Effects of dry vegetation coverage estimated from the 2 MODIS Soil Tillage Index on dust occurrence: 3 Verification by surface synoptic observations

In drylands, the dry vegetation coverage affects dust occurrence by modulating 33 threshold friction velocity (or wind speed) for dust emission. However, there has been little research into quantifying the effect of dry vegetation coverage on dust occurrence. This study investigated spatial and temporal variations of dust occurrence and three definitions of strong wind frequency over the Gobi Desert and surrounding regions in March and 37 April, months when dust occurrence is frequent, during 2001–2021. We evaluated the 38 effects of variations in dry vegetation on dust occurrence by using the threat scores of 39 forecasted dust occurrences for each strong wind definition. Our results indicate that dry 40 vegetation, which was derived from the MODIS Soil Tillage Index, affects dust occurrence 41 more remarkably in April than in March. In March, land surface parameters such as soil 42 freeze-thaw and snow cover, in addition to dry vegetation coverage, should be considered 43 to explain dust variations in that month. However, use of the threshold wind speed 44 estimated from dry vegetation coverage improved the prediction accuracy of dust 45 occurrence in April. Therefore, we propose that the dry vegetation coverage is a key factor 46 controlling dust occurrence variations in April. The findings imply that estimation of dry 47 vegetation coverage should be applied to dust models.

Erodibility is influenced by various parameters, including vegetation coverage, snow cover, 68 and soil moisture. The other factor on which dust occurrence depends is erosivity (i.e., ability 69 of wind to cause erosion represented by wind speed), and the relation between erosivity and 70 dust occurrence has been widely studied (e.g., Kurosaki and Mikami, 2003). Although strong 71 3 winds have been reported to significantly affect dust occurrence in desert regions (Kim and 72 Kai, 2007), many studies have also demonstrated that changes in erodibility, rather than 73 erosivity, control variations in dust occurrence (e.g., Kurosaki et al., 2011;Liu et al., 2020). 74 As one of the erodibility factors for dust occurrence, the presence of vegetation affects the 75 threshold friction velocity, which is defined as the minimum friction velocity required for dust  The threshold wind speed is the minimum wind speed required for the initiation of sand 132 saltation and dust occurrence. We used three definitions of threshold wind speed in this 133 study. In the first definition, a constant threshold wind speed value of 6.5 m s -1 ( 6.5 ) is  we cannot obtain friction velocity from synoptic observations. Therefore, we assumed that

159
(1) roughness length 0 is temporally constant at each observatory and (2) dust emissions where ( , ) 5% is the 5th percentile of ( , ). We can expect temporal variation of 183 one of the two roughness elements, stone cover, to be almost none; therefore, its effect can friction velocity on a rough land surface to that in the absence of roughness elements (Eq.

192
(6)), as proposed by Raupach et al. (1993): where (STI) 5% is the 5th percentile of (STI), which is obtained from the 5th percentile where FO is the number of predicted dust occurrence events, when wind speeds exceeded 236 the threshold wind speed, that were verified by observations; FX is the number of predicted 237 dust occurrence events that were not observed; and XO is the number of dust occurrence 238 events that were observed but not predicted. We defined TS in three ways (TS 6.5 , TS 5% , 239 and TS (STI) ), which were calculated from the number of predicted dust occurrence events 240 obtained by using the three definitions of threshold wind speed. The average DOF over the study region was also higher in April (3.39±1.95%) than in 252 March (2.61±1.51%) (Fig. 2c-d), consistent with previous studies (e.g., Kurosaki  March or April, SWF 6.5 ranged from 10% to about 40%, but SWF 5% and SWF (STI) 264 values were always lower than 20% (Fig. 2c-d). SWF 6.5 was obviously larger than both 265 SWF 5% and SWF (STI) ; therefore, both 5% and (STI) were higher than 6.5 m s -1 . Correlations of DOF with SWF 6.5 ( COR 6.5 ), SWF 5% ( COR 5% ), and SWF (STI)

276
(COR (STI) ) in April were generally stronger than those in March (Fig. 3) March. This is probably because other factors such as snow cover and soil freeze-thaw 282 process also play important roles in affecting the threshold wind speed for dust occurrence 283 in March (e.g., Kurosaki and Mikami, 2004;Kong et al., 2021). In April, however, COR (STI)

284
(R 2 = 0.88) was stronger than COR 5% (R 2 = 0.78). Moreover, in April, the intercept of 285 COR (STI) approached the theoretical value of zero compared with that of COR 5% . The 286 reason for the more theoretical correlation between DOF and SWF (STI) was that effects of 287 the interannual variation of dry vegetation on threshold wind speed were taken into account.

288
The SWF (STI) was more reliable to explain the variation of DOF. This result suggests that 289 the dry vegetation effectively influences DOF in April. 290 We explored the threat scores at station scale to evaluate the effects of dry vegetation on 291 the interannual variations of DOF. At most stations in March, TS 6.5 values were lower than 292 0.2 and TS 5% and TS (STI) values were generally higher than 0.2 (Fig. 4a-c). Use of the 293 spatiotemporally constant 6.5 resulted in dust occurrence predictions with low accuracy.

294
Threat scores increased from TS 6.5 to TS 5% throughout the study region except at two 295 stations (Tsogt-Ovoo in Mongolia and Ejin-Qi in Inner Mongolia) (Fig. 4d). However, 296 increases from TS 5% to TS (STI) were smaller at 11 observatories, and decreases were 297 found at 10 stations (Fig. 4e). This result indicates that in addition to the dry vegetation