfall. Proc. Natl. Acad. Sci. USA, 114, 12 681–12 684, https://doi.org/10.1073/pnas.1716222114.
Feng, S., Q. Hu, and W. Qian, 2004: Quality control of daily meteorological data in China, 1951–2000: A new dataset. Int. J. Climatol., 24, 853–870, https://doi.org/10.1002/joc.1047.
Fritsch, J. M., and Coauthors, 1998: Quantitative precipitation forecasting: Report of the Eighth Prospectus Development Team, U.S. Weather Research Program. Bull. Amer. Meteor. Soc., 79, 285–299, https://doi.org/10.1175/1520-0477(1998)079<0285:QPFROT>2.0.CO;2.
Han, H. T., and Z. L. Li, 2012: Research progress on quality control methods of ground real-time meteorological data (in Chinese). J. Arid Meteor., 30, 261–265.
Hong, J. S., C. T. Fong, L. F. Hsiao, Y. C. Yu, and C. Y. Tzeng, 2015: Ensemble typhoon quantitative precipitation forecasts model in Taiwan. Wea. Forecasting, 30, 217–237, https://doi.org/10.1175/WAF-D-14-00037.1.
Hsiao, L. F., M. J. Yang, and C. S. Lee, 2013: Ensemble forecasting of typhoon rainfall and floods over a mountainous watershed in Taiwan. J. Hydrol., 506, 55–68, https://doi.org/10.1016/j.jhydrol.2013.08.046.
Jia, L., Z. Jia, F. Ren, C. Ding, M. Wang, and T. Feng, 2020: Introducing TC intensity into the DSAEF_LTP model and simulating precipitation of super‐typhoon Lekima (2019). Quart. J. Roy. Meteor. Soc., 146, 3965–3979, https://doi.org/10.1002/qj.3882.
Jia, L., F. Ren, C. Ding, Z. Jia, M. Wang, Y. Chen, and T. Feng, 2022: Improvement of the ensemble methods in the dynamical–statistical–analog ensemble forecast model for landfalling typhoon precipitation. J. Meteor. Soc. Japan, 100, 575–592, https://doi.org/10.2151/jmsj.2022-029.
Jie, W., Y. Xu, L. Yang, Q. Wang, J. Yuan, and Y. Wang, 2020: Data assimilation of high-resolution satellite rainfall product improves rainfall simulation associated with landfalling tropical cyclones in the Yangtze River Delta. Remote Sens., 12, 276, https://doi.org/10.3390/rs12020276.
Kidder, S. Q., S. J. Kusselson, J. A. Knaff, R. R. Ferraro, R. J. Kuligowski, and M. Turk, 2005: The tropical rainfall potential (TRaP) technique. Part I: Description and examples. Wea. Forecasting, 20, 456–464, https://doi.org/10.1175/WAF860.1.
Lee, C.-S., L.-R. Huang, H.-S. Shen, and S.-T. Wang, 2006: A climatology model for forecasting typhoon rainfall in Taiwan. Nat. Hazards, 37, 87–105, https://doi.org/10.1007/s11069-005-4658-8.
Li, B., and S. X. Zhao, 2009: Development of forecasting model of typhoon type rainstorm by using SMAT (in Chinese). Meteorology, 35, 3–12.
Liu, C. C., 2009: The influence of terrain on the tropical rainfall potential technique in Taiwan. Wea. Forecasting, 24, 785–799, https://doi.org/10.1175/2008WAF2222135.1.
Lonfat, M., R. Rogers, T. Marchork, and F. D. Marks, 2007: A parametric model for predicting hurricane rainfall. Mon. Wea. Rev., 135, 3086–3097, https://doi.org/10.1175/MWR3433.1.
Maddox, R. A., C. F. Chappell, and L. R. Hoxit, 1979: Synoptic and meso-α scale aspects of flash flood events. Bull. Amer. Meteor. Soc., 60, 115–123, https://doi.org/10.1175/1520-0477-60.2.115.
Marks, F. D., G. Kappler, and M. DeMaria, 2002: Development of a tropical cyclone rainfall climatology and persistence (RCLIPER) model. Preprints, 25th Conf. on Hurricanes and Tropical Meteorology, San Diego, CA, Amer. Meteor. Soc., 327–328.
Ren, F. M., and C. Y. Xiang, 2017: Review and prospect of researches on the prediction of precipitation associated with landfalling tropical cyclones (in Chinese). J. Mar. Meteor., 37, 8–18.
Ren, F. M., B. Gleason, and D. R. Easterling, 2001: A technique for partitioning tropical cyclone precipitation (in Chinese). J. Trop. Meteor., 17, 308–313.
Ren, F. M., Y. M. Wang, X. L. Wang, and W. J. Li, 2007: Estimating tropical cyclone precipitation from station observations. Adv. Atmos. Sci., 24, 700–711, https://doi.org/10.1007/s00376-007-0700-y.
Ren, F. M., W. Y. Qiu, X. L. Jiang, L. G. Wu, Y. L. Xu, and Y. H. Duan, 2018: An objective track similarity index and its preliminary application to predicting precipitation of landfalling tropical cyclones. Wea. Forecasting, 33, 1725–1742, https://doi.org/10.1175/WAF-D-18-0007.1.
Ren, F. M., C. Ding, D. L. Zhang, D. L. Chen, H. L. Ren, and W. Y. Qiu, 2020: A dynamical-statistical-analog ensemble forecast model: Theory and an application to heavy rainfall forecasts of landfalling tropical cyclones. Mon. Wea. Rev., 148, 1503–1517, https://doi.org/10.1175/MWR-D-19-0174.1.
Ren, Z. H., Q. Zhang, F. Gao, and Y. Yu, 2018: CMA meteorological data quality control system (in Chinese). Adv. Meteor. Sci. Technol., 8, 54–55.
Singh, K. S., and B. Tyagi, 2019: Impact of data assimilation and air–sea flux parameterization schemes on the prediction of Cyclone Phailin over the Bay of Bengal using the WRF‐ARW model. Meteor. Appl., 26, 36–48, https://doi.org/10.1002/met.1734.
Titley, H. A., H. L. Cloke, S. Harrigan, F. Pappenberger, C. Prudhomme, J. C. Robbins, E. M. Stephens, and E. Zsoter, 2021: Key factors influencing the severity of fluvial flood hazard from tropical cyclones. J. Hydrometeor., 22, 1801–1817, https://doi.org/10.1175/JHM-D-20-0250.1.
van Oldenborgh, G. J., and Coauthors, 2017: Attribution of extreme rainfall from Hurricane Harvey, August 2017. Environ. Res. Lett., 12, 124009, https://doi.org/10.1088/1748-9326/aa9ef2.
Wang, J., Y. Xu, L. Yang, Q. Wang, J. Yuan, and Y. Wang, 2020: Data assimilation of high-resolution satellite rainfall product improves rainfall simulation associated with landfalling tropical cyclones in the Yangtze River Delta. Remote Sens., 12, 276, https://doi.org/10.3390/rs12020276.
Wang, Y. M., F. M. Ren, and X. L. Wang, 2006: The study on the objective technique for partitioning tropical cyclone precipitation in China. Meteor. Mon., 32, 6–10.
Wilks, D. S., 1995: Statistical Methods in the Atmospheric Sciences: An Introduction. International Geophysics Series, Vol. 59, Elsevier, 467 pp.
Xu, D. S., Z. T. Cheng, and S. X. Zhong, 2014: Study of the coupling of cumulus convection parameterization with cloud microphysics and its influence on forecast of typhoon (in Chinese). Acta Meteor. Sin., 72, 337–349.
Yamaguchi, M., J. Chan, I. J. Moon, K. Yoshida, and R. Mizuta, 2020: Global warming changes tropical cyclone translation speed. Nat. Commun., 11, 47, https://doi.org/10.1038/s41467-019-13902-y.
Ying, M., W. Zhang, H. Yu, X. Lu, J. Feng, Y. Fan, Y. Zhu, and D. Chen, 2014: An overview of the China Meteorological Administration tropical cyclone database. J. Atmos. Oceanic Technol., 31, 287–301, https://doi.org/10.1175/JTECH-D-12-00119.1.
Yu, X., S. K. Park, Y. H. Lee, and Y. S. Choi, 2013: Quantitative precipitation forecast of a tropical cyclone through optimal parameter estimation in a convective parameterization. SOLA, 9, 36–39, https://doi.org/10.2151/sola.2013-009.
Zhong, Y., H. Yu, W. P. Teng, and P. Y. Chen, 2009: A dynamic similitude scheme for tropical cyclone quantitative precipitation forecast (in Chinese). J. Appl. Meteor. Sci., 20, 17–27.
Zhu, L., Q. L. Wan, X. Y. Shen, Z. Meng, F. Zhang, and Y. Weng, 2016: Prediction and predictability of high-impact western Pacific landfalling Tropical Cyclone Vicente (2012) through convection-permitting ensemble assimilation of Doppler radar velocity. Mon. Wea. Rev., 144, 21–43, https://doi.org/10.1175/MWR-D-14-00403.1.
