Subseasonal Prediction of Early-summer Northeast Asian Cut-off Lows by BCC-CSM2-HR and GloSea5

Northeast Asian cut-off lows are strong cyclonic lows above cold surface anomalies near the east coast of Eurasia (Palmén and Newton, 1969). They often bring heavy precipitation and persistent cool weather over large areas (Hu et al., 2011; Gao et al., 2014; Xie and Bueh, 2015). Particularly, they are one of the most important circulation systems that are responsible for some of the most catastrophic floods (Zhao and Sun, 2007). Active Northeast Asian cut-off lows in early summer of 2021 induced severe convective weather in southeast China, giving rise to heavy casualties and large property losses. Given these strong impacts on regional temperatures and precipitation extremes, skillful subseasonal prediction of Northeast Asian cut-off lows is of great importance.

Observational studies have shown that Northeast Asian cut-off lows occur throughout the whole year and are most active during early summer (May and June). They are often accompanied by strong blocking highs over the Ural mountains and Okhotsk Sea, and a strong subtropical jet over East Asia (Hu et al., 2011; Xie and Bueh, 2015). As suggested by previous studies, the formation mechanisms for Northeast Asian cut-off lows exhibit strong seasonal differences. During winter, cut-off lows are often coupled with the quasi-stationary East Asian trough (Song et al., 2016). An upper-tropospheric Rossby wave train propagates across northern Eurasia and interacts with preexisting surface cold anomalies over central Siberia, which helps to intensify the East Asian trough. During early summer, cut-off lows are often initialized by a Rossby wave train originating from the subpolar North Atlantic, and are then reinforced rapidly by zonal local wave activity (LWA) flux convergence and local baroclinic eddy generation (Nie et al., 2022). In late-summer (July and August), Northeast Asian cut-off lows are strongly affected by the northward progressing East Asian summer monsoon, as this is when diabatic heating plays a more dominant role (Lin and Bueh, 2021). External factors, such as cold anomalies of offshore sea surface temperature and cold anomalies of land surface temperature over west Asia in the preceding spring, are also suggested to play a role in affecting summer-mean cut-off lows (Wang et al., 2018).

The study by Pinheiro et al. (2022) comprehensively evaluated the simulation ability for global cut-off lows in Coupled Model Intercomparison Project Phase 5 (CMIP5) and Phase 6 (CMIP6). It was shown that CMIP5 models systematically underestimate the cut-off low intensity. CMIP6 models simulate the spatial distribution of cut-off lows more realistically. The improvement of cut-off low simulation in CMIP6 models is mainly attributed to the improved representation of the background zonal wind. In this study, we aim to investigate the subseasonal prediction skill for Northeast Asian cut-off lows. Particularly, we focus on the early summer when East Asian cut-off lows are most active. During this period, the East Asian summer monsoon has still not reached Northeast Asia, and the stationary East Asian trough is very weak and thus exerts negligible impact on the formation of Northeast Asian cut-off lows. The dynamics of cut-off lows during this period is primarily determined by regional eddy-mean flow interaction (Nie et al., 2022). We show that both BCC-CSM2-HR and GloSea5 tend to underestimate the amplitude of cut-off lows and regional eddy-mean flow interaction responsible for the cut-off low evolution, and therefore may underestimate the large impacts on extreme temperature and precipitation events in Northeast Asia.

The paper is organized as follows. In section 2, we describe the data and diagnostic methods. Subseasonal prediction skill of early-summer Northeast Asian cut-off lows, and model ability to capture the essential dynamics for the cut-off low evolution are investigated in section 3. Section 4 summarizes our results.

In this study, we analyze cut-off low statistics derived from both reanalysis data and ensemble hindcast. Specifically, we use daily (1200 UTC) geopotential height, zonal wind and temperature ($ 1^\circ\times1^\circ $) on constant pressure levels over the period 1979−2020 from the European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis version 5 (ERA5). The ensemble hindcast is from two subseasonal to seasonal forecasting systems, one from the Beijing Climate Center (BCC-CSM2-HR model) and the other from the UK Met Office (GloSea5). The atmospheric component of the BCC-CSM2-HR model has a horizontal resolution of T266 (~45 km) with 56 vertical levels. The model details are described in Wu et al. (2022). A four-member ensemble of forecasts is run for the period 2005−2020, initialized twice a week. The climate model at the core of GloSea5 (MacLachlan et al., 2015) is the Hadley Center Global Environmental Model version 3 (HadGEM3), which has atmospheric resolution of $ 0.83^\circ\times0.55^\circ $ and 85 quasi-horizontal vertical levels. The oceanic resolution is $ 0.25^\circ\times0.25^\circ $ with 75 quasi-horizontal levels. A three-member ensemble of forecasts is run for the period 1993–2015, initialized from the 1st, 9th, 17th, and 25th day of each month.

Different cut-off low detection algorithms are designed to detect various aspects of cut-off lows throughout their lifetimes. In this study, we compare three different cut-off low indices using the ERA5 reanalysis and select one of them to analyze the subseasonal prediction skill for cut-off lows in the climate models.

● The most common cold vortex index is defined by satisfying the following criteria: at least one enclosed 500-hPa geopotential height ($ Z_{500} $) contour is observed over the key region (115°−145°E, 35°−55°N ), and the temperature at 500 hPa in the key region is lower than its surroundings (Hu et al., 2011).

● The meridional reversal index is designed similar to the definition of a blocking high by Tibaldi and Molteni (1990). We search for the meridional cyclonic reversal of 500-hPa geopotential height in the key region (115°−145°E, 35°−55°N). Specifically, for a longitude with a cut-off low, the meridional difference of geopotential height (GHGN) must hold the following criteria:

where Z denotes geopotential height, $\phi_{\rm{N}}=(\phi_{\rm{c}}+\frac{3}{2}\delta\phi)+\Delta$ and $\phi_{\rm{M}}=(\phi_{\rm{c}}+\frac{1}{2}\delta\phi)+\Delta$. Here, $\phi_{\rm{c}}$ is the center latitude 45°N, and $ \delta \phi=15^\circ $ based on the typical scale of cut-off lows. The definition allows a shift of $ \Delta=-4^\circ, 0^\circ, 4^\circ $ latitude to account for cut-off lows that are not located directly on $ \phi_{\rm{c}} $. Equation (1) requires that westerly flow is present poleward of the cut-off low region.

● The local wave activity index is designed following Nie et al. (2022) by detecting the large-amplitude cyclonic meandering of geopotential height over a key region of Northeast Asia (115°−145°E, 35°−55°N). To be more specific, the local wave activity of geopotential height at longitude $ \lambda $ and latitude $ \phi_{\rm{e}} $ is defined as:

Here, $\phi_{\rm{e}}$ is the equivalent latitude of the geopotential height contour at 500 hPa. ${z}_{\rm{e}}(\lambda, \phi) = z(\lambda, \phi)-Z_{500}(\phi_{\rm{e}})$ is an eddy term describing the deviation of the geopotential height contour from the eddy-free, zonally symmetric basic state. More details on ${\rm{LWA}}_{Z500}$ can be found in Chen et al. (2015). The daily local wave activity index is calculated as the normalized time series of the domain-averaged cyclonic component of ${\rm{LWA}}_{Z_{500}}$ over (115°−145°E, 35°−55°N).

For all three indices, the circulation anomalies associated with cut-off lows are investigated through lagged composites of these fields for strong cut-off lows. Figures 1a-c compare the composite maps of geopotential height and its anomaly for strong cut-off lows based on the three indices. There are 437 cut-off low days selected by the traditional cold-vortex index, 366 days selected by the meridional reversal index, and 375 days selected by the local wave activity index. Note that strong cut-off lows based on the local wave activity index are selected when the wave activity index is larger than its time mean by one standard deviation. The $ Z_{500} $ shows a strong trough and negative anomalies in Northeast Asia for all three indices. The intensity of the composite trough is weakest for the traditional cold-vortex index (Fig. 1a). The meridional reversal index searches for meridional reversal of $ Z_{500} $, and thus it is often accompanied by a strong blocking high to the north of the trough (Fig. 1b). The $ Z_{500} $ shows an enclosed contour in the trough center, and the intensity of the $ Z_{500} $ anomaly is relatively strong. The pattern correlations of the $ Z_{500} $ anomaly in the key region based on the local wave activity index and the other two indices are 0.78 and 0.91, respectively. Figures 1d-f further display the composite maps of temperature anomalies at 500 hPa for strong cut-off lows based on those three indices. It is clear that the temperature is lower than its surroundings. The negative temperature anomalies in the key region are strongest for the local wave activity index (Fig. 1f). The above analyses suggest that our results are not sensitive to the choice of cut-off low index.

Figure 1.  Diagnosis of cut-off lows. (a) Composite map of the total $ Z_{500} $ (black contours with interval of 50 m) and $ Z_{500} $ anomaly (shading, units: m) for strong cut-off lows measured by the cold-vortex index over Northeast Asia in early summer. (b), (c) as in (a) but for the composite map based on (b) meridional reversal index and (c) local wave activity index. (d)−(f) as in (a)−(c), but for the composite maps of temperature anomalies at 500 hPa (units: K). Values that are significant at the 95% confidence level are highlighted with black dots.

More importantly, the local wave activity is a dynamical field that can measure the cyclonic waviness of mid-tropospheric geopotential height contour, through which the budget for wave activity can be explicitly diagnosed. It can also yield a daily two-dimensional (longitude−latitude) map. Thus, the local wave activity is a good tool for measuring the intensity of cut-off lows (Martineau et al., 2017; Nie et al., 2022). Previous studies have widely employed this approach to diagnose the blocking high, wave events, and cut-off lows (Chen et al., 2015; Nakamura and Huang, 2018; Chen et al., 2022; Nie et al., 2022). In the following analyses, we decide to use the local wave activity index to measure the intensity of East Asian cut-off lows and refer to this index as the Northeast Asian cut-off low index. The circulation characteristics associated with cut-off lows are investigated through lagged composites of these fields for the positive phase of the local wave activity index (larger than its time mean by one standard deviation).

The daily cut-off low index in the model is defined as the local wave activity index similar to that in the ERA5 reanalysis. In order to evaluate the spatial patterns reproduced by the climate models, we apply composite analysis of $ Z_{500} $ against the cut-off low index in the model. More specifically, we firstly select the first-day output at each forecast date for each ensemble member. Then we get a time series of the cut-off low index for all those days in all ensemble members. Next, we select the strong cut-off lows from this time series when the value is larger than its time mean by one standard deviation. To get as many samples as possible, we use the entire hindcast period to perform the composite analysis (2005−2020 for BCC-CSM2-HR and 1993−2015 for GloSea5). Anomalous patterns of other variables associated with the cut-off low are obtained by composites of the variables against the cut-off low index for each ensemble member, and then averaged.

In this study, the subseasonal prediction skill for Northeast Asian cut-off lows is assessed by calculating the temporal correlation coefficients between the observed and forecast ensemble-mean cut-off low indices on a daily basis. Here, the ensemble-mean cut-off low index in the hindcast is calculated as follows. Firstly, we compute the cut-off low index for each ensemble member using the model output from the first 20 days. We then average the cut-off low index for all ensemble members. The persistence forecast using the autocorrelation of the observed cut-off low index provides a baseline forecast, and we consider a prediction skill useful only when it exceeds that of the persistence forecast.

Figure 2 compares the spatial pattern of Northeast Asian cut-off lows represented in the BCC-CSM2-HR and GloSea5 models. Here, the spatial pattern in the models is defined as the composite map of the total $ Z_{500} $ (contours) and $ Z_{500} $ anomaly (shading) for strong Northeast Asian cut-off index at the first-day output during early summer (192 days for BCC-CSM2-HR during 2005−2020 and 72 days for GloSea5 during 1993−2015). In both models, the total $ Z_{500} $ shows a strong trough in Northeast Asia, and the $ Z_{500} $ anomalies exhibit negative values in the trough center. Comparison between Fig. 2a and Figs. 2b, c illustrates that the spatial patterns of Northeast Asian cut-off lows in the models are in good agreement with the ERA5 reanalysis, but the negative $ Z_{500} $ anomalies are weaker than those in the ERA5 reanalysis. The pattern correlations in the key region between the $ Z_{500} $ anomalies from the two models and the reanalysis are r=0.9 for BCC-CSM2-HR and r=0.98 for GloSea5, respectively. This suggests that both models can capture the spatial pattern of Northeast Asian cut-off lows, but they underestimate the intensity. Figures 2d-f further compare the composite maps of 500-hPa temperature ($ T_{500} $) anomalies associated with cut-off lows in the reanalysis and forecasting systems. It is shown that both models can simulate the low temperature anomalies in the key region, but they underestimate the anomaly amplitude.

Figure 2.  Spatial pattern of Northeast Asian cut-off lows in the: (a) ERA5 reanalysis, (b) BCC-CSM2-HR model, and (c) GloSea5 model. Here, the spatial pattern is defined as the lag 0-day composite map of the total $ Z_{500} $ (black contours with interval of 50 m) and $ Z_{500} $ anomaly (shading, units: m) for strong Northeast Asian cut-off low index with a lead time of 0 days during early summer. (d)−(f) as in (a)−(c), but for the 500-hPa temperature ($ T_{500} $) anomalies. Values that are significant at the 95% confidence level are highlighted with black dots.

The above results demonstrate that the models are, to a large extent, able to reproduce the observed spatial pattern of Northeast Asian cut-off lows. We next focus on the subseasonal prediction skill. Figure 3a compares the time evolution of regional-mean LWA, starting from 1 June 2015, between the ERA5 reanalysis and two models. The two models can basically simulate the rapid growth of LWA in the first few days and its decay in the following days. Both models underestimate the strength of LWA from 7 June to 15 June. Figure 3b further displays the correlation coefficient between the cut-off low index in the ERA5 reanalysis and two climate models. The model prediction skill is much better than the observed self-persistence of the cut-off low index. The e-folding decorrelation time scale is 10.2 days for the BCC-CSM2-HR model and 11.4 days for the GloSea5 model, suggesting that the prediction skill limit for Northeast Asian cut-off lows is roughly 10−12 days in the current models. Note that the model prediction skill for cut-off low intensity may depend on the initial state of cut-off low evolution.

Figure 3.  Predictability of cut-off lows. (a) Time evolution of regional-mean LWA over Northeast Asia from 1 June 2015 in both the ERA5 reanalysis and ensemble-mean forecasts. (b) Temporal correlation coefficient between the cut-off low index in the ERA5 reanalysis and the forecast ensemble-mean daily cut-off low index in the BCC-CSM2-HR model and GloSea5 model. The magenta dashed line denotes the auto-correlation of the observed cut-off low index. The gray dashed line denotes a base line with a correlation coefficient of 1/e, which helps to determine the e-folding decorrelation time scale of the cut-off low index in observation and assess the prediction limit of the cut-off low index in the forecast systems.

We next examine the model ability to capture the essential dynamics responsible for the Northeast Asian cut-off low evolution. Following Nie et al. (2022), we apply lagged composite analysis of $ Z_{200} $ against the cut-off low index in both the reanalysis and models. The time evolution of early-summer Northeast Asian cut-off lows can be roughly divided into three stages: initial stage (lags −6 to −3 days), rapid growth stage (lags −2 to −1 days), and decay stage (lags 0 to 4 days). The left column of Fig. 4 displays the observed composite maps of $ Z_{200} $ anomalies during these three stages. During the initial stage, the $ Z_{200} $ anomalies display a cyclonic anomaly in the subpolar North Atlantic, an anticyclonic anomaly in the Ural region, and a cyclonic anomaly in Northeast Asia (Fig. 4a), implying a Rossby wave train propagating from the subpolar North Atlantic to Northeast Asia [see the wave activity flux plot of Fig. 2 in Nie et al. (2022)]. During the rapid growth stage, the negative $ Z_{200} $ anomalies become much stronger in Northeast Asia (Fig. b4), amplifying the preexisting cyclonic low. During the decay stage, the negative $ Z_{200} $ anomalies in Northeast Asia become gradually weaker and move downstream.

Figure 4.  Observed and modeled life cycle of cut-off lows. (left) Lagged composites of the anomalous $ Z_{200} $ (contours, contour interval: 8 m) and $ T_{500} $ (shadings, units: K) for strong early-summer Northeast Asian cut-off lows in the ERA5 reanalysis. The middle and right columns are the same as the left column but for the composite maps in BCC-CSM2-HR model and GloSea5 model, respectively. Values that are significant at the 95% confidence level are highlighted with black dots.

The middle column and right column of Fig. 4 show the composite $ Z_{200} $ anomalies associated with the cut-off low evolution in the BCC-CSM2-HR model and GloSea5 model, respectively. During the initial stage, as shown in Figs. 4d and g, both models can capture the alternative cyclonic and anticyclonic anomalies from the subpolar North Atlantic to Northeast Asia, but the amplitudes of $ Z_{200} $ anomalies, especially the Ural anticyclonic anomalies, are underestimated in the two models. During the rapid growth stage, the negative $ Z_{200} $ anomalies in Northeast Asia are strengthened compared to the initial stage but are still weaker compared to observations. During the decay stage, the negative $ Z_{200} $ anomalies in Northeast Asia move downstream and are also weaker than the observation. The evolution of 500-hPa temperature anomalies (shadings) suggests that the cold temperature anomalies are always associated with negative geopotential height anomalies. The amplitudes of temperature anomalies are underestimated by the two models. Through the above composite analysis, we find that the two models can essentially capture the dynamics responsible for the three-step evolution of early-summer Northeast Asian cut-off lows, but the amplitudes are generally underestimated.

We next try to understand the possible reason for the underestimated intensity of cut-off lows in the three-step evolution. As suggested by Nie et al. (2022), the dynamical processes controlling the evolution of early-summer cut-off lows are strongly modulated by the changes of background flow. In early summer, Northeast Asia is located at the eastern exit of the midlatitude jet and to the north of the subtropical jet. This region is characterized by weak wind speeds, and favors frequent formation of cyclonic anomaly and energy accumulation. Figure 5 shows the difference of climatological 200-hPa zonal wind between the model and reanalysis. It is shown that both models underestimate the strength of the Eurasian midlatitude jet and the East Asian subtropical jet. The weaker midlatitude jet does not favor Rossby wave trains propagating along the jet (Hoskins and Karoly, 1981; Seager et al., 2003; Barnes and Hartmann, 2011; Nakamura and Huang, 2018). Also, the weaker subtropical jet is not favorable for baroclinic eddy generation below the jet (Lee and Kim, 2003). Therefore, the underestimated background flow in the models may lead to less efficient regional eddy-mean flow interaction essential for the cut-off low evolution.

Figure 5.  Errors in modeled wind climatology. (a) Difference of 200-hPa zonal wind climatology between the BCC-CSM2-HR model and ERA5 reanalysis (shading, units: m s⁻¹) during May and June. (b) as in (a) but for the GloSea5 model. The black contours in the two panels denote the 200-hPa zonal wind climatology in the ERA5 reanalysis.

Northeast Asian cut-off lows are important cyclonic systems that can bring cool weather and precipitation extremes over large areas. Skillful subseasonal forecasting of Northeast Asian cut-off lows is of great importance for subseasonal prediction in regional extremes. Using two dynamical forecasting systems (BCC-CSM2-HR and GloSea5), this study assesses the simulation ability and subseasonal prediction skill for early-summer Northeast Asian cut-off lows. It is shown that both models can represent the spatial pattern of Northeast Asian cut-off lows well but with weaker amplitude compared to the observation. The skillful prediction time scales for cut-off low intensity are roughly 10.2 days for the BCC-CSM2-HR model and 11.4 days for the GloSea5 model. We also show that both models can essentially capture the three-stage processes responsible for the cut-off low evolution, namely the initial stage with a Rossby wave train propagating from the subpolar North Atlantic to Northeast Asia, a rapid growth stage with local eddy-mean flow feedback, and a decay stage with energy dispersion. However, the models tend to underestimate the amplitude at each stage, and therefore fail to simulate the large impacts on extreme temperature and precipitation events in Northeast Asia. Since the local eddy-mean flow interaction is often modulated by background flow, we argue that the weaker eddy-mean flow interaction is a result of the underestimation of wind speed of the Euraisan midlatitude jet and East Asian subtropical jet, which both play important roles in providing a propagating path for the Rossby wave train and amplifying the local eddy generation. Future work will be conducted to explore the relation between the background flow bias and cut-off low intensity using more climate models with large ensembles.

Acknowledgements. This work was jointly supported by the National Key Research and Development Program of China (2021YFA0718000), NSF of China under Grant No. 42175075, and the UK-China Research & Innovation Partnership Fund through the Met Office Climate Science for Service Partnership (CSSP) China as part of the Newton Fund.