L.A.D. Buddik Bndurthn , Lu Wng ,∗ , Xun Zhou , Yifeng Cheng , Lin Chen

a Key Laboratory of Meteorological Disaster, Ministry of Education (KLME)/Joint International Research Laboratory of Climate and Environmental Change

(ILCEC)/Collaborative Innovation Center on Forecast and Evaluation of Meteorological Disasters (CIC-FEMD), Nanjing University of Information Science and Technology, Nanjing, China

b Laboratory for Regional Oceanography and Numerical Modeling, Qingdao National Laboratory for Marine Science and Technology, Qingdao, China

Keywords:Intraseasonal oscillation Sri Lanka Southwest monsoon Subseasonal forecast

ABSTRACT

1. Introduction

Sri Lanka is located in the southernmost part of the Indian subcontinent, close to the major Indian Ocean sea lanes. The country covers 65 610 km2, with 64 740 km2of land and 870 km2of water. The climate is strongly affected by the South Asian monsoon. For the people living in this country, the year can be divided into four seasons according to monsoon variation. The southwest monsoon (SWM) mainly occurs from May to September (five months), while the northeast monsoon prevails from December to February (three months). March—April is the first inter-monsoon period, and October—November is the second.During the SWM season, the island suffers its most rainfall, which influences the agriculture, irrigation, hydropower generation, and economy of this country deeply. Therefore, understanding the mechanisms of the SWM variation and the current skill of models to forecast it is of great importance to the people of Sri Lanka.

The rainfall of Sri Lanka has been reported to typically experience three to four active periods in a single SWM season ( Webster et al.,1998 ), indicating a strong subseasonal oscillation. This is understandable, as the Indian Ocean shows the strongest atmospheric intraseasonal oscillation (ISO) in the world ( Madden and Julian, 1971 ). The northward propagating boreal summer ISO from the equator to the Bay of Bengal can modulate the active and break phases of the South Asian monsoon significantly ( Lawrence and Webster, 2002 ). However, to the best of our knowledge, a complete study on the ISO of the SWM over Sri Lanka is still lacking. Thus, it is imperative to reveal the dominant modes of the ISO of the SWM, which is one aim of the present study.

Subseasonal forecasting of rainfall is of great practical significance,but is also a considerable challenge to the meteorological community( Coelho et al., 2018 ). In recent years, the World Weather Research Program/World Climate Research Program has established the Subseasonal to Seasonal Prediction project (S2S) towards a better understanding of the forecast skills of current dynamic models and improving them at the subseasonal-to-seasonal time scale ( Vitart et al., 2017 ). In past years, the subseasonal forecast skills of different models with respect to the global monsoon have been evaluated (e.g., Liu et al., 2014; Nicolas et al., 2017;Li et al., 2020 ). For example, the NECP Climate Forecast System, version 2, has been reported to be skillful in forecasting global monsoon indices at leads of about two weeks, while apparent interannual differences exist ( Liu et al., 2014 ). Nevertheless, the subseasonal forecast skills of current dynamic models for the SWM over Sri Lanka remain unknown. As the European center for Medium-Range Weather Forecasts (ECMWF)model is superior in subseasonal forecasts in many aspects among the S2S models (e.g., Kim et al., 2014 ; Zhou et al., 2019 ; Xie et al., 2020 ),we employed it to evaluate the subseasonal forecast skill for the SWM in Sri Lanka based on its reforecast experiment data.

The remainder of this paper is organized as follows. The data and methods are described in Section 2 . In Section 3 , the dominant modes of the ISO of the SWM are revealed, and then Section 4 the subseasonal forecast skills for SWM indices are assessed using the ECMWF model.Finally, a summary is given in Section 5 .

2. Data and methods

2.1. Data

The observational data employed in this study were the (i) daily precipitation over the land areas of Sri Lanka from the Climate Hazards Group Infrared Precipitation with Station Data (CHIRPS) product, at a horizontal resolution of 0.05° × 0.05°, (ii) daily precipitation from the 1DD (1° daily) product of the Global Precipitation Climatology Project(GPCP) dataset, version 1.1 ( Huffman et al., 2001 ), and (iii) daily wind at 850 hPa from ERA-Interim ( Dee et al., 2011 ) archived using the horizontal resolution of 1.5° × 1.5° The time span of these datasets was from 1999 to 2010. The daily outgoing longwave radiation (OLR) data( Liebmann and Smith, 1996 ) from NOAA’s polar-orbiting satellites on 2.5°× 2.5°grids were also used. To evaluate the subseasonal forecast skill of the SWM over Sri Lanka by dynamic models, the reforecast experiment result produced by the ECMWF model from the S2S database( Vitart et al., 2017 ) was analyzed. Specifically, five perturbed integrations from version CY41R1 were employed. This version of the Integrated Forecast System was first integrated for 10 days with a resolution of TL639 (～32 km) in the atmospheric model, and for the rest of the forecast days with a resolution of TL319 (～64 km). The hindcast results during 1999—2010 were compared with observations.

2.2. Extracting the dominant ISO modes

An empirical orthogonal function (EOF) analysis on the daily precipitation anomalies over Sri Lanka during the monsoon season was conducted to extract the dominant intraseasonal modes of the SWM.Prior to the EOF analysis, the daily climatology of precipitation was removed, and then a five-day running mean was calculated to remove the synoptic-scale signals (e.g., Yang et al., 2010 ; Cheng et al., 2020 ). The results were found to be insensitive to a change in number, such as four or seven (figures not shown). A spectral analysis was conducted on the time series of principal components of each year, and then the multi-year average was calculated to reveal the dominant intraseasonal periods of these modes. Composite analysis was also used, based on the intraseasonal rainfall events selected according to the PC time series, to show the evolutionary characteristics of intraseasonal rainfall and circulation anomalies.

2.3. Evaluating the subseasonal forecast skill

To assess the forecast skill for the SWM over Sri Lanka in the ECMWF model, the monsoon indices in the reforecast experiment were compared with observations using the temporal correlation coefficient(TCC), which was estimated as follows:

Here,Xiis the observed monsoon index andfiis the predicted monsoon index for a lead time ofτdays.Nis the number of forecasts, and an overbar represents the time average. The daily monsoon indices were calculated using a five-day running mean to remove synoptic-scale signals. Thet-test was used to evaluate the forecast skill for the monsoon indices, as in previous studies (e.g., Xue et al., 2010 ).

3. Dominant intraseasonal modes of the swm over Sri Lanka

Prior to revealing the dominant intraseasonal modes of the SWM over Sri Lanka, we first display the climatology during the SWM season(May-September). Fig. 1 (a) shows the climatology of rainfall in Sri Lanka using CHIRPS data. Abundant rainfall is confined to over the southwestern part of the island, while it is less over the northern part due to the hilly topography in the central part of Sri Lanka. Fig. 1 (b) shows the climatology of rainfall derived from GPCP data and the 850-hPa winds in ERA-Interim over a larger region. The rainfall distribution over Sri Lanka is similar to that shown in Fig. 1 (a), but the amplitude is less than one-third of that derived from the satellite data. This is due to the coarser resolution in the GPCP data than the CHIRPS data.

Fig. 2 (a) displays the multi-year mean power spectrum of the Sri Lanka rainfall anomaly from May to September. Based on the 99% confidence level, the significant frequency peaks mostly lie in the 10—35-day band. This is consistent with previous studies that reported a dominant biweekly mode during the South Asian summer monsoon ( Goswami and Mohan, 2001 ; Annamalai and Slingo, 2001 ). The GPCP data also show a significant 10—35-day oscillation (figure not shown).

EOF analysis was then applied to the 10—35-day filtered rainfall anomaly derived from CHIRPS during the SWM season. The two leading EOF modes account for 66% and 10% of the total variance, respectively. The two modes are distinct from each other, as well as from the remaining modes, according to the rule of ( North et al., 1982 ). Fig. 2 (b)presents the horizontal pattern of the first EOF mode (EOF1). It is primarily characterized by prominent positive rainfall anomalies confined to over the southwestern coastal area of the island, similar to the climatology shown in Fig. 1 (a). This suggests that the subseasonal oscillation of the rainfall anomalies could contribute to the accumulated seasonal rainfall. The second EOF mode (EOF2) is characterized by positive anomalies in the southern coastal area (figure not shown). The lead—lag correlation coefficients between the first and second principal components (PC1 and PC2) are insignificant on each day, indicating the independence of the two leading modes (figures not shown).

The evolutionary characteristics of the 10—35-day rainfall mode during the SWM season were investigated through composite analysis.Strong rainfall events were first identified as when the standardized time series of PC1 exceeded 1.5, and the peak day was defined as day 0. Thirty-three cases were selected with these criteria, and then the 10—35-day filtered rainfall and circulation anomalies were composited for these events. Fig. 3 presents the evolution of anomalous rainfall, OLR and 850-hPa winds associated with EOF1 from day − 6 to day 0. On day 0, Sri Lanka shows a positive rainfall anomaly over its southwestern coastal area, which is identical to EOF1. Meanwhile, the island is under the control of a large-scale cyclonic anomaly with a center to its southwest. It can also be seen that weak southwesterlies prevail over the southwestern coastal area on this day, indicating that the SWM would enhance. Enclosed cyclonic circulation forms over the equatorial eastern Indian Ocean on day − 4, and then propagates westward. Also, the rainfall over Sri Lanka changes from negative to positive. Corresponding to the westward cyclone is a northward active convection anomaly (represented by a negative OLR anomaly), with its center near 4°S on day − 6,and at about 5°N on day 0. This suggests that the subseasonal change in the rainfall anomaly over Sri Lanka may be related to the westward propagation of atmospheric ISO.

Fig. 1. May-September mean fields of (a) rainfall (shading; units: mm d − 1 ) in Sri Lanka derived from CHIRPS data, and (b) rainfall (shading; units: mm d − 1 ) from GPCP data and 850-hPa winds (vectors; units: m s − 1 ) from ERA-Interim data. The white rectangle covers the area of (5°—12°N, 79°—90°E).

Fig. 2. (a) Multi-year summer (May—September) mean power spectrum of the Sri Lankan rainfall anomaly after removal of the climatology and synoptic fluctuations by a 5-day running mean. The red dashed line denotes the 99% confidence level. (b) The horizontal pattern of the first EOF mode of the 10—35-day rainfall anomaly over Sri Lanka during the SWM season.

4. Subseasonal forecast skill for monsoon indices in Sri Lanka

The aim of this next part of the study was to reveal the subseasonal forecast skill of current numerical models (as represented by the ECMWF model) with respect to the SWM in Sri Lanka. Before the evaluation,we needed to define some useful indices that could reflect the dominant characteristics of the SWM and that were easy to calculate using gridded model data. As shown in Fig. 1 (b), the monsoon season—averaged rainfall over Sri Lanka is strongly related to more rainfall in its surrounding area and a large-scale southwesterly over the Indian Ocean. We compared the seasonal variation of the area-averaged rainfall over the land areas of Sri Lanka based on CHIRPS data and that over a larger box area(LBA; see white box in Fig. 1 (b)) based on the GPCP data (see Fig. S1(a)).The fluctuations of the two rainfall time series are generally consistent,with a correlation coefficient of 0.78. Therefore, the LBA-averaged rainfall time series based on GPCP data is used as the monsoon rainfall index in the following assessment. Next, the seasonal variation of the 850-hPa zonal and meridional winds averaged over the LBA were investigated,based on reanalysis data (see Fig. S1(b)). The two time series are both positive during May to September, indicating a strong southwesterly wind. Therefore, in the following evaluation, the LBA-averaged zonal wind at 850 hPa is used as the monsoon wind index.

Fig. 3. Composite evolution of anomalous precipitation (shading; units: mm d − 1 ), OLR (green contours; units: W m − 2 ) and 850-hPa winds (vectors; units: m s − 1 )from day − 6 to 0 with an interval of 2 days. Day 0 denotes the peak day of the PC1 for each selected subseasonal event. The letter C stands for cyclonic circulation anomaly.

Fig. 4(a,b) shows the prediction results of the two monsoon indices during the monsoon season (May − September) at different lead times.For the rainfall index, it is closer to the observation during the days of monsoon onset and retreat, while the biases become larger during the prevailing monsoon period. Generally speaking, the predicted evolution of the wind index is better than that of the rainfall index, as its fluctuation is closer to its observed counterpart. However, the intensities of the two indices are both overestimated, and the biases increase as the lead time increases. This is a common problem in many other models (e.g.,Wu et al., 2017 ).

Next, the prediction skill of the model with respect to the two indices in each year was investigated. Fig. 4 (c,d) shows the TCC between the predicted and observed monsoon indices as a function of lead time.The multi-year mean results show that the forecast skill of the rainfall index drops below the 99% confidence level on about 13 days, and that of the wind index does so on more than 30 days. Therefore, the forecast skill for the wind index is higher than that of the rainfall index on average. Note that the forecast results beyond 30 days are not exhibited because of the data availability. Furthermore, the forecast skills for both indices show remarkable interannual differences. For example, it takes 4 days for the rainfall index to fall into the range of unskillful predictions in 2002, whereas it takes 29 days in 2000. Also, the forecast of the wind index in 2003 only shows skill within two weeks,while it is skillful at leads of more than four weeks for most of the other years.

Next, the most skillful and the least skillful cases based on the TCC of the rainfall index were analyzed to understand the interannual difference in the prediction skill for Sri Lankan monsoonal rainfall. Fig. 5 displays the pattern correlation coefficients between the rainfall and LBA rainfall index and the regressed 850-hPa wind field against the rainfall index in 2000 and 2002 for the observation and prediction at different lead times, separately. In 2000, the observed stronger-than-normal rainfall near Sri Lanka over the white box is strongly correlated with the positive rainfall anomaly west of the southernmost end of the Indian subcontinent.

Such a rainfall pattern is further related to a large-scale cyclonic circulation anomaly ranging from 60°E to 100°E, which is favorable for rainfall occurrence as it provides low-level convergence. The ECMWF model captures the observed rainfall—circulation relationship reasonably at all leads. In contrast, the observed stronger-than-normal rainfall near Sri Lanka in 2002 is related to a positive rainfall anomaly over the Arabian Sea and another one over the Maritime Continent. The Sri

Fig. 4. (a) Seasonal evolution of the LBA (larger box area; see Fig. 1 (b)) rainfall (units: mm d − 1 ) during May to September based on GPCP rainfall (black curve)and the predicted results with different leads. The green curve represents the prediction averaged over the lead time of 0—4 days, the blue one denotes that of 5—9 days, the orange one denotes that of 10—14 days, and the pink one denotes that of 15—19 days. (b) As in (a) but for the LBA zonal wind (units: m s − 1 ) at 850 hPa.The values in brackets represent the biases of the predictions (i.e., forecasts minus observation) averaged from May to September at different lead days. Temporal correlations between observation and forecasts at different lead days for the (c) rainfall index and (d) 850-hPa zonal wind index. Black dashed lines denote statistical significance of the correlation at the 99% confidence level. Shown the three points running mean of lead days.

Lankan area is under the control of an easterly anomaly, which originates from the West Pacific. Such an easterly anomaly is part of a cyclonic circulation anomaly centered near 95°E, which is a response to the enhanced convection near the Maritime Continent ( Gill, 1980 ). Another cyclonic circulation anomaly centered at 65°E is apparent, which is related to the enhanced convection over the Arabian Sea. Nevertheless, the easterly winds near Sri Lanka are replaced by westerly winds in ECMWF, which is a part of the large-scale circulation over the Arabian Sea and Bay of Bengal. This is due to an overestimated link between the circulation and rainfall over the Arabian Sea and an underestimated link between the circulation and rainfall near the Maritime Continent.The incorrect relationship between the circulation and rainfall pattern may explain the unskillful prediction of rainfall near Sri Lanka in 2002.

The meridional propagation of the intraseasonal OLR anomaly over 60°—90°E was also investigated. As shown in Fig. S2, there is an obvious northward propagation of convection in 2000, while it is weaker in 2002. This may also result in the higher skill in 2000 than in 2002. As an improved subseasonal prediction system is expected to emerge from consideration of the seasonal evolution of the monsoonal ISO ( Liu et al.,2020 ), we will further study the seasonal evolution of the local intraseasonal variability and its predictability in the future.

5. Summary and conclusions

Sri Lanka, a small island country located near the southernmost end of the Indian subcontinent, is controlled by the SWM from May to September when Sri Lanka suffers the most accumulated rainfall in a year. Compared with extensive studies on the ISO of the Indian monsoon, less attention has been paid to the ISO of the SWM over Sri Lanka.Based on observational data, this study investigated the dominant intraseasonal mode of the SWM. It also evaluated the forecast skill of current dynamic models in simulating the SWM on a subseasonal time scale by using the ECMWF reforecast data from the S2S project.

Results indicated that the leading mode of SWM rainfall in Sri Lanka shows a significant variability on a 10—35-day time scale. It accounts for 66% of the fractional variance based on EOF analysis. The horizontal distribution of the leading mode of the rainfall anomaly is similar to its climatology, with the maximum over the southwestern part of the island, suggesting a contribution of the intraseasonal rainfall to the season-accumulated rainfall. Composite results based on the dominant EOF mode indicated that the development of the biweekly rainfall anomaly in Sri Lanka is associated with a westward-propagating anomalous cyclonic circulation.

Fig. 5. (a—e) Correlation coefficients (shading) between GPCP rainfall and the LBA (larger box area; see Fig. 1 (b)) rainfall index, and the regressed 850-hPa winds(vectors) against the rainfall index in the most skillful year (2000). Among them, (a) shows the observation and (b—e) the predictions at different lead days, which are marked in the upper-right corner of each panel. (f—j) As in (a—e) but for the least skillful year (2002).

Two monsoonal indices are proposed for evaluating the forecast skillone is the area-averaged rainfall, and the other is the area-averaged 850-hPa zonal wind. Generally speaking, the ECMWF model is more skillful in predicting the wind index than the rainfall index, with the former showing a skill beyond 30 days and the latter about two weeks.The forecast skills exhibit prominent interannual differences for both indices. A preliminary investigation of the most skillful and the least skillful cases suggested that a correct simulation of the large-scale circulation response to tropical convection is crucial for the subseasonal prediction of monsoonal rainfall over Sri Lanka.

Funding

This work was jointly supported by the National Key Research and Development Program of China [grant number 2019YFC1510004 ], the National Natural Science Foundation of China (NSFC) [grant number 41975108 ], and the NSFC-Shandong Joint Fund for Marine Science Research Centers [grant number U1606405].

Supplementary materials

Supplementary material associated with this article can be found, in the online version, at doi: 10.1016/j.aosl.2021.10062 .