Evaluation of CMPA precipitation estimate in the evolution of typhoon-related storm rainfall in Guangdong, China

Extreme precipitation often causes or induces flooding, landslides, and mudslides (Hong et al. 2006; Wu et al. 2012). Heavy rainfall driven by tropical cyclones (also called typhoons) frequently cause casualties and enormous property losses in coastal areas (Negri et al. 2005). For example, Katrina struck the Gulf Coast of Florida, Alabama, Mississippi and Louisiana on August 2005 (Knabb et al. 2005), Morakot hit Taiwan and Fujian Provinces of China on August 2009 (Van Nguyen & Chen 2011), and Yasi hit the Queensland coast of Australia on February 2011 (Chen et al. 2013d). All brought huge amounts of rainfall and caused heavy casualties and serious economic losses.

Accurate measurement of precipitation with fine resolution and reasonable accuracy is essential but challenging (Li et al. 2013). Automatic rain gauge networks can provide near-real time measurements, while the representativeness of a sparse and uneven distribution of gauges is worthy of consideration (Li & Shao 2010). Weather radar is capable of monitoring local precipitation under fine spatial resolution and reliable quality, although it is affected by atmospheric conditions and high terrain in mountainous areas (Zhang et al. 2011a). With the development of satellite precipitation retrievals, high resolution and multi-sensor precipitation products come out in succession and are available in near-real time, such as PERSIANN (hourly and 0.04°; Sorooshian et al. 2000), the US NOAA CPC Morphing Technique (CMORPH) (30 minutes and 8 km; Joyce et al. 2004), Naval Research Laboratory (NRL)-Blend (3 hours and 0.25°; Turk & Miller 2005), the Tropical Rainfall Measuring Mission (TRMM) Multi-satellite Precipitation Analysis (TMPA, 3 hours and 0.25°; Huffman et al. 2007), and the Global Precipitation Measurement products (4 km and 30 minutes; Sorooshian et al. 2011). For continental China, version 1.0 of the merged precipitation data of the Climate Prediction Center Morphing Technique and gauge based precipitation (CMPA, 0.1° and 1 hour) are available for recent years and demonstrate high agreements with ground observations (Shen et al. 2014). These products greatly improve hydrological simulation and flood prediction due to their large coverage and relatively high spatial resolution (Anagnostou 2004; Hossain & Lettenmaier 2006; Behrangi et al. 2011; Jiang et al. 2012).

Satellite precipitation estimates have considerable importance in many hydrological and meteorological applications (Tian et al. 2007). The satellite-based precipitation products usually need verification before being applied in hydrologic modeling (Turk et al. 2008). The comprehensive evaluation to satellite precipitation products has significant implications. On the one hand, the knowledge of error characteristics is beneficial to the improvement of retrieval algorithm (Dinku et al. 2011). On the other hand, the error characteristics offer suggestions on how the data are properly used in hydrologic and climatic models since errors may vary notably among different seasons and climate settings (Dai et al. 2007; Zhou et al. 2008; Stampoulis & Anagnostou 2012). Satellite precipitation estimates, after bias adjustment, can provide better performance in hydrological modeling (Su et al. 2008; Yong et al. 2010).

The current satellite rainfall estimates have limitations in detecting extreme rainfall but with great potential. For instance, the four products of TMPA 3B42 V6, 3B42 RT, PERSIANN-CCS and CMORPH greatly underestimates the peak rain rate for the extreme rainfall brought by the Typhoon Morakot in Taiwan and Fujian Provinces, China (Chen et al. 2013a). Both TMPA 3B42 V6 and CMORPH considerably underestimates the moderate and heavy rainfall and overestimates the light rainfall of typhoon-related precipitation over mainland China (Yu et al. 2009). The TMPA 3B42 V7 performs better over ocean than over land, and underestimates the frequency of heavy rain over land according to the assessment study using PACRIAN rain gauges over low-lying atolls and coastal islands in the tropical Pacific basins (Chen et al. 2013e). Meanwhile, Habib et al. (2009) highlight the potential of TMPA products and demonstrate the significant improvement from the real-time version to the research version in a study using six tropical-related heavy rainfall events in Louisiana, USA. Prat & Nelson (2013) quantify the contribution of tropical-related rainfall over land to the annual rainfall around the world using TMPA 3B42 V7 estimate, and this contribution could be better quantified if the underestimate of peak rain rate of TMPA is adjusted. In addition, the emergence of CMPA estimate has not received enough attention, although it shows high agreements with ground observations (Shen et al. 2014) and is used in a few studies (Yu & Liu 2015; Chen et al. 2015, 2016). Moreover, few studies use the CMPA to quantitatively analyze the typhoon-related heavy rainfall in China.

Therefore, the primary objective of this study is to evaluate the performance of the hourly CMPA product on typhoon-related storms using an independent, dense rain gauge network in Guangdong Province, China. Seven typhoon-related storm events were selected to compare the CMPA estimates with dense gauge recordings over 1,330 stations. This study intends to investigate the error characteristics of the CMPA estimate on typhoon storms through comprehensive assessment, such as quantifying total rain depth, rainfall hyetograph, spatiotemporal pattern and storm evolutions, and so on.

There were a total of seven typhoon events which made landfall in Guangdong Province from 2013 to 2015 (Table 1). Three typhoon rainfall events of Utor, Usagi and Linfa (Figure 1) had a wider impact on the entire Guangdong Province than the other four typhoons of Mujigae, Kalmaegi, Rammasun and Rumbia (Figure A1, available with the online version of this paper), and are presented in the main text. The results for the other four typhoons are provided in the auxiliary material and cited in the text with capital letter of A, e.g. Figure A1.

Typhoon Utor was a long-lived storm, which originated as a tropical disturbance in the west Pacific on August 8 and became a super typhoon on 10 August. After passing through the Philippines, Utor strengthened again and landed on Yangjiang at 7:50 UTC on August 14, 2013, with a peak wind speed of 42 m/s, then moved northward and eventually dissipated on August 18, 2013 (Figure 1). It is one of the most extreme typhoon rainfall events in Guangdong Province in the last 50 years, causing 1.33 billion Yuan property loss and 84 people dead or missing (AOF 2014). The average rainfall in Guangdong is more than 170 mm (Table 1).

Typhoon Usagi is one of the strongest typhoons that has landed on east Guangdong in the last 40 years. It landed on Shanwei at 11:40 UTC on September 22, 2013 with a peak wind speed of 45 m/s, and moved northwest across Guangdong before dissipating in the morning of September 23, 2013 (Figure 1). Typhoon Usagi brought strong winds and high tide surges (1.63 m at Zhelang station), 100–200 mm of rain to most areas of Guangdong Province (Table 1), caused 5.86 billion Yuan property loss and affected five million people with 30 people dead (AOF 2014).

Typhoon Linfa became a typhoon at 12:00 UTC on July 8, 2015 in the northeast of the South China Sea. It landed on Shanwei at 04:15 UTC on July 9, 2015 with a peak wind speed of 35 m/s (Figure 1). It then moved westward and brought an average of 48 mm of rain to east Guangdong Province (Table 1).

Typhoon Mujigae originated as a tropical disturbance in the South China Sea and moved toward the northwest. It made landfall near Zhanjiang City, Guangdong Province at 5:00 UTC on October 4, 2015 with a peak wind speed of 50 m/s, and then dissipated 22 h later (Figure A1(a)). It brought 100–200 mm rainfall to the western and central areas of Guangdong Province (Table 1).

Typhoon Kalmaegi developed on September 12, 2014 as a tropical depression in the west Pacific. It later made landfall on northeast Philippines at 11:00 UTC on September 14. After crossing the South China Sea, Kalmaegi strengthened again and landed on Hainan Island and Zhanjiang City successively in the morning of September 16, with a peak wind speed of 40 m/s (Figure A1(a)). It brought an average of 103 mm of rain to southwestern Guangdong Province (Table 1).

Typhoon Rammasun was a long-lived storm, which originated in the west Pacific on July 12, 2014, and then moved westward. After crossing the Philippines, Rammasun took a northwest turn and passed the South China Sea. It strengthened and landed on Zhanjiang at 11:00 UTC on July 18, 2014, with a peak wind speed of 60 m/s. It dissipated on July 20, 8 days after its original development (Figure A1(c)). It brought an average of 59 mm of rain to the western areas of Guangdong Province (Table 1).

Typhoon Rumbia became a typhoon at 02:00 UTC on July 1, 2013 in the central South China Sea. It landed on Zhanjiang at 21:30 UTC on July 1 with a peak wind speed of 30 m/s (Figure A1(d)). Then it moved northward and brought nearly 100 mm of rain to the Leizhou Peninsula of Guangdong Province (Table 1).

The dense rain gauge network is maintained by the Department of Water Resources in Guangdong Province (www.gdwater.gov.cn/). All rain depth data are accumulated hourly with a precision of 0.1 mm, and the temporal coverage starts from October 2012 to present. There are over 1,330 stations with valid rain depth recorded for the seven typhoon-related storm events from 2013 to 2015. To our best knowledge, these gauge observations were not incorporated in the interpolation process of the CMPA estimate. It is rational and significant to validate the CMPA estimates using rain depth data from the stations in this study.

The CMPA data are merged hourly precipitation products with 0.1 × 0.1° spatial resolution (Shen et al. 2014; http://cdc.nmic.cn/home.do) and are available from 2008 to present. The merging process is as follows: (1) the hourly precipitation measurements from more than 30,000 automatic weather stations over China are interpolated into 0.1 × 0.1° grids; (2) the 8 km CMORPH precipitation estimates (Joyce et al. 2004) every 30 minutes are summed into hourly accumulation and re-sampled into 0.1 × 0.1° grids; (3) the re-sampled CMORPH data are adjusted to correct biases by using the probability density function based on gauge measurements; (4) the final hourly 0.1 × 0.1° products are generated by merging the adjusted CMORPH data with the gauge data using the improved optimum interpolation technique. The corresponding root mean squared difference (RMSD) and correlation coefficient (CC) are 0.6 mm/hour and 0.8, respectively (Shen et al. 2014).

The typhoon-affected rainfall is first defined according to the trajectories of typhoon centers before carrying out the normal comparison and other evaluations. The time and trajectories of the typhoon events were retrieved from the Department of Water Resources in Guangdong Province. Rainfall observed within a radius of 400 km from the typhoon center is treated as typhoon-related rainfall in this area, while a 500-km radius is applied in the northwest of Australia (Lonfat et al. 2004; Dare et al. 2012). If the distance of a rain gauge from the typhoon center during any hour is less than 400 km, that particular hour of that gauge is defined as a ‘typhoon hour’. The typhoon-related storms normally last more than 24 hours (Table 1). Only those gauges with at least 12 typhoon hours are used in the following evaluation. The 12 hours represent a range of roughly 300 km since the mean moving speed of typhoon after landfall is 20–25 km/h. In practice, gauges with less than 12 typhoon hours usually have much less rainfall and are less affected by the typhoon event than those with over 12 typhoon hours.

The common grid-to-point comparison method is utilized in comparing the grid data of CMPA (0.1 × 0.1°) to the point gauge observations (Li et al. 2013). To reduce the scale errors, only those grids that contain at least one gauge are extracted for further evaluation. The mean value of all gauge measurements within a single grid is adopted if there is more than one gauge in a CMPA grid (Yong et al. 2010, 2014). Rain or no-rain hour is defined by a threshold of 0.1 mm/h. The total number of the affected gauges and grids are listed in Table 1.

The positive bias and negative bias will be offset in RB, which is used to describe the accumulative bias of the CMPA estimate. The standard mean bias is measured by RMSD. RB, RMSD and CC are used together to compare the CMPA estimate with rain gauge observations in this study and are also widely used in the radar and satellite rainfall validations (Yong et al. 2010; Jiang et al. 2012; Wang et al. 2013b; Shen et al. 2014).

POD shows how often the observed rain events are correctly detected by CMPA. FAR indicates the ratio of false alarm of CMPA. The perfect values of POD and FAR are 1 and 0, respectively.

Probability distribution functions (PDFs) provide detailed information on the distribution of rainfall frequency and accumulated volume as a function of rain rate (Chen et al. 2013c). The probability distribution function by occurrence (PDFc) is the ratio of the rain hours within a rain rate interval to the total rain hours. The probability distribution function by volume (PDFv) is the ratio of the rain depth within each rain rate interval to the event-total rain depth. PDFc and PDFv are computed using the ‘Hit’ pairs of gauges and CMPA grids in Table 3.

The K-means cluster algorithm has been applied to characterize temporal patterns of rainfall in mainland China (Yin et al. 2014), Taiwan (Yu et al. 2014), and South Korea (Lee et al. 2012). It sets initial centers and calculates the minimum squared distance from the samples to its centers iteratively. Finally, all gauges are classified into three clusters (hyetograph types) by the K-means cluster analysis (Hartigan & Wong 1979) using the time series of hourly gauge observations for each rain event in this study. The three clusters are considered as the three parts of typhoon-related storms, the eyewall, the rainband and the outer rainband (Lonfat et al. 2004). The three hyetograph types for each event form a spatial pattern that corresponds to the three stages of the storm evolution, the initial landfall (storm center), advance and the dissipation. The according CMPA grids’ values are extracted to compare with the gauge observations within each type or area for the seven storms.

Typhoon Utor landed on Yangjiang and brought intense rain to the west of the typhoon centers, forming a storm center in Maoming and with a peak rain depth over 400 mm (Figure 2(a), 2(d) and 2(g)). Then, the rain belt moved along the east of the typhoon track. Another rain belt was in the far east of the track. CMPA captures the storm centers and rain belts as gauges do, but has better spatial coverage and details.

Typhoon Usagi landed on Shanwei and brought intense rain along the coast areas from Huizhou, Shanwei and Shantou to the southern Fujian Province (Figure 2(b), 2(e) and 2(h)). Different from the Utor, the storm center is about 250 km to the northeast of the typhoon track and in the Hanjiang Watershed, which is administered together by Guangdong and Fujian Provinces. Beneficial to the larger spatial coverage, CMPA better captures the storm center than the gauges from Guangdong Province. Usagi continued to move northwest across Guangdong and formed a light rain belt in the north of the track.

Typhoon Linfa also landed on Shanwei and moved westward (Figure 2(c), 2(f) and 2(i)). It brought intense rain of 100–200 mm in the north along the track, forming two storm centers in Jieyang and Huizhou, respectively. CMPA works best for this event by CC = 0.91, RB = 0.63% and RMSE = 18.76 mm.

In the storm center (Type 1) for the Utor event, the peak rain rate occurs in the 9th–11th hours, when the Utor landed. CMPA greatly underestimates the peak rain and leads to a smaller event total rain by −15.54%, although it has a similar estimate prior to and post the peak time (Figure 7(b1)). Within the three peak hours from the 9th–11th hours, the mean rain rates from gauges were 22.7, 24.4 and 23.0 mm/h, while they were 10.5, 13.7 and 14.0 mm/h from CMPA. During the advance (Type 2) and dissipation stages (Type 3), CMPA estimates generally match the gauge observations by RB of 9.61 and 2.11% and CC of 0.95 and 0.99, respectively (Figure 7(c1) and 7(d1)). For the Usagi and Linfa events, CMPA estimates are overall in agreement with the gauge observations (Figure 7(b2)–7(d2) and 7(b3)–7(d3)). They are slightly larger than the gauge observations to Type 2 and Type 3 for the Usagi event and Type 3 for the Linfa event. There were similar patterns for the three types of hyetographs of the auxiliary four events (Table 4 and Figure A6). In other words, CMPA underestimate the intense rain and overestimate the light rain (Table 4).

The current satellite rainfall estimates have great potential in hydrologic and climatic applications in spite of their limitations in detecting extreme rainfall (Habib et al. 2009; Chen et al. 2013a). They usually need validation before applying in hydrologic modeling (Turk et al. 2008). In validation, more studies focus on the differences in rainfall amount, intensity and spatial varies between satellite estimates and ground observations, and fewer studies investigate the evolution of rainfall events (Wang et al. 2013b). The lately emerging CMPA product generated for continental China has higher spatial and temporal resolution (hourly and 0.1°) than the current common satellite precipitation estimates (3-hourly or daily and 0.25°) and also show better performance in quantifying extreme rainfall, but few studies have applied it to investigate the typhoon-related extreme rainfall in southern China.

This study carried out a comprehensive evaluation to the CMPA product in quantifying the typhoon-related extreme rainfall using a dense rain gauge network in Guangdong Province, China. All seven typhoon events making landfall in Guangdong Province from 2013 to 2015 were analyzed. The three typhoon storm events of Utor, Usagi and Linfa had a wider impact on the entire Guangdong Province than the other four typhoons of Mujigae, Kalmaegi, Rammasun and Rumbia, and are illustrated in the main text. The results for the other four typhoons are provided in the auxiliary material. CMPA shows consistent performance for all seven events. The primary results are summarized below.

The hourly and event-total precipitation from CMPA is generally in agreement with gauges. CMPA captures the spatial patterns of typhoon-related rain depth. CMPA underestimates intense rain and overestimates light rain. CMPA underestimates the occurrence of light rain and heavy rain, while overestimates the occurrence of moderate rain between 2 and 16 mm/h. CMPA shows similar high POD in all seven storm events. Meanwhile, it has a relatively large FAR and small missing ratio. The miss count has a larger proportion in the low rain rate ranges, leading to a smaller mean rain rate for the miss count than that of the false count.

All gauges are further classified into three types of hyetograph, which form a spatial pattern that corresponds to the three stages of the storm evolution, the initial landfall (storm center), advance and the dissipation. Even though the performance of CMPA may vary from case to case, CMPA generally underestimates rainfall of Type 1 and overestimates rainfall of Type 3 for most events. Overall, CMPA is in agreement with the spatiotemporal evolution of gauge observation.

Why does CMPA perform slightly better along the landfall track than that farther away from the track? This might relate to the physical mechanisms and topographic effects on the typhoon-related rainfall. More studies are needed to investigate various types of typhoon-related rainfall events that land on different regions and move towards different directions.

CMPA greatly underestimates the intense rainfall larger than 16 mm/h although it provides relatively accurate information on the spatiotemporal evolution of the typhoon-related extreme rainfall. This implies that caution should be taken when applying it in hydrologic modeling for flooding forecasting. Chen et al. (2013a) indicated that a ground radar network performs better than satellite-based products for extreme typhoon rainfall. Further studies with a combination of the CMPA estimate and ground radars are highly recommended. Moreover, hydrologic modeling and stream flow measurements are additional validation and constraints to total rainfall volume that might not be properly represented by gauges, especially in mountainous areas.

This study was partially supported by the Natural Science Foundation of China (#41371404; #51379224; #41301419), the Water Resource Science and Technology Innovation Program of Guangdong Province (#2016-19), and the Open Foundation of the State Key Laboratory of Desert and Oasis Ecology (#Y371163), Xinjiang Institute of Ecology and Geography, Chinese Academy of Sciences. We thank all researchers and staff for providing and maintaining the meteorological data and the CMPA product from China Meteorological Administration, National Meteorological Information Center, and Water Resources Department of the Guangdong Province. Comments provided by two anonymous reviewers and the editor greatly improved this manuscript and are highly appreciated by all authors.