Abstract: In this paper,probability forecast equations of 72 hour's dust forecast in two grades(general dust and strong dust) in spring in Ningxia Province,China were established using probability regression method based on T639 model products from February to May in 2010-2013 and the observation data of sand in 25 stations in Ningxia in the same time period. The forecast result from February to May in 2014 was evaluated using B score,B_s score and B_ias method. The evaluation results show:in dust prediction,the principal concerns are potential vorticity in the middle and lower troposphere,frontogenetical function and water vapor flux in the horizontal direction for general dust,as well as vorticity in 150 hPa in the tropopause and humidity factor for strong dust storm,this illustrates that strong dust weather forecast in Ningxia should consider overall the weather system in high and low levels and cooperation of dynamic heat and water vapor factors;the B_s values are positive for both general dust and strong dust,whose values are 0.19-0.76 and 0.001-0.53,respectively,indicating this method has prediction ability on sand of classification in spring in Ningxia,and the prediction effect decreases with the extension of forecast period;in general,the greater the climatic probability,the larger the region and the stronger the dust storm process is,the more obvious the forecast effect is. With the forecast time approaching,model adjustment, the forecast ability also improved significantly,the method in this paper can basically forecast the 6 large-scale sand and dust weather processes in Ningxia,and can forecast 24 hours in advance for strong dust weather in a single station;so the probability classification forecast have actively guidance and practical application values on dust intensive forecast in Ningxia,and can be extended to grading precipitation,fog and haze and other small probability weather forecast;but because sand dust climate probability is small in Ningxia,especially the strong dust,the missing probability is high,and forecast probability error is always less than what it is. Therefore, it is very necessary to obtain a much longer history samples to improve prediction equation,strengthen the analysis and selection of forecast factors that have clear physical meaning. By combining with the subjective experience of the forecaster,the improvement of numerical weather forecast models and the statistical methods,the prediction ability of this method will be further enhanced.

Key words: probability event regression analysis, graded sand, forecast test