| Tianshan mountainous area is located in the alpine region,Northwest China,where solid precipitation plays an important role in total precipitation volume.Snowfall,the major component of solid precipitation,is one of predominant water resources in mountains.As global warming intensifies,precipitation forms are changing correspondingly,with the increasing of rainfall and the decreasing of snowfall or sleet,which may greatly increase the risk of flood disaster in early spring,shortage of water resources in summer and all kinds of secondary disasters.Based on daily mean temperature and precipitation labeled with different types from 27 meteorological stations in Tianshan mountainous area during 1950s-1979,combined with altitude,this paper analyzed the spatial distribution of critical temperature(CT)of precipitation transformation using frequency intersection method and threshold temperature(TT)which separate solid from liquid precipitation using probability guarantee method.The threshold temperature was used to separate the snowfall and rainfall from 1980 to 2015 in Tianshan Mountain area,analyzing the temporal and spatial variations of different precipitation patterns since 1960 s.Furthermore,using meteorological,geographical and topographic variables of 27 sites during the cold seasons from 1980 to 2015 in the Tianshan Mountains,we explored their potential influence to snowfall applying two methods,Multiple Linear Regression(MLR),a conventional method;and Random Forest(RF),a non-parametric and non-linear machine learning algorithm.The primary results are as follows:(1)Different stations had different CT values and TT values in a certain temperature range during 1950s-1979.The range of CT values observed in Tianshan mountainous area was found from 1.57 to 5.70 oC while TT values from 1.36 to 5.48 oC.The difference CT values and TT values is less than 1 oC except for the Kumishi station.Through an in-depth comparative analysis of the spatial distribution of CT and TT values,this study could provide reference for the forecasting of different precipitation forms and selection of parameters in the hydrological modelling in Tianshan mountainous area.(2)Both CT and TT values increased with the rising of altitude significantly.But the altitude was not the only factor which impact the change of them,and other meteorological factors such as atmospheric pressure,atmospheric humidity may impact on it.As a whole,CT and TT values in the north slope of Tianshan mountainous area were less than that in the south slope.(3)It was indicated that the changes in snowfall and rainfall exhibited a significantly upward trend with an increasing by 0.35mm/a and 0.59mm/a(P<0.01)since 1960 s in Tianshan mountainous area,respectively.Spatially,snowfall and rainfall in most of stations showed an increasing trend in general with 55.6% and 51.9% stations statistical significance at the level of 0.05,respectively.These stations that significantly increase are basically located on the north slope,and the change of the south slope is not obvious.(4)The different regression approaches were utilized in this study.Experimental results showed that RF model produced a better performance and simulation power than conventional MLR method.In contrast with MLR model,RF model got a higher R2 value(R2 = 0.74)and d value(d=0.93)and lower bias error(RSR =0.51).It was indicated that MLR model can only explain the variation of partial snowfall,and the relationship between snowfall and predictors tends to be nonlinear.It was showed that RH,temperature and longitude were the first three most important variables affecting snowfall simulation.Moreover,elevation,aspect and latitude were of secondary importance,followed by slope and wind speed in both models.In despite of difference between the results of two models,but the main factors affecting snowfall remain the same. |
PDF Full Text Request