Roles of Climate Variability on the Rapid Increase of Winter Haze 1 Pollution in North China after 2010 2

North China experiences severe haze pollution in early winter, resulting in many premature deaths and considerable 10 economic losses. The number of haze days in early winter in North China (HDNC) increased rapidly after 2010 but declined 11 slowly before 2010, reflecting a trend reversal. Global warming and emissions were two fundamental drivers of the long-term 12 increasing trend of haze, but no studies have focused on this trend reversal. The autumn SST in the Pacific and Atlantic, 13 Eurasian snow cover and central Siberian soil moisture, which exhibited completely opposite trends before and after 2010, 14 were proven to stimulate identical trends of meteorological conditions related to haze pollution in North China. Numerical 15 experiments with a fixed emission level confirmed the physical relationships between the climate drivers and HDNC during 16 both decreasing and increasing periods. These external drivers induced a larger decreasing trend of HDNC than the observations, 17 and combined with the persistently increasing trend of anthropogenic emissions, resulted in a realistic slowly decreasing trend. 18 However, after 2010, the increasing trends driven by these climate divers and human emissions jointly led to a rapid increase 19 in HDNC. 20

Monthly mean meteorological data from 1979 to 2018 were obtained from NCEP/NCAR reanalysis datasets (2.5°×2.5°), 60 including the geopotential height at 500 hPa (H500), vertical wind from the surface to 150 hPa, surface air temperatures (SAT), 61 wind speed, and special humidity at the surface (Kalnay et al., 1996) (Fig. 1b). The simulation results 141 were highly correlated with HD NC and showed the feature that the trend in P2 was stronger than that in P1, indicating that 142 meteorological conditions drove the trend change of haze pollution.

143
As shown in Figure S3a, the preceding autumn SST in the Pacific, associated with the detrended HD NC , presented a

164
These similar trend changes suggest that the North Pacific SST might have been a major driver of the abrupt change in HD NC .

165
It is clear that SST P underwent a significant trend change around 2010 (Fig. 4a). Thus, the persistent decline in SST P during P1 166 (at a significant rate of -0.2 °C/10 yr; Table 1)

202
The area-averaged September-November soil moisture over central Siberia was denoted as the Soilw index, whose correlation 203 coefficients with HD NC were -0.57 and -0.60 from 1979 to 2018 before and after detrending, respectively (above the 99% 204 confidence level). Negative Soilw anomalies could induce a positive phase of EA/WR, and the associated anticyclonic 205 circulations occurred more frequently and more strongly (Fig. S7). Correspondingly, the local vertical and horizontal 206 dispersion conditions were limited. With increasing moisture, pollutants can more easily accumulate in a confined area. The 207 spatial linear trend of soil moisture also shifted from increasing to decreasing in 2010, opposite to the trend of HD NC (Fig. 3e,   208 f). The change rate of Soilw was 38.8 mm/10 yr (opposite that of HD NC ) during P1, and the rate of change became more intense 209 (-51.8 mm/10 yr) during P2, physically driving a similar large change in HD NC (Fig. 4d).  (Table   214 1). Thus, the physical relationships between HD NC and these four factors were long-standing and did not disappear as the trend 215 changed. These four external factors had completely opposite trends in P1 and P2. Excluding SST A , the amplitudes of the 216 change trends of the other three indices in P2 were obviously stronger than those in P1 and were identical to those of HD NC 217 (Table 1) (Table 1). Among the four indices, the 233 correlation coefficients between SST P and Snowc (Pair 1) and between SST A and Soilw (Pair 2) were high, while the rest were 234 insignificant. The variance inflation factors of the four factors in the MLR model were less than 2, showing that the collinearity 235 among them was weak. When selecting one factor from both Pair 1 and Pair 2 to refit HD NC , the correlation coefficient between 236 the fitted and observed HD NC and the trends of the fitted HD NC in P2 worsened (Fig. S9). Therefore, these four external factors   above the 99% confidence level, "**" indicates that the CC was above the 95% confidence level, and "*" indicates that the 423 CC was above the 90% confidence level.
424 Table 2. The contribution rate of fitted HD NC and each external forcing factor to the trend of HD NC in P1 and P2, respectively. above the 99% confidence level, "**" indicates that the CC was above the 95% confidence level, and "*" indicates that the 453 CC was above the 90% confidence level.