Promising prediction of the monsoon trough and its implication for tropical cyclone activity over the western North Pacific

The monsoon trough (MT) is generally recognized as a feeding ground for tropical cyclones (TCs) over the western North Pacific (WNP). In view of the many challenges that remain in current seasonal TC forecasting, it would be a profound benefit to understand the predictability of variations in the MT and the implications of this for the seasonal prediction of TC activity. This study reveals that high predictability of the MT is shown by the current atmosphere–ocean coupled forecasting system, with the correlation coefficient being 0.84 for the model-ensemble prediction with observations from 1960 to 2005. This high predictability arises mainly from the tropical dipole sea surface temperature over the Maritime Continent and tropical Pacific Ocean, which favors convection around the warm pool and further excites the vorticity anomalies over the WNP. It is further found that good knowledge of the MT could provide promising prediction of TC activity over the WNP, including the occurrence and energy of TCs. The findings of this study suggest that coupling between the WNP circulation and tropical ocean acts as an important source of seasonal predictability in the WNP, and highlight the importance of the MT for seasonal prediction of TCs over the WNP.


Introduction
Tropical cyclones (TCs) occur more frequently over the western North Pacific (WNP) than any other basin of the tropical oceans. About 30 TCs form over the WNP every year and their frequency is high between July and November (Chan 2005). Frequent TCs over the WNP have a great impact on people's livelihoods and cause huge economic loss over East Asia. Many attempts have been made to seasonally predict TC activity using different methods for different regions, but this remains a challenge for modern science and technology (e.g. Camargo et al 2007, Zhan et al 2012, Camp et al 2015. In view of the high frequency of and great damage caused by TC activity over the WNP, better means of forecasting occurrence and associated predictability would carry huge benefits for society and the economy in the region. Variation of the monsoon trough (MT) exerts a significant influence on the interannual variability of TC activity over the WNP (Chen et al 2004, Wu et al 2012, Molinari and Vollaro 2013. Due to the intertropical convergence zone between the lowertropospheric off-equatorial westerlies and the northern trade easterlies, more than 70% of WNP TCs form within the MT (Molinari and Vollaro 2013). This is often accompanied by a warm sea surface temperature (SST), low-level relative vorticity, water vapor and advection (Chen et al 2004), providing a favorable environment for the enhancement of lowlevel synoptic-scale wave disturbances that are key to the occurrence of TCs (Chen and Huang 2009). Regarding interannual variation, the longitudinal shift of the MT has a significant effect on the mean location of TC genesis (Wu et al 2012). In addition, the MT often varies in association with the WNP subtropical high and further modulates the track of TCs (Song et al 2013). Thus, profound impacts of the MT on TC activity suggest that an insightful understanding of seasonal predictability of the MT Original content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence.
Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. would aid better prediction of TC activity over the WNP.
Many forecasting approaches and methods, including statistical, dynamical and hybrid dynamical-statistical models, have been applied in recent decades for forecasting seasonal TC activity over the WNP (e.g. Vitart 2006, Li et al 2013, Camp et al 2015, Manganello et al 2016, Zhan and Wang 2016. Advances in prediction of seasonal TC activity largely rely on better understanding and capturing slowly evolving climate signals, such as vertical wind shear, changes in the El Niño-Southern Oscillation (ENSO) and other variations in tropical SST. ENSO and its associated tropical air-sea interaction are currently the main sources for coupled prediction over the WNP (Wang et al 2008, Li et al 2012, Kosaka et al 2013, Li et al 2014. In view of the teleconnections between tropical SST and large-scale circulation in the WNP, Wang et al (2013) showed high predictability of the WNP subtropical high by using an empirical model based on central Pacific SST anomalies and their feedback with the Indo-Pacific warm pool, further indicating that high predictability of the WNP subtropical high would favor the prediction of summer tropical storm activity over the WNP. Their work prospectively implies the feasibility and importance of applying the predictability of large-scale circulation in the WNP to the prediction of TC activity over the WNP (Zhan et al 2012).
Although it is one of the most direct and important factors modulating TC activity over the WNP, predictability of the MT and its sources have not been well understood so far. The correlation between the MT and WNP subtropical high is modest, around 0.5, suggesting relatively independent variation of the MT and subtropical high. In view of its intimate and direct contributions to the frequency and energy of TCs (Chen et al 2004, Wu et al 2012, Molinari and Vollaro 2013, the implication that the MT has an effect on the predictability of TC activity over the WNP, which also remains elusive, needs further validation. In this study we make a comprehensive assessment of the how changes in the MT can be used to predict TC activity during the TC season over the WNP. We use five coupled prediction models from the ENSEMBLE-based prediction of climate changes and their impacts (ENSEMBLES) retrospective forecasting system (van der Linden and Mitchell 2009) to investigate the predictability of the MT and explore its profound impact on the prediction of TC activity over the WNP.

Models, hindcasts and observational datasets
ENSEMBLES is an EU-funded integrated retrospective forecast (hindcast) project, which comprises five fully coupled atmosphere-ocean prediction models from the European Centre for Medium-Range Weather Forecasts (ECMWF), the UK Met Office (UKMO), the Leibniz Institute of Marine Sciences at Kiel University (IFM-GEOMAR), Météo-France (MF) and the Euro-Mediterranean Center for Climate Change (CMCC-INGV). It includes 45 ENSEMBLE members with nine initial conditions for each model, which were initialized using realistic estimations of their observed states. More information on the model components and initialization can be found in van der Linden and Mitchell (2009) and Doblas-Reyes et al (2010).
The corresponding hindcasts were initialized on 1 May with a seven-month-long integration for each year from 1960 to 2005; this provided large enough forecast sample sizes for a comprehensive understanding of the predictability of the MT. A simple composite method for all 45 ENSEMBLE members was used to calculate the multi-model ensemble (MME) prediction, excluding potential uncertainties in different initial perturbations and models. This study investigated the TC season from July to November.
Several observational datasets were used in this study to validate the model hindcast, including: (i) monthly mean precipitation from the Global Precipitation Climatology Project (GPCP) between 1979 and 2005 (Adler et al 2003), (ii) monthly mean circulation from the National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) reanalysis products (Kalnay et al 1996) and (iii) the National Oceanic and Atmospheric Administration (NOAA) extended reconstructed monthly mean SST V3b dataset (Smith and Reynolds 2004) for 1960-2005 In addition, to estimate TC activity we used the International Best Track Archive for Climate Stewardship (IBTrACS) v03r06 data to construct the frequency of occurrence of TCs, the accumulated cyclone energy (ACE) (Bell et al 1999) and other TC activity indices. In order to intuitively illustrate TC activity, the IBTrACS data for TC activity were first binned into each 2.5°Â 2.5°grid box. To diminish the bias due to assessment of TC intensity, this study mainly focused on typhoons with a 10 m averaged wind speed greater than 64 knots. Moreover, the results achieved here can apply similarly to tropical storms with wind speeds greater than 32 knots.

Predictability of the monsoon trough
Variation of the MT can be well represented by lowlevel cyclonic relative vorticity Lau 1992, Wu et al 2012). Thus, we first examine the predictive skill of 850 hPa relative vorticity as shown by the grid-point temporal correlation between the MME prediction and observation for July-November for 1960-2005 ( figure 1(a)). This suggests that the interannual variation of low-level relative vorticity is skillfully Environ. Res. Lett. 12 (2017) 074027 predicted by current coupled models. The skill over most of the MT region, in which there is large positive relative vorticity and interannual variability, exceeds the 95% confidence level according to Student's t-test. Figure 1(b) shows the interannual variation of the MT index between the model prediction and observation, defined as the 850 hPa relative vorticity averaged over the region (5°-20°N, 130°-180°E) following Wu et al (2012) but excluding the Philippines. The models exhibit a remarkable ability to predict MT seasonal activity. The MT index is reasonably reproduced in most of the hindcast years, and the correlation between the ensemble mean and observation is 0.84 over the whole 46 years. This skillful prediction can also be found in all five models in ENSEMBLES, with the prediction correlation ranging from 0.63 (MF) to 0.84 (ECMWF) (table 1). In addition, the MT is sustainably predictable during the typhoon season from July to November, despite the increase in prediction lead months. The correlation between the MME prediction and observation is 0.79 for July and 0.55 for November, which still exceeds the 99% confidence level. This sustainability in the MT prediction suggests a high predictability associated with various types of slowly varying prediction sources (e.g. tropical SSTs). The signalto-noise ratio of the MT index, which is defined as the standard deviation of the ensemble mean divided by the averaged ensemble member standard deviation (Kumar 2009, Scaife et al 2014, is 0.91, further supporting the high predictability of the MT.  Furthermore, sources of the predictability of the MT are identified via linear regression of various fields onto the normalized MT index, as shown in figure 2. The MT is characterized by strong low-level cyclonic circulation over the tropical WNP, together with the east-west-oriented dipole of SST and precipitation anomalies between the Maritime Continent and the tropical Pacific Ocean. The correlation coefficient between the MT index and tropical dipole SST, defined as the difference in SST between (5°S-5°N, 170°W-120°W) and (10°S-10°N, 110°-150°E) (figure 2(e)), is 0.88 for observations during 1960-2005. The close teleconnection patterns imply a positive feedback between the MT and dipole SST over the Maritime Continent and tropical Pacific Ocean. The anomalous warm SSTover the central eastern Pacific could excite a westward-propagated Rossby wave, enhancing convection around the Philippine Sea and thus positive vorticity over the tropical WNP with a low-level cyclonic anomaly. This positive vorticity, on the other hand, promotes strong westerly winds along the equator, which would further reinforce the dipole SST over the tropical Pacific Ocean by transporting warm water in the surface sea to the east.
The above teleconnections associated with the MT can be well captured by the models, including the lowlevel anomalous cyclone over the MT region (figures 2 (b)), the dipole precipitation (figures 2(d)) and the SST pattern (figures 2(f)) around the Maritime Continent and tropical Pacific Ocean. This implies that the positive feedback between the MT and SST dipole over the Maritime Continent and tropical Pacific Ocean acts as the primary source of the predictability of the MT. The dipole-like SST anomalies over the Maritime Continent and central eastern Pacific Ocean, which could favor convection around the warm pool and excite vorticity anomalies over the WNP, make a significant contribution to the skillful prediction of the MT. The corresponding correlation coefficient is 0.81 between the predicted dipole SST and MT index. Furthermore, in view of the slowly varying SST, positive feedback can persist for months during the typhoon season, corresponding well to sustainable predictive skill from July to November (table 1).
The dipole SST exhibits a certain relationship with the ENSO in view of the significant anomalously warm SST found in the tropical eastern Pacific (figures 2(e) and (f)). The correlation coefficient between the dipole SST pattern and the Niño 3 index (5°S-5°N, 150°W-90°W) is 0.88 for observations. Nevertheless, compared with the traditional ENSO, the dipole SST demonstrates a relatively larger teleconnection with MT variation. The correlation coefficient between the MT and Niño 3 index is 0.78 (0.71) for observations (MME prediction), lower than for the correlation with the dipole SST. This further suggests that the dipole SST between the Maritime Continent and tropical Pacific Ocean comprises a more predictable signal for seasonal predictability of the MT. In addition, a relatively closer relationship of the MT index is shown with the tropical eastern Pacific part of the dipole SST

Implication for prediction of TC activity
While providing favorable dynamic and thermodynamic conditions for TC activity (Chen et al 2004, Wu et al 2012, Molinari and Vollaro 2013, the detailed contributions of the MT over the WNP need to be verified via a statistical correlation between the abovedefined MT index and TC activity at each grid point ( figure 3). It is found that the MT exerts a large influence on both the frequency of occurrence of TCs and the ACE, with a significant positive correlation around the Philippine Sea. The key regions of TC activity related to the MT correspond well to those regions with a large interannual variability of relative vorticity. When the MT is stronger with a positive relative vorticity anomaly, more TCs and increased TC intensity tend to occur around the Philippine Sea, and vice versa. The correlation coefficients between the MT index and the averaged TC frequency and ACE over the tropical WNP (10°-30°N, 135°-160°E) are 0.61 and 0.62, respectively. In view of the significant teleconnections with the tropical dipole SST over the Maritime Continent and central eastern Pacific Ocean, as shown above (figure 2), this further suggests that the MT acts as a key factor in bridging the tropical air-sea interaction and TC activity over the WNP. Based on skillful seasonal prediction of the MT, we attempt to predict TC activity directly using the MT index as the only predictor. Statistical transfer of largescale features of the model into other target predictions, including some regional rainfall and mesoscale systems, has been successful (e.g. Zhu et al 2008, Ndiaye et al 2011. A cross-validated reforecast, built by a linear regression method leaving one target year out for prediction (Michaelsen 1987), was used here to test the predictive capacity of TC activity averaged over (10°-30°N, 135°-160°E). Figure 4 shows the time series of observed and reforecast TC frequency  1972 and 1997 and negative anomalies in 1973 and 1999, can be well reproduced. Furthermore, the MT could also be capable of predicting the location of TCs over the WNP, as a similar predictive skill could be achieved for the location of TC formation (the average longitude over the WNP when a TC is generated), with a temporal correlation coefficient of 0.60 shown by the crossvalidated reforecast. The high predictability suggests that the MT is a promising means for seasonal prediction of TC activity over the tropical WNP. A cross-validated reforecast of TC activity was also performed using the tropical dipole SST over the Maritime Continent and tropical Pacific Ocean as a predictor, instead of the MT index. The reforecast skill is 0.54 for both the number of TCs occurring and ACE index. In addition, the correlation coefficients of the number of TCs and the ACE index with the tropical dipole SST are 0.64 and 0.65, respectively, similar to those with the MT index. This implies that the above tropical dipole SST makes a remarkable contribution to the seasonal prediction of TC activity over the WNP. The models are quite capable of reproducing the tropical dipole SST, with a predictive skill 0.85 for the MME result. The contributions of the tropical dipole SST described here are similar to those in Zhan and Wang (2016), which demonstrates the importance of tropical SST for the prediction of ACE over the WNP based on a hybrid dynamical-statistical prediction.
Nevertheless, compared with the TC activity predicted by the tropical dipole SST, a more skillful reforecast is achieved by the MT index. This further suggests that variation of the MT could effectively bridge tropical SST and WNP TC activity, and comprise more predictable signals for the seasonal prediction of TC activity over the WNP.
This study reveals that TC activity predicted by the MT tends to be skillful over a wide region of the tropical WNP, including frequency, energy and formation location. In comparison, the WNP subtropical high is more closely related to variation of the TC track and favors the prediction of tropical storm occurrence over the subtropical WNP around the west front of the subtropical high . In addition, the faithful reforecast demonstrated in this study employed the MT as the only predictor and directly used the output of model hindcasts. Greater prediction skill would be expected if an empirical model related to interannual variation of the MT is applied, in view of the quite high correlation between subtropical WNP tropical storms and the subtropical high in summer established by a physically based empirical model, which included tropical air-sea interactions and the North Atlantic Oscillation as predictors .

Summary and discussion
This study assesses the predictive skill of variation in the MT during the TC season, identifies the sources of predictability and investigates the implications for the  1960 1970 1980 1990 2000 1965 1975 1985 1995 2005 1960 1970 1980 1990 2000 1965 1975 1985 1995 2005  Environ. Res. Lett. 12 (2017) 074027 seasonal prediction of TC activity over the WNP. It is found that the remarkable predictive skill of the MT is seen in current coupled models, mainly as a result of its significant positive feedback with the dipole SST over the Maritime Continent and tropical Pacific Ocean. Our further findings reveal that this positive interaction between the MT and tropical oceans acts as a primary source for the seasonal prediction of TC activity over the WNP, including the frequency, intensity and formation location of TCs, and implies huge potential for understanding the predictability of TCs.
As an important environmental factor for TC activity over the WNP, the MT could provide a key basis for further improving the seasonal prediction of TC activity. Furthermore, it is anticipated that a comprehensive TC seasonal forecast system could be built by combining the MT with other TC-related factors, such as the WNP subtropical high and vertical shear, based on dynamical or statistical downscaling; this deserves further work.
Using a multi-model approach, this study delivers a definitive statement on the current capacity of global coupled climate models to predict the MT and on further development for TC activity forecast. A simple statistical cross-validated prediction is used here. While we build a reasonable predictive link between the MT to TC activity, high predictability of the MT and the associated statistical downscaling method further imply a convenient means for climate services to produce a seasonal forecast of TCs.