Key message

  • East Asia has experienced a resurgence of serious rice planthopper outbreaks in recent years. In one of the worst-affected areas, the Lower Yangtze Valley of China, the number and timing of brown planthopper immigrants from further south has a major influence on the risk of local outbreaks.

  • Here we show how seasonal outbreak risk can be predicted from indices of the intensity of the western Pacific subtropical high-pressure system, a major atmospheric circulation system that drives the synoptic weather patterns affecting planthopper immigration.

  • Better prediction allows time for plant protection agencies and other stakeholders to make essential preparations in high-risk seasons.

Introduction

The brown planthopper (BPH), Nilaparvata lugens (Stål), is the pre-eminent insect pest of rice in Asia, due to devastating feeding damage to the crop (‘hopperburn’) and through the transmission of virus diseases of rice (Bottrell and Schoenly 2012; Cheng 2009, 2015; Otuka 2013; Heong et al. 2015). During heavy outbreaks, BPH can cause almost total crop failure, damaging up to 20 million ha of rice crops annually in China alone (Hu et al. 2011, 2014; Lu et al. 2017). Besides the crop damage itself, there are serious issues associated with insecticide use on planthoppers in China [e.g. high levels of resistance to many insecticidal compounds (Zhang et al. 2014; Wu et al. 2018), disruption of natural enemy control (Cheng 2015) and public health problems (Huang et al. 2015)]. BPH cannot overwinter in temperate regions of China, Korea and Japan; outbreaks in these regions each summer are initiated by a series of five long-range windborne migrations originating in winter-breeding areas in the Indochina Peninsula (Cheng et al. 1979; Kisimoto and Sogawa 1995; Otuka 2013), particularly Central Vietnam (14–19°N; Hu et al.

Methods

Light trap data

Daily planthopper catch data from standardised 20-W ‘black light’ (UV) traps located at the plant protection stations of 222 counties in China were obtained from the National Agro-Tech Extension and Service Centre (NATESC), which has been continuously collecting data since 1977. In this study, data from 8 stations (Fig. S1), which have complete data cover from 1978 to 2003, were used in the correlation analyses. From 2004 onwards, a new light trap design was gradually introduced throughout China (over a period of several years) to replace the ‘traditional’ black light traps. The efficiency of the new traps for catching planthoppers was greatly affected, and during exploratory analyses it proved difficult to take account of changes in the light trap type in the regression models (G. Hu, unpublished analyses). Therefore, we have restricted our analyses to the standardised data set (1978–2003) for the models developed to forecast BPH abundance from WPSH intensity and SSTA indices (see below) and have refrained from including later light trap data in our analyses due to its questionable reliability.

BPH immigration levels and concentration zones

To represent BPH migration activity, light trap catch data from the 222 stations (Fig. S1) from 1 April to 10 August of each year from 1977 to 2003 were extracted, and the summed catch for each 5-day period (Table S1) for each station was calculated. In order to assign any given station as a ‘concentration and landing zone’ (i.e. a station which received a major immigration of BPH), we calculated the 90th percentile value of catches (termed ‘BPH90th’) in each of the 26 5-day periods from 1 April to 10 August. We then produced a time series of 702 of these BPH90th values (the 26 5-day periods × 27 years, Table S1) which we termed ‘TS.BPH’.

As formulated in the classic paper by Cheng et al. (1979), BPH was expected to expand its range each year from overwintering areas in Indochina by up to five cycles of northwards migration. To examine these migration cycles further, the seasonal variation during the years 1977–2003 was decomposed from our time series (TS.BPH) by using linear models. The form of this linear model was:

$$\hbox{Log}10 \, \left( {TS.BPH + 1} \right) \, = \, \beta_{1} V_{P1} + \beta_{2} V_{P2} + \beta_{3} V_{P3} + \;\cdots + \beta_{26} V_{P26}$$
(1)

where VP1, VP2, VP3, …, VP26 are the 26 periods of 5-day BPH catches between 1 April and 10 August (Table S1) and β1, β2, β3, …, β26 are the estimates of the model parameters. Each spell of rapid population growth in this seasonal variation of planthopper numbers was defined as a migration cycle.

To explore the seasonal variation in the position of BPH concentration and landing zones, the latitude time cross section of the relative 2-D binned kernel density of stations in a concentration zone was estimated. A heat map showing intensity of BPH outbreaks by latitude and seasonal time period (see Fig. 2b) was generated from raw count data for 5-day periods from all 222 recording sites, using a 2-dimensional kernel smoother to produce density estimates on a regular grid from the irregularly spaced raw data. An outbreak was flagged if a count exceeded the 90th percentile value for the seasonal time period and year in question. The densities shown on the heat map are the probability of an outbreak at the specified time and latitude, minus the probability of any relevant data being available, to allow for uneven data collection. Only outbreak densities which are greater than data availability densities are shown, in order to filter out sporadic density peaks.

Fig. 2
figure 2

a The upper panel shows the mean latitude of the ridge of the WPSH between 110 and 120°E, and the mean seasonal variation in 5-day BPH catches in eastern China (Table S1). The lower panel shows the changes between neighbouring periods; for example for time period i + 1, the value plotted is equal to the value at period i + 1 minus the value at period i. The five migration ‘steps’ during the planthoppers’ migrations were interspersed by periods of rapid population growth. The ‘two abrupt jumps’ of the WPSH are meteorologically defined (e.g. Ding and Chan 2005; Tao and Wei 2006; Ding et al. 2007) and denote the beginning and end of the Meiyu season in the Yangtze River Valley. b Latitude date cross section of the relative 2-D binned kernel density of the trap** stations (relative density of traps per unit of latitude for each 5-day period) in a BPH concentration zone, based on the data from 222 county plant protection stations between 1977 and 2003. Any given BPH trap** station in any 5-day period was defined as a planthopper ‘concentration and landing zone’ if the number of BPH in the 5-day catches was greater than or equal to the BPH90th (i.e. the 90th percentile value in that period of that year). c Latitude date cross section of 5-day mean precipitation between 110 and 120°E. d Latitude date cross section of 5-day mean winds at 850 hPa height, between 110 and 120°E. (Color figure online)

Full size image

Meteorological data and WPSH indices

The Climate Prediction Center Merged Analysis of Precipitation (CMAP) data, including the monthly and 5-day global-gridded precipitation means since 1979, were obtained from National Oceanic and Atmospheric Administration’s (NOAA) Earth System Research Laboratory (http://www.esrl.noaa.gov/). Their monthly and daily global-gridded data, including the geopotential height and u- and v-winds, were derived from National Centers for Environmental Prediction (NCEP)/National Center for Atmospheric Research (NCAR) reanalysis data from 1948 to 2011. The CMAP and NCEP/NCAR data have a spatial resolution of 2.5°.

Monthly global mean sea surface temperature (SST) since 1854 was obtained from the NOAA Extended Reconstructed SST (version 3b), which has a spatial resolution of 2.0°. Monthly mean North Atlantic Oscillation (NAO) index since January 1950 was obtained from NOAA’s Climate Prediction Center. The sea surface temperature anomaly [SSTA(IO–WNP)] index represents the April–May mean dipolar SSTA difference between the Indian Ocean (IO, 10°S–10°N, 50°E–110°E) and the West North Pacific (WNP, 0°–15°N, 120–160°E). The El Niño-Southern Oscillation development index (ENSOdevelop) denotes the May–March SSTA in the central Pacific (15°S–5°N, 170°W–130°W).

The monthly indexes of WPSH (110°E to 180°E) from 1951 to 2010 were obtained from the China Meteorological Data Sharing Service System (http://cdc.cma.gov.cn/). WPSH is described using five indices: (VA7, VI7, VR7, VN7 and VW7 to represent the area, intensity, mean ridge, northern edge and westward extension of the WPSH, respectively).

The five-day mean ridge of the WPSH was calculated using daily NCEP/NCAR reanalysis data. The region of 5860 gpm (geopotential metres) at 500 hPa was used to describe the WPSH. The boundary between the east wind and west wind was defined as the dynamic parameter used to describe the location of the WPSH ridge (Song et al. 2001) and was calculated by the following equation:

$$u = 0;\quad\frac{\partial u}{\partial y} = 0$$
(2)

where u is the speed of zonal wind and coordinate (x, y) is its location in the 2-D dimension of u-wind grid data. The location of the five-day mean ridge was calculated by the Grid Analysis and Display System (version 2.0.2, http://grads.iges.org/grads/).

Regression models of BPH immigration levels in the Lower Yangtze River Valley

As most BPH migrated into the LYRV in July (see results and Figs. 2b, 4a), light trap catches of BPH during 1978–2003 from five stations in the LYRV, namely Dongzhi, Huizhou, Gaochun, Nantong and Fengxian (Fig. S1), in July were used to assess immigration into this area (i.e. the response variable VJul). The five monthly indices of the WPSH circulation system (i.e. VA7, VI7, VR7, VN7 and VW7; Table S2) were the potential explanatory variables used to build a regression model for exploring the quantitative relationship between WPSH and BPH migration (see Box 2 for list of variables).

Box 2 List of variables used in the regression models
Full size table

Previous studies have established that the WPSH variation is primarily controlled by central Pacific cooling/warming and that there is a positive atmosphere–ocean feedback between the WPSH and the Indo-Pacific warm pool ocean. The WPSH intensity index in summer (June–August) can be predicted based on SSTA(IO–WNP), NAO index and ENSO development index (i.e. VSSTA(IO–WNP), VNAO and VENSO in Table S2) (Wang et al. 2013). These three indices were the potential explanatory variables used to build a forecast model, several months in advance, for planthopper immigration levels in July.

The densities in the source areas have great impact on the immigrant density of the planthoppers in a seasonally invaded area (i.e. the LYRV). Most migrants arriving in the LYRV in July came from northern South China, i.e. northern Guangxi, northern Guangdong, southern Hunan, southern Jiangxi and south-eastern Guizhou (Cheng et al. 1979; Hu et al. 2011). The emigratory adults were the third generation after the initial colonisation of northern South China 2 months earlier, due to their population cycle of macropterous-brachypterous-macropterous forms (Cheng et al. 1979). Light trap data from three stations, i.e. Tianzhu, Quanzhou and Qujiang (see Fig. S1), were used to represent the abundance of emigrants from this source region in May. The log-transformed sum of light trap catches in these three stations formed another potential explanatory variable (VMaylg) (Table S2) in both regression models.

Data exploration was applied following the protocol described by Zuur et al. (2010). The presence of outliers, auto-correlation in the response variables and collinearity were examined, and the type of relationship was also investigated. Because there was much collinearity in these variables (Fig. S3), each potential parameter was tried in turn and was chosen if it had the best Akaike information criterion (Table S3). Negative binomial generalized linear model (GLM) was applied after the initial Poisson GLMs indicated over-dispersion. Auto-correlation and over-dispersion in the Pearson residuals of the fitted model were checked, justifying the use of a negative binomial GLM.

Results

As mentioned above, the seasonal expansion of BPH from its overwintering areas in Indochina can be schematised by five cycles of northwards migration interspersed by rapid population growth (Fig. 2a, Table S1). We are concerned here with the factors influencing the last two migration ‘steps’ in China, extending from mid-June to early-July (fourth ‘step’), and from mid-July to late July (fifth ‘step’) (Fig. 2a). In the fourth step, the concentration and landing zones were located in northern South China (north of the Tropic of Cancer in Guangxi and Guangdong Provinces,  ~  25°N) and the LYRV (Figs. 2b, S2d–h). The number of migrants decreased as the migration distance increased and so BPH catches in the LYRV were much smaller than those in northern South China (Fig. S2d–h). The fifth migration step comprises the main movement into the LYRV (Figs. 2b, S2j-k).

Influence of rainfall and low-level jets on the northward migration

Major zonal rainfall belts with their associated downdrafts, rain and cold temperatures form a barrier to BPH flight and promote concentration and landing (e.g. Crummay and Atkinson 1997; Hu et al. 2007). Just before the fourth migration step, the concentration zone was still in South China (Fig. 2b). In mid-June, during the ‘Meiyu season’, the rainfall belt moves to the LYRV (Fig. 2c) causing the concentration zone to shift towards the north in what constitutes the fourth migration step (Figs. 2b, S2d–h). After the Meiyu season is over (around 10 July: Ding and Chan 2005), the rain belt is situated further north again, over the Jiang-Huai Valley (30 to 34°N, i.e. the region between Yangtze and Huai rivers) for 10 days or more (Ding and Chan 2005) (Fig. 2c). The concentration zone consequently shifts to the LYRV—a movement representing the fifth N. lugens migration step (Figs. 2b, S2j–k). Thus, the seasonal movement of the East Asian rain belt determines the temporal and spatial distribution of immigrants by alternately allowing and impeding migration (Fig. 2b, c).

A spatial map of correlation coefficients between the immigration levels in the LYRV and the July precipitation during 1979–2003 revealed a significant correlation between catches and rainfall in the region immediately to the north (see green-filled region in Fig. 3). The LYRV is located at the southern fringe of the rain belt, which thus forms a natural barrier to migration at this time (Figs. 2c, 3).

Fig. 3
figure 3

Simultaneous correlation map between BPH immigration levels in the Lower Yangtze River Valley and precipitation (green), and low-level jet (LLJ) days (red) in July. The BPH immigration level is defined as the cumulative sum of light trap catches from five plant protection stations (blue triangles) during 1978–2003. LLJ days are days when the 850 hPa south-westerly wind speed was greater than 12 m/s. The light and dark green/red areas indicate significance at 1 and 5% levels, respectively. (Color figure online)

Full size image

The northward migration process itself is facilitated by the development of strong south-westerly winds, particularly the development of nocturnal LLJs which provide rapid aerial transport (Watanabe and Seino 1991; Feng et al. 2002; Qi et al. 2010). Before the fourth migration step, strong south-westerlies were confined to South China but after mid-June the zone of south-westerlies expanded to the north, and the winds strengthened (Fig. 2d) promoting migration to the LYRV. Consistent with this scenario, we found that the number of LLJ days (i.e. those with wind speed ≥ 12 m/s at 850 hPa) in southeast China was significantly correlated with the immigration level in the LYRV in July during 1979–2003 (see red-filled region in Fig. 3).

Association of the WPSH system with the northward migration

Previous studies have shown that the rain belt distribution in eastern Asia is regulated by the WPSH (e.g. Ding and Chan 2005; Tao and Wei 2006; Ding et al. 2007). Based on results from these studies, we hypothesised that movements of the WPSH would influence the development of south-westerly airstreams and the location of rain belts, and through them BPH movement and concentration (see previous section). The WPSH moves northwards in a stepwise fashion each year, and during summer it exhibits two independent and abrupt movements; the first of these occurs in mid-June when its ridge jumps northward abruptly from South China to the Yangtze River basin, heralding the Meiyu season in the latter region (and much further afield in Korea and southern Japan). The second jump usually occurs in late July, when the WPSH shifts to its most northern position (> 30°N), marking the end of the Meiyu season in the Yangtze River valley and the start of the rains in north China. The association of these movements of the WPSH and the position of the rain belt and the development of the south–westerly airstream are shown in Fig. 2a–d.

The northward shift of the BPH concentration zone coincided with the advance of the WPSH (Fig. 2a). The concentration zone was located north of the WPSH ridge close to the 5860 gpm contour at 500 hPa altitude, whereas the WPSH ridge itself was located south of 30°N (Fig. S2a–l). As the WPSH moved northward in June and July, the BPH concentration zone showed a corresponding northward shift (Fig. S2).

We investigated the relationships between the position and intensity of the WPSH and the trap catches arising from the fourth and fifth migration steps. At the time of peak catches of BPH across the 50 stations from the LYRV (Fig. 4a), the average position of the WPSH ridge was 26.53° ± 0.48°N (95% confidence interval, n = 326) (Fig. 4b). The distribution of latitudes where ridges were located during these peak catches was considerably tighter (i.e. more concentrated) than that in the period from June to early August as a whole (mean 25.23° ± 0.52°N, (95% CI, n = 378)) (compare the green and red bars in Fig. 4b), and these two latitude distributions were significantly different (Fligner-Killeen test: χ2 = 18.01, df = 1, p < 0.001). Therefore, the location of the WPSH and, in particular, its two abrupt jumps (see above) were critical for immigration levels in the LYRV.

Fig. 4
figure 4

a Histogram of 5-day catch peaks of BPH recorded at the 50 stations in the Lower Yangtze Valley before mid-August during 1977–2003. b Histogram of the latitude of WPSH ridges when 5-day catch peaks were recorded at these stations (solid green bars), and during the whole period from June to early August (hollow red bars). c Histogram of the latitude of WPSH ridges in years with a strong WPSH (solid green bars) and a weak WPSH (hollow red bars). (Color figure online)

Full size image

To analyse this association further, we distinguished years in which the WPSH intensity in July was classified as ‘strong’ (intensity was ≥ the 3rd quartile value of all WPSH intensities in July 1977–2003), and years when the WPSH intensity in July was classified as ‘weak’ (intensity was ≤ the 1st quartile value) (Table S2). The latitudinal distribution of ridges in strong years (27.20° ± 1.22°N, (95% CI, n = 42)) was similar to the spatial distribution of ridges during the period of BPH peak catches in the LYRV (Fligner-Killeen test: χ2 = 0.004, df = 1, p = 0.95) (compare the green bars in Fig. 4b, c). By contrast, the latitudinal distribution of ridges in weak years (27.19° ± 1.79°N, (95% CI, n = 42) was significantly different to that during peak catches of BPH (Fligner-Killeen test: χ2 = 9.062, df = 1, p = 0.003) (compare the green bars in Fig. 4b with the red bars in Fig. 4c). Latitudinal distributions of ridges in weak and strong years were also significantly different (Fligner-Killeen test: χ2 = 6.434, df = 1, p = 0.011; Fig. 4c). These analyses indicated that WPSH intensity also has significant influence on immigration.

Prediction models based on WPSH-related climatic indices

Having demonstrated the influence of the WPSH and associated weather factors on immigration into the LYRV, we then developed GLMs to provide forecasts of immigration levels in July—specifically to predict the cumulative light catches in July (VJul). The optimal model proved to be:

$${\text{Log }}\left( {V_{\text{Jul}} } \right) \, = \, 7.056 \, + \, 0.023V_{I7} + \, 0.224V_{\text{Maylg}}$$
(3)

where VI7 is a WPSH intensity index and VMaylg represents light catches in source areas in South China in May. The ‘generalised R2’ for this model = 0.34, and the Pearson’s correlation coefficient (r) of July BPH catches and model predictions was 0.60 (df = 24, p = 0.001). The model suggests that an enhanced WPSH indicates an increase in immigration into the LYRV in July and that a model incorporating a measure of WPSH intensity and an estimate of numbers of emigrants in source areas can effectively predict the immigration levels (Fig. 5).

Fig. 5
figure 5

Interannual variations in the immigration levels of BPH in the Lower Yangtze Valley in July (solid black line) in comparison with the catches in South China in May (solid green line) and the WPSH intensity index in July (solid blue line). Also shown is the predicted BPH immigration level in July (red dotted line) based on the WPSH intensity in July and catches in South China in May; uncertainty in the predicted values (the forecast standard error) is shown by grey shading. The Pearson’s correlation coefficient (r) of July BPH catches and model predictions was 0.60 (df = 24, p = 0.001). (Color figure online)

Full size image

We then sought to discover whether a WPSH-related climatic index, such as the ENSO or NAO indices, could provide a longer-range forecast. The most efficient index proved to be the SSTA(IO–WNP) (Fig. 6a). The optimal model was:

Fig. 6
figure 6

a Interannual variations in the immigration levels of BPH in July (solid black line) in comparison with the SSTA(IO–WNP) index for April–May (solid blue line). Also shown is the predicted immigration level in July (red dotted line) based on values of the SSTA(IO–WNP) index for April–May and catches in South China in May (see the solid green line in Fig. 5). The Pearson’s correlation coefficient (r) of July BPH catches and model predictions was 0.59 (df = 24, p = 0.002). b Interannual variations in the BPH immigration levels in the Lower Yangtze Valley in July (solid black line) were significantly correlated to catches late in the season (solid red line) (r = 0.63, df = 24, p < 0.001). (Color figure online)

Full size image
$${\text{Log}}\left( {V_{Jul} } \right) = \, 7.796 \, + \, 1.904V_{{{\text{SSTA(IO{-}WNP}})}} + 0.232V_{\text{Maylg}}$$
(4)

The ‘generalised R2’ for this model = 0.41, and the Pearson’s correlation coefficient (r) of July BPH catches and model predictions was 0.59 (df = 24, p = 0.002). There was no advantage in including the other indices in the model as the NAO and ENSO indices had very low correlations with the July catch (Fig. S4). Thus, the magnitude of immigration into the LYRV could be predicted with a reasonable degree of accuracy several months ahead of the arrival date by using the SSTA index and spring catches of BPH from the source area.

Finally, immigration levels in the LYRV in July were significantly correlated with the level of BPH infestation in the late season, i.e. after mid-August (r = 0.63, p < 0.001) (Fig. 6b). This is interesting because previous studies showed that BPH catches in this region in the late season (late August and early September) are a mixture of locally produced insects and immigrants from further afield (Hu et al. 2014). Nonetheless, the high correlation indicates that high reproduction within this extended area is the main reason for the size of the late infestations and migrations can be considered as internal movement within one population region (see also Hu et al. 2014).

Discussion

Our previous work has demonstrated that BPH populations in the LYRV originate initially with spring migrations from central Vietnam to South China (Hu et al. 2015; Hu et al. 2017; Lu et al. 2017). We therefore recommend that a more flexible approach to pest control, based on forecasting of risk and involving significantly reduced quantities of chemical insecticides, is the way forward.

Authors contribution

GH, JWC and BPZ conceived the research ideas and designed the methodology; XC, WCL, FZ, XWW, FX, MCX and XNC collected the data; GH, MHL, HKW and KSL analysed the data; GH, BPZ, DRR and JWC wrote the manuscript. All authors approve publication.