Numerical Study on the Effect of Warm Ocean Anomalies in the South China Sea on the Heavy Rainfall Event on Hainan Island

The previous work has found that the development of an extended TD, which resulted in an abnormally prolonged and heavy rainfall event on Hainan Island from 1 October to 9 October in 2010, was significantly influenced by warm oceanic anomalies from surface to subsurface layer in the SCS. Numerical experiments using the Weather Research and Forecast Model (WRF) were conducted to understand the mechanisms of TD development during this event. A control experiment with an observed warm SST pattern simulated a reasonably extended TD, while a sensitivity experiment using the climatological SST distribution reproduced only a weak and short-lived TD, thus highlighting the importance of air-sea interactions and persistent warm upper-ocean anomalies for the genesis and intensification of the TD.


Introduction
The South China Sea (SCS) is home to the Hainan Island, which has altitudes of more than 500 meters in its southern region.The autumn is the island's primary rainy season, with maximum rainfall occurring during September and October nearly twice as much as it in May and June.This may be due to the summer monsoon's southward retreat in the autumn as well as the approach and passage of tropical cyclones (TCs) around the island [1,2].During the autumn rainy season, heavy rainfall events occurred in Hainan Island and the surrounding areas (e.g., Vietnam) were caused by abnormal southerlies associated with tropical depressions (TDs), along with northerly anomalies associated with winter monsoon cold surges [3].And topographic effects amplified the rainfalls.
From 1 October to 9 October in 2010, Hainan Island saw an exceptionally prolonged and intense rainfall event.On southeastern Hainan Island, the total amount of precipitation during these 9 days exceeded 600 mm, with extremes of over 1000 mm in coastal locations, shattering previous records for daily precipitation on eastern Hainan Island simultaneously.It is discovered that the main cause of this protracted heavy rainfall event (abbreviated as OCT10) was an extended TD surviving for 138 hours (12 UTC on 4 October to 06 UTC on 10 October), significantly longer than the TDs' 78-hour climatological lifetime [4].The warm oceanic anomalies from surface to subsurface layer in the SCS were illustrated as significant elements in the development of the prolonged TD.The warm sea surface temperature (SST) anomalies over the SCS aroused the formation of anomalous surface southerlies associated with the TD.And the warm oceanic anomalies in the subsurface manifested as extraordinarily thicker warm subsurface layer under the positive SST anomalies, which accumulated adequate upper-ocean heat content to refrain the SST cooling effect triggered by the TD.Thus the available enthalpy fluxes were ample to sustain the extended TD.In the composite analyses, the precipitation in the cases without the warm oceanic anomalies in the central SCS off the south-central coast of Vietnam was dramatically less than that in the "Warm eddy" cases.
The objective of this study is verifying the importance of warm ocean anomalies in the SCS on the formation and reinforcement of the TD and the resultant fortified effect of such extended TD on the heavy rainfall event on Hainan Island by means of numerical simulation.Section 2 describes the data and model used in this study.The simulation results and related comparisons are discussed in Section 3. A summary and discussions are given in Section 4.

Data
To demonstrate the atmospheric conditions of the heavy rainfall event, 1°×1°-resolution 6-hourly atmospheric circulation data were taken from National Centers for Environmental Prediction (NCEP) Final (FNL) Operational Global Analysis products.The NCEP FNL data were also used for the initial and boundary conditions in the numerical experiments (U.S. National Centers for Environmental Prediction, updated daily: NCEP FNL Operational Model Global Tropospheric Analyses, continuing from July 1999.Dataset ds083.2published by the CISL Data Support Section at the National Center for Atmospheric Research, Boulder, CO, available online at http://dss.ucar.edu/datasets/ds083.2/).The observed daily rainfall data with a resolution of 0.1°×0.1°were obtained from the Integrated Multi-satellite Retrievals (IMERG) products for Global Precipitation Measurement (GPM) [5].
For ocean features, daily SST data at a spatial resolution of 1/4° were derived from NOAA optimum interpolation sea surface temperature (OISST) datasets [6].Such SST data were also utilized for surface boundary forcing in the numerical modeling.
At the air-sea interface, high-resolution (0.25°×0.25°) daily surface latent heat (LH) and sensible heat (SH) fluxes were acquired from the dataset of Woods Hole Oceanographic Institute's Objectively Analyzed air-sea heat Fluxes (WHOI_OAFlux) [7].The WHOI_OAFlux product was developed by fusing surface moorings, ship reports, atmospheric model reanalyzed surface meteorology, and satellite observations.

Model descriptions and experiments
We performed numerical experiments using the Weather Research and Forecast Model (hereafter WRF) Advanced Research version.The WRF model, which is a nonhydrostatic fully compressible mesoscale atmospheric dynamic model [8], was configured using multiple nested domains connected by two-way feedbacks, and with numerous choices of physics schemes.In this study, three nested domains were employed, based on Mercator map projections with horizontal resolutions of 27, 9, and 3 km.The finest (highest-resolution) domain used in the WRF simulations is shown in Figure 1a.Except that the Kain-Fritsch (KF) cumulus parameterization scheme [9] was used only in the two outermost domains (with relatively coarser resolutions), all the other same physical processes were adopted for each domain, including the Morrison 2-moment microphysics scheme [10,11], the Rapid Radiative Transfer Model scheme for longwave radiation [12], the Dudhia scheme for shortwave radiation [13], the Yonsei University (YSU) scheme for planetary boundary layer (PBL) physics [14], and the Noah land-surface model [15].Two ensemble experiments were conducted using different SST patterns as varying surface boundary conditions.The control experiment was run to realistically simulate the physical processes associated with TD activity and heavy rainfall during the OCT10 event, using the observed daily SST distribution (Figures 1g-1l) as an underlying boundary forcing condition.In contrast, the sensitivity experiment (without a warm SSTA) was run during the same period, with climatological daily SST patterns as the surface boundary forcing condition (Figures 1m-1r); this experiment was designed to highlight the impact of warm ocean anomalies in the SCS on TD development during the OCT10 event, by comparing the results of the sensitivity experiment with those of the control experiment.The same initial atmospheric conditions and identical model lateral boundary conditions were derived from NCEP FNL data for both the control and the sensitivity experiments.In all experiments, the model was integrated from 0000 UTC 27 September to 0000 UTC 10 October (to cover the entire period of the OCT10 event), and without using a bogus vortex scheme and special observed assimilations.The simulation results prior to 0000 UTC 29 September were excluded, as they constituted the model's spin up or initial adjustment period; outputs over the subsequent 11 days were analysed in this study, with a focus on the behaviour of simulated TD under different experimental conditions.

Simulation results
Figure 1 shows variations in the observed and simulated daily surface winds and rainfall for both the control and sensitivity experiments during the OCT10 event.Also displayed are the SST patterns, which act as surface boundary forcing conditions during the course of each experiment.As mentioned above, observed daily SSTs are used in the control experiment to show how the warm SSTAs in the SCS influenced the genesis and growth of the extended TD.Prior to the genesis of the TD and during its developing stage (1-5 October, as shown in Figure 1a-1d and 1g-j), both the SCS and the western North Pacific east of 120°E were covered almost entirely by very high SSTs (greater than 29.5°C), with cold waters intruding into the northern SCS from the southeastern China Coast.On subsequent days (7-9 October), the SSTs decreased somewhat, and relatively cold SSTs (28°C -29°C) were present in several areas of the central and southern SCS (Figure 1e and 1f, 1k and 1l).However, the climatological daily SST patterns used in the sensitivity experiment are different from those used in the control experiment.Except for the warm waters in the southernmost portion of the SCS (near the equator) and in the western North Pacific east of 120°E, the SSTs on each day were all below 29°C over the entire SCS, and even lower than 27.5°C in the northwestern SCS around Hainan Island, thus exhibiting a decreasing tendency from the southeastern to northwestern SCS (Figure 1m-1r).In addition, there are no significant day-to-day variations in the climatological SST patterns during the entire experimental period of the OCT10 event.
The simulated spatial distributions of surface winds and rainfall in the control experiment (Figure 1g-1l) were consistent with observations (Figure 1a-1f).The simulations reproduced the strong surface southwesterlies over the southern SCS and northeasterlies over the northern SCS on 29 September (Figure 1a and 1g), which formed large-scale horizontal shear flow and cyclonic vorticity over the SCS.Afterwards, the southwesterlies intensified and extended northward to the northeastern SCS in the control run, where they joined the northeasterlies, thus generating a closed cyclonic circulation centered around 12°N, 111°E (Figure 1h), north to the center of the observed TD (Figure 1b).Note that evident convergence of enhanced southwesterlies and southeasterlies occurred over the warmest SST region west of Luzon Island.Subsequent intensification of the TD and its northward migration to the vicinity of Hainan Island, during 3-5 October (Figure 1c and d), were also well captured by the control simulation (Figure 1i and 1j), with the heaviest rainfall also occurring on these days.As compared with previous days, the reduced winds on 7 and 9 October indicated a weakening of the TD in the observation (Figure 1e and 1f).While the converging surface wind fields in the control-run did not weaken significantly during 7-9 October (Figure 1k and 1l), due to the model error accumulation since there were no assimilation in the experiments.On the other hand, the control simulation underestimated daily rainfall around the TD, as compared with observations, and a slight discrepancy in the location of the simulated TD center is noted for most days.However, the control simulated flow pattern and the TD intensity during each day (Figure 1g-1l) were in agreement with observations (Figure 1a-1f), demonstrating that the WRF model is capable of qualitatively capturing the observed characteristics of atmospheric circulation and precipitation associated with TD development.
In contrast to the control experiment, in the sensitivity experiment, the simulated southwesterlies over the southern SCS were very weak on both 29 September and 1 October (Figures 1m and 1n).Consequently, the closed cyclonic circulation centered around 15°N did not appear until 3 October (Figure 1o), and it occurred two days later than in the control experiment.The weak TD subsequently moved westward rather than northward, and by 5 October its center was located over the eastern Indochina Peninsula (Figure 1p).Note that the strengths of the southwesterlies and southeasterlies on the eastern side of the TD were still weak during these days, as compared with the control experiment, which may be due to the cooler climatological SSTs that largely reduced the SH flux from the ocean to the atmosphere, such that no significant temperature gradients formed in the atmospheric boundary layer [16].In contrast to the situation in the control experiment, the TD in the sensitivity experiment already dissipated on 7 October (Figure 1q), and northeasterlies prevailed over the SCS instead.Furthermore, two days later, the SCS was dominated by a saddle-backed wind field.Note that this weak TD existed for only four days.In fact, the development of such a short-lived TD might depend largely on favorable atmospheric factors, as the sensitivity experiment was run under the same atmospheric initial and lateral boundary conditions as the control experiment.
The thermodynamical role of the SSTAs in TD development can be further demonstrated by the temporal variations of surface enthalpy fluxes in the simulations (Figure 2).The time series of the area-averaged LH flux in the control experiment showed an increase during 1-6 October (on the order of 100-300 W m -2 ), reaching a maximum of 340 W m -2 when the TD had strengthened to its highest intensity on 6 October.Afterwards, the area-averaged LH flux gradually decreased, while still larger than 200 W m -2 , indicating considerable heat and moisture transferring into the upon atmosphere to maintain the TD due to the positive SSTAs.Although the simulated LH fluxes in the control run were somewhat larger than the observed ones (100-280 W m -2 , with maximum on 5 October) during the duration of the TD, both of them were in the order of magnitude, which also demonstrated that the control experiment is capable of reproducing the development of the TD in the observation.On the other hand, in the sensitivity experiment, the LH flux remained more or less constant during the period when the short-lived TD was discernable, only 120 W m -2 .Note that the simulated LH fluxes in the sensitivity run were in accord with the observed LH fluxes in climatology (140-180 W m -2 ).The main reason for the slight difference between the above two may be the difference in TD center position.Such LH flux under the cooler climatological SSTs was nearly 30% of that in the control run forced by the dramatically warm anomalies.
Figure 2. Time series of area-averaged (over an area within a 300-km radius of the surface vortex center) daily sensible heat (SH) flux (W m -2 ) during the period 30 September-9 October for the observation in the OCT10 event (green solid lines; indicated by "SH in 2010" in the legend) and in the climatology (green dash lines; indicated by "SH in climatology" in the legend), control experiment (red solid lines; indicated by "SH in CTRL-run" in the legend) and the sensitivity experiment (red dashed lines; indicated by "SH in SSTC-run" in the legend); latent heat (LH) flux (W m -2 ) for the observation in the OCT10 event (purple solid lines; indicated by "LH in 2010" in the legend) and in the climatology (purple dash lines; indicated by "LH in climatology" in the legend), control experiment (blue solid lines; indicated by "LH in CTRL-run" in the legend) and the sensitivity experiment (blue dashed lines; indicated by "LH in SSTC-run" in the legend); and enthalpy (SH plus LH) flux (W m -2 ) for the observation in the OCT10 event (orange solid lines; indicated by "SH+LH in 2010" in the legend) and in the climatology (orange dash lines; indicated by "SH+LH in climatology" in the legend), control experiment (black solid lines; indicated by "SH+LH in CTRL-run" in the legend) and the sensitivity experiment (black dashed lines; indicated by "SH+LH in SSTC-run" in the legend).
Since there was no obvious distinction between the SH flux in the control run and sensitivity experiment (alike only 10 W m -2 ), the enthalpy fluxes in both of the control and sensitivity experiments exhibited variations similar to those of the LH fluxes.Thus, high SSTs (above 29.5°C) in the SCS may be a prerequisite for the formation and intensification of TCs, at least during the months of September and October.Although the sensitivity experiment with cooler SSTs (below 29°C) also showed the formation of a short-lived TD, the cooler SSTs could not assist continued intensification of the TD.
To further substantiate the influence of warm SSTAs on TD-related rainfall, we investigated the time series of area-averaged simulated daily precipitation over the northwestern SCS around Hainan Island during the OCT10 event (Figure 3).The control experiment reproduced rainfall amounts of approximately 25-30 mm in the first two days, a pronounced increase from 2 to 6 October (on the order of 20-60 mm; coinciding with the maximum intensity of the TD), and a subsequent gradual decrease.The temporal variations in the simulated daily rainfall showed a resemblance with the observations, although the simulated amounts were less than observed amounts on corresponding days.As compared with observations, the control simulation underestimated daily rainfall around the TD, showing possible shortcomings of cumulus parameterization and microphysical parameterization schemes in simulating extreme precipitation.However, the simulated daily rainfall in the sensitivity experiment was much less than that in the control experiment, especially for the strong TD period of 4-7 October.The differences between the simulated daily rainfall in the sensitivity and control experiments demonstrate that the persistent warm SSTAs in the SCS helped prolong the TD, and as a result, enhanced the rainfall associated with the extended OCT10 TD event.

Summary and discussion
To validate the importance of warm ocean anomalies in the SCS on the formation and intensification of the TD and the resultant fortified effect of such extended TD on the heavy rainfall on Hainan Island, a WRF model was used to conduct numerical experiments with and without the observed warm SSTAs in the SCS, to understand the mechanism for TD development during the OCT10 event.The control experiment using the observed warm SST patterns reasonably reproduced the extended TD, while the sensitivity experiment with the climatological SST distributions reproduced only a weak short-lived TD, indicating the importance of air-sea interactions and persistent warm SST (actually upper-ocean) anomalies for the genesis and intensification of the TD.The simulated daily rainfall in the sensitivity experiment was much lower than that in the control experiment, especially for the intense-precipitation period of 4-7 October.The variations in the daily rainfall simulations between the sensitivity and control experiments illustrate that the persistent warm SSTAs in the SCS indeed contributed to the extension of the TD, which in turn increased the associated rainfall.
Because the control experiment using the WRF model was forced only by anomalous warm SSTs in the SCS, the forcing represented a one-way interaction rather than two-way interactions between the atmosphere and the ocean; the strong surface wind anomalies might have been generated by changes in the boundary layer virtual temperature distribution, as a direct response to the warm SSTA distribution.Therefore, the interrelationships between oceanic subsurface conditions, the translation speed, and the air-sea enthalpy flux required for intensification of TCs in the SCS should be further explored on the basis of more TC cases and the use of a fully coupled atmosphere-ocean model.

Figure 1 .
Figure 1.(a)-(f) Distribution of obseverd daily sea-surface temperatures (SSTs) (color scale, °C), which acted as a surface boundary forcing in the control experiment, and daily-mean FNL surface winds at 10 m above sea level (vectors, m s -1 ) and observed rainfall from GPM dataset (grey color scale with filling patterns, mm), shown at 2-day intervals for the period 29 September-9 October 2010.(g)-(l) As in (a)-(f), except for the control experiment, and (m)-(r) for the sensitivity experiment.The typhoon symbol marks the daily mean position of the tropical depression (TD) center.

Figure 3 .
Figure 3.Time series of area-averaged simulated daily precipitation (mm) over the northwestern South China Sea (SCS) (16°-21°N, 105°-113°E) around Hainan Island, for the observation (green bars; indicated by Obs Rainfall in the legend), the control experiment (red bars; indicated by CTRLrun in the legend), the sensitivity experiment (blue bars; indicated by SSTC-run in the legend), and the difference between the control and the sensitivity experiment results (black bars; control -sensitivity results).