运用再分析资料（NCEP/NCAR）分析了新疆区域高空年均气温变化特征,同时运用逐步回归法对新疆高空冷月（1月）、暖月（7月）实测气温数据进行缺测插补,并通过交叉验证法对插补数据进行精度检验,在此基础上分析了新疆区域冷月、暖月高空气温变化特征,运用插补重建后的高空实测数据对1月、7月再分析资料（NCEP/NCAR）进行精度验证。结果表明：（1） 新疆区域高空缺测插补后的数据精度较高效果较好,能够反映新疆高空气温变化的客观事实。（2） 通过对新疆区域对流层（低、中、上）、平流层下层的冷、暖月和年平均变化趋势分析来看,随着高度的上升气温变化趋势由升温转为降温,并且随着高度的上升升温率在减小,减温率在增大;气温越高的月份,随着高度的升高,气温变化趋势发生改变的高度就越高;2000年后,对流层各层多为偏暖年,平流层下层多为偏冷年。（3） 850 hPa和700 hPa年均温转折点均发生在1996年,由冷转暖;100 hPa发生在1995—1997年,由暖转冷;500 hPa和300 hPa均未出现明显的突变年份。（4） 1月、7月再分析资料（NCEP/NCAR）与实测数据相关系数大多在0.9以上,总体误差相对较小。

Here, we analyzed the mean annual variations in upper-air temperature features in the Xinjiang region with the NCEP/NCAR reanalysis dataset. We used stepwise regression to interpolate the missing values in the measured data and verified them using a cross-validation method. We analyzed the change in upper-air temperature both in the coldest (January) and the warmest (July) months. Additionally, we verified the NCEP/NCAR data’s accuracy in both January and July with the upper-air measured data. The results indicated that the measured data can describe the upper-air temperature in Xinjiang more accurately after interpolation. We examined the average temperature trends in the troposphere (at the bottom, middle, and upper layers) and the lower stratosphere on both the coldest and warmest months and throughout the year. As the height rises, the temperature trend changes from warming to cooling temperatures. The critical height at which the temperature trend reverses is higher during the warmer months. In most years after 2000, the air temperature exhibited a warm anomaly in the troposphere a cold anomaly in the lower stratosphere. The turning point in the annual mean temperature anomaly, which turned from cold to warm, occurred in 1996 at both 850 hPa and 700 hPa levels. However, between 1995 and 1997, it happened at the 100 hPa level, showing a warm to cold trend. The temperature anomaly did not change abruptly between 500 hPa and 300 hPa. The NCEP/NCAR data and measured data in January and July had a correlation coefficient mostly above 0.9, indicating a relatively small error.