Rainfall pattern over the middle of Indochina Peninsula during 2009–2010 summer monsoon

: Rainfall patterns during summer monsoon in 2009 and 2010 over the middle of the Indochina Peninsula (ICP) are investigated using calibrated daily accumulated radar rainfall (CDARR). Empirical orthogonal function (EOF) analysis applied to CDARR shows that the first three modes explain 40% of the total rainfall variance. The pattern of the first EOF mode is only positive over the radar observation area with a large value near the foot of the Annam range in the eastern region of the radar site. The second EOF mode is a dipole pattern that has positive and negative regions in the eastern and western regions of the radar observation area, respectively. The third EOF mode also shows a dipole pattern with positive and negative areas in the southern and northern regions of the observation area, respectively. Composite analysis results suggest that the first EOF mode is possibly produced by a difference in positive vorticity, in which the difference in the southerly wind component likely causes orographic rainfall in the eastern region of the radar site. In addition, the second and third EOF modes are possibly produced by differences in westerly and southwesterly wind components, respectively.


INTRODUCTION
This study aims to determine a dominant spatial pattern of summer monsoon rainfall over the middle of the Indochina Peninsula (ICP) with a focus on the topography surrounding the radar observation area. The rainy season begins over the middle of the ICP earlier than that in the surrounding region, where summer monsoon begins in mid-May over the western coast of the ICP and the central South China Sea (SCS) (Matsumoto, 1997). The understanding of rainfall systems is important because agriculture in this region is significantly affected by the natural variability of rainfall. Although the contribution of the agriculture sector to the gross domestic product of countries in this region has continually declined recently, agriculture is still the most important economic activity in the ICP (MRC, 2010). In addition, natural disasters occur regularly in this region, particularly flooding during the summer monsoon season in Cambodia, Laos, Thailand, and Vietnam during the past decade (Asian Disaster Reduction Center, 2013). The annual expense due to flood damage is about 60-70 million US dollars in these countries (MRC, 2010). A massive flood in Chao Phraya River Basin, Thailand, in 2011 was an example of severe flooding in the region that was caused by the maximum ever recorded rainfall in the country (Komori et al., 2012). Hoyos and Webster (2007) suggested that a tall and narrow mesoscale mountain affects orographic rainfall during the Indian summer monsoon. Xie et al. (2006) discussed the orographic anchoring of monsoon convection, revealed by the Tropical Rainfall Measuring Mission Precipitation Radar (TRMM PR), to occur on the western coasts of India, Indochina, and the Philippines, and in the foothills of the Himalayas and the Annam range. Thus, major local maxima of the rainfall of summer monsoon have been detected at the windward side of mesoscale mountains. In this context, it would be interesting to study how two mesoscale high mountain ranges near the eastern and western coasts of the ICP, between which a large plain is situated, affect rainfall.
Spatio-temporal variations in rainfall related to mountains in the ICP have been studied by Satomura (2000), who used a numerical model to determine that diurnal variations in precipitation are governed by organized squall lines moving eastward from mountains during the night over the middle ICP. Okumura et al. (2003) studied diurnal variations in radar echo by using Omkoi radar in northern Thailand. They detected a phase delay of the peak of the echo area in time with distance from the mountains at the lee side of the mountain range. This result is consistent with that determined by a numerical simulation conducted by Satomura (2000), in addition to a phase delay of the peak radar echo area detected near Vientiane, Laos . In addition, Yokoi and Satomura (2008) and Yokoi et al. (2007) detected mesoscale mountain effects on intraseasonal variation (ISV) in precipitation over the ICP by using Omkoi radar and rain gauge data, respectively. They concluded that the sharp gradient of 30-60-day variation over the mountain ranges in the ICP is produced by enhancement (suppression) effects at the windward (lee) sides of the mountains.
Although the use of radar echo for rainfall analysis has an advantage over rain gauges in terms of spatial resolution, the rainfall observed by radar contains inherent errors, as reviewed by Wilson and Brandes (1979). They calculated a multiplicative adjustment factor to account for the differences in the two measurement methods. Despite such calibration results in improvements, they suggested that a degree of spatial error remains, which can be removed by local adjustment between gauge and radar measurements through inverse distance weighting. However, they also discussed that the interpolation error occurring with distance from the gauge needs to be investigated.
Although the gridded precipitation data is useful for rainfall study, it has some disadvantages. Rainfall estimates products produced from satellite data show poor performance at daily time scale validated by rainfall observed by gauge (Dinku et al., 2010). In addition, the reliability of gauge-based daily precipitation datasets also depend on many factors such as distribution of gauge stations, accessibility of gauge observations, and quality control of gauge observations (Xie et al., 2007;Yatagai et al., 2009). However, many methods of rainfall bias correction have been applied to these gridded precipitation data to improve the accuracy of rainfall estimates in different parts of the ICP (e.g., Ngo-Duc et al., 2013;Ono et al., 2013).
In this study, radar echoes observed in Vientiane are calibrated by a dense rain gauge network in Thailand and are subjects of an empirical orthogonal function (EOF) analysis. While the research on radar echoes calibrated by a rain gauge network in the northwestern region of the ICP has been successfully performed by Yokoi et al. (2011), the present study is the first to accomplish the calibration of rainfall over the middle of the ICP by using a rain gauge network and EOF analysis of radar echo data. Even though the calibration method that we used is very simple, it may be useful and practical for hydrological and meteorological staffs in this region.

DATA AND METHOD
The C-band doppler radar used in this study is located in Vientiane, Laos, 102°34'14.2''E, 17°58'15.9''N at a height of 168 m above the mean sea level, and is operated by the Department of Meteorology and Hydrology (DMH), Laos ( Figure 1a). This radar observed four or eight volume scans per hour consisting of 12 angles with elevation angles of 0.5°-23° during the study periods of July 1 to September 30 in 2009 and 2010. Rain gauge data with a time resolution of 15 min was provided by the Thai Meteorological Department (TMD), which operates 161 gauges within 200 km from the radar site in the Thailand territory, as shown in Figure 1b. The GTOPO30 global digital elevation model with a horizontal resolution of 30'' was used for topographic data. ERA-Interim reanalysis data from the European Centre for Medium-Range Weather Forecasts (ECMWF) at 1.5° × 1.5° resolution (Dee et al., 2011) were used for meteorological data analysis.
In this study, radar echo data blocked by topography, which were greater than 50% of the beam's cross section, were excluded from our calculation of the constant altitude plan position indicator (CAPPI). Prior to interpolation to the Cartesian grid of CAPPI, the radar echo intensity was converted to the rainfall rate (hereafter radar rainfall) by using a common empirical radar-rainfall relationship, Z = 200R 1.6 (Marshall et al., 1955), where Z is the radar reflectivity factor in mm 6 m −3 and R is the rainfall rate in mm h −1 . The radar rainfall within a range of 0.15 to 100 mm h −1 was interpolated into CAPPI at 3-km height by using the weighted interpolation method (Cressman, 1959) with horizontal and vertical influence radii of 2.5 and 1.5 km, respectively. The horizontal resolution of CAPPI was 1 km with 400 × 400 grids. We considered that our radar rainfall grid data is not contaminated by ice phase because the use of CAPPI height and vertical influence radius are considerably below a seasonal mean of freezing level for Black dots indicate the locations of rain gauges in the observation range, the radar site is marked by a triangle, and observation radius of 200 km is indicated by a circle the study area. Daily accumulated radar rainfall (DARR) and daily accumulated rainfall observed by gauge (DARG) within a range of 10 to 200 km from the radar site were used to calculate linear regression for each month with the criteria that both DARR and DARG were greater than 0 mm day −1 . DARR was then calibrated by multiplying the slope of the regression line (conversion factor (CF), hereafter) and the calibrated DARR was abbreviated as CDARR, hereafter.
As shown in Figure 2a, DARR was smaller than DARG during the study period, and the CF determined each month varied from 1.6 to 1.9 (not shown). The root mean square difference (RMSD) between CDARR and DARG of each month was reduced by approximately half after the application of the CF (not shown). As shown in Figure 2b, the strong underestimation of radar rainfall was improved. In fact, RMSD was reduced from approximately 6 mm day −1 to 3 mm day −1 after application of the CF. An accumulation of rainfall averaged over all gauge positions of all months by using CDARR showed nearly the same result as that using DARG (Figure 2d), with a difference of 98 mm. Thus, calibration resulted in clear improvement (Figure 2d).
Spatial distribution of accumulated average rainfall from July to September of 2009 and 2010 was calculated using DARG and CDARR (Figure 3a and 3b). These figures show a well corresponding pattern of rainfall distribution over the radar observation area: while a larger amount of rainfall was located in the east of the radar observation area near the foot of the Annam range, a smaller amount of rainfall was located over plain in the southern part of the radar observation area.
EOF analysis (e.g., Murakami, 1980;Singh, 2004) was employed to analyze the dominant mode of daily precipitation patterns. This analysis was applied to the cube roots of CDARR because the cube root of rainfall rather than the original rainfall is closer to a normal distribution pattern (Stidd, 1953). The scores of each EOF mode were normalized by its standard deviation.

RESULTS AND DISCUSSIONS
CDARR during a 184-day period from July 1 to September 30 of 2009 and 2010 was analyzed by EOF. The Radar rainfall, the radar site is marked by a triangle and the color bar is used to explain rain rate of both gauge and radar rainfall. The rain rate unit is mm. The 500 m and 1,000 m contours are indicated by gray and black lines, respectively Figure 2. Results of rainfall observed by gauge and radar before and after application of the conversion factor (CF) for all months during the study period. Scatter diagram of (a) rainfall observed by gauge and radar before CF application and (b) that observed after CF application. Accumulated average rainfall at positions of gauges observed by gauge and radar (c) before and (d) after CF application, represented by blue and red lines, respectively first three dominant modes of EOF were selected because they explained 40% of the total rainfall variability during the study period. Higher EOF modes were not selected because the fourth EOF mode explained only 3.5% of the total rainfall variability. The first EOF mode, the only positive mode in the study area (Figure 4a), explained 27% of the total rainfall variability. The largest positive values were located near the foot of the Annam range in the eastern region of the radar observation area, which is the windward side of the Annam range for the southwesterly summer monsoon wind ( Figure 5). Relatively small positive values were located over the westernmost radar observation area, which includes the eastern region of the Luangprabang range and the lee side of the mountain range for the southwesterly summer monsoon wind. The first EOF mode showed is similar to the spatial distribution of accumulated average radar rainfall as shown in Figure 3b, which are large positive values located over the east of the radar observation area and relatively small positive values over the plain.
The second and third EOF modes showed dipole patterns of rainfall, which explain 7% and 6% of the total rainfall variability, respectively. The pattern of the second EOF mode showed an east-west contrast (Figure 4b). Many small negative areas were detected near the Luangprabang range in the western region of the radar observation area, which also includes the lee side of the mountain range for the southwesterly summer monsoon wind. The maximum positive values were located over the easternmost part of the radar observation area, which includes the foot of the Annam range and the windward side of the mountain for the southwesterly summer monsoon wind.
In contrast to the pattern of the second EOF mode, that of the third EOF mode (Figure 4c) revealed a north-south contrast with large positive values located over the plain in the southern part of the radar observation area. In addition, large negative values were detected near the foot of the Annam range in the northern region of the radar observation area. Figures 6a, 6b, and 6c show the scores of the first, second, and third EOF modes, respectively. The large positive (negative) days with scores above (below) the percentile at 90% (10%) are hereafter referred to as LPD (LND). In order to find causes making LPD and LND, the effect of tropical cyclone on each EOF mode is examined. Firstly, the best tracks of tropical cyclone provided by RSMC Tokyo -Typhoon Center (2013) were used to determine the center of the tropical cyclone. When the centers of tropical cyclones were located within a radius of 500 km from the radar station, those LPD or LND of each EOF score are determined as affected days. Then, those affected days are excluded before recalculation of EOF. The recalculated EOF results show a small change of the first three dominant EOF patterns with a small change of EOF variance from those of EOF analysis including affected days (not shown). This result suggests that the rainfall patterns of each EOF mode are not affected by tropical cyclones. Even if the distance between radar station and the center of tropical cyclone is changed to 700 km, the EOF mode and the variance of each EOF mode were not much changed.
Next, LPD and LND were used to create a composite  (Figure 7). The statistical significance for differences in vorticity and wind between LPD and LND composites was assessed with the use of the student's t-test. The black arrows shown in Figure 7c, 7f, and 7i indicate differences in wind between LPD and LND of each EOF mode with statistical significance at the 95% confidence level. The areas where vorticity was statistically significant at the 95% confidence level were located inside of the black dash lines shown in Figure 7c, 7f, and 7i. Most parts of the radar observation area in the first EOF mode have the statistical significance of the differences in vorticity composites. In addition, the east of the radar observation area in the second EOF mode shows statistical significance of the differences in vorticity. However, the difference in vorticity of the third EOF mode over the radar observation area is not statistically significant at the 95% confidence level. The composite pattern in LPD of the first EOF mode (Figure 7a) indicated that positive vorticity was located over the radar observation area as part of a larger positive vorticity over the northern part of Vietnam with southwesterly wind. The pattern in LND of the first EOF mode (Figure 7b) showed negative vorticity over the radar observation area. The difference in wind and vorticity between LPD and LND (Figure 7c) possibly indicated a stronger southerly wind blowing into the southern part of the radar observation area with larger positive vorticity.
The large-scale rainfall pattern of the first EOF mode is consistent with the difference in vorticity between LPD and LND, which showed positive vorticity over the radar observation area (Figure 7c). Although the southerly wind is not statistically significant over the foot of the Annam range, the larger positive values near the foot of the Annam range, which faces southwest, may be possibly attributed to orographic rainfall enhancement because the stronger southerly wind component in LPD is considered to induce more orographic rain in that area. The smaller positive values of the first EOF mode located in the westernmost region of the radar observation area (Figure 4a) were likely caused by the lee side rainfall suppression by the Luangprabang range and the westerly prevailing wind.
The composite vorticity in LPD in the second EOF mode (Figure 7d) was also positive but weaker than that in LPD of the first EOF mode (Figure 7a). Westerly wind was dominant in LPD in the second EOF mode, whereas southwesterly wind was dominant in this composite map (Figure 7e).
The dipole pattern of the second mode (Figure 4b) was possibly produced by the wind difference between LPD and LND (Figure 7f), which showed stronger westerly wind in LPD. Though the southerly wind is not statistically significant over the foot of the Annam range (Figure 7f), orographic rain possibly occurred near the foot of the Annam range when stronger westerly wind blew toward the range. The scattered smaller negative values shown in the second EOF mode in the western region of the radar observation ( Figure 4b) area possibly suggest rainfall suppression after the prevailing wind was blocked by the Luangprabang range.
The difference in vorticity between LPD and LND of the third EOF mode over the radar observation area ( Figure  7i) may not be worth discussion because it was not statistically significant at the 95% confidence level. The LPD pattern in the third EOF mode indicates that southwesterly wind was dominant (Figure 7g). Figure 7h shows the stronger southwesterly wind as well as the large positive vorticity located over the radar observation area in LND. Figure 7i shows that the strong northeasterly wind was dominant over the radar observation area, indicated by the difference between LPD and LND. The rainfall over the Annam range in the northern and eastern regions of the radar observation area was possibly reduced by the weaker southwesterly wind (Figures 7g and 7i), which suggests that this EOF pattern probably represents orographic rain.

CONCLUSIONS
The radar echo data observed near Vientiane, Laos, from July 1 to September 30 of 2009 and 2010 were converted to radar rainfall by using the common empirical radarrainfall relationship. The radar rainfall was then calibrated by rain gauges with conversion factors defined each month. CDARR was used to examine the rainfall pattern over the middle of the ICP during the summer monsoon of 2009-2010. EOF analysis was applied to the cube root of CDARR to extract the dominant modes for the 184-day study period. Three dominant EOF modes were selected because they accounted for 40% of the total rainfall variability. The analysis results are summarized as follows: 1. The first EOF mode showed rainfall over a large area of the radar observation region. The second and third EOF modes indicated dipole patterns. 2. The first EOF mode was possibly produced by differences in positive vorticity. Differences in the southerly wind component likely caused orographic rainfall in the eastern region of the radar observation area near the foot of the Annam range. 3. The second and third EOF modes were possibly produced by the differences in westerly and southwesterly wind components, respectively. The strong westerly and southwesterly wind likely caused orographic rainfall in the eastern and northern region of the radar observation area for the second and third EOF modes, respectively. 4. The rainfall patterns of each dominant EOF mode were not affected by tropical cyclones.

ACKNOWLEDGMENTS
The radar data were provided by DMH, Laos. Rain gauge data were provided by TMD, Thailand. The ERA-Interim reanalysis data used in this study were provided by ECMWF.  LND of the (b) first, (e) second, and (h) third EOF modes; and differences between LPD and LND of the (c) first, (f) second, and (i) third EOF modes. The vorticity unit is 1 × 10 −6 s −1 , and the wind vector unit is m s −1 . Open circles and open triangles represent the 200-km radar observation range and the radar site, respectively. The reference vector at the bottom right of each figure represents 10 m s −1 . The black arrow of the (c) first, (f) second, and (i) third EOF modes are difference in wind between LPD and LND with statistical significance at the 95% confidence level. The black dash lines of the (c) first, (f) second, and (i) third EOF modes indicates boundary of difference in vorticity between LPD and LND with statistical significance at the 95% confidence level