GEOGRAPHICAL RESEARCH ›› 2015, Vol. 34 ›› Issue (9): 1675-1684.doi: 10.11821/dlyj201509006

• Orginal Article • Previous Articles     Next Articles

Analysis of spatial stationary characteristics of air temperature data in different time scales, seasons and its influence on interpolation performance

Chunxia GUO1,2(), Yunqiang ZHU1,3(), Wei SUN4   

  1. 1. Institute of Geographic Sciences and Natural Resources Research, CAS, Beijing 100101, China
    2. University of Chinese Academy of Sciences, Beijing 100049, China
    3. Jiangsu Center for Collaborative Innovation in Geographical Information Resource Development and Application, Nanjing 210023, China
    4. Agricultural Information Institute of CAAS, Beijing 100081, China
  • Received:2015-04-13 Revised:2015-07-08 Online:2015-09-15 Published:2015-09-15


Spatial stationary is a hypothesis for most geo-statistical processes. In order to explore the influence of spatial non-stationary on air temperature interpolation performance, a set of air temperature data in different time scales (including daily average air temperature, monthly average air temperature of January and July and annual average air temperature in 2010) are used. First of all, stepwise regression analysis is adopted to select the most important regression parameters for each data set of temperature. Then trend line analysis is used to estimate whether the air temperature data meets the assumption of spatial stationary. Finally, ten-cross validation is carried out by using the interpolation methods of ordinary linear regression, ordinary kriging, and regression kriging respectively. According to the results, the conclusion can be summarized as follows: 1) Daily average, monthly average air temperature of January, and annual average air temperature data present spatial non-stationary characteristic and an obvious change trend in the north-south direction; while daily average and monthly average air temperature of July are spatial stationary. Interpolation accuracy of daily average air temperature in July, which is spatial stationary, is significantly higher than that in January, which is spatial non-stationary. 2) In general, all the three interpolation methods obtain the best prediction results on annual dataset, then monthly datasets, worst on daily datasets, because the spatial structure of the daily air temperature dataset is more non-stationary than that of the monthly and annual temperature datasets. In terms of time series, the interpolation error reduces with the decrease of the degree of reduction. 3) Regression kriging achieves higher interpolation accuracy on each dataset in general than ordinary kriging, furthermore the improvement of interpolation accuracy achieved by regression kriging is more obvious on non-stationary datasets than on stationary datasets. 4) Distribution of air temperature interpolated by various techniques presents significantly difference in daily time scale, but in monthly and annual scales, there is no significant difference. Values in long time series, which are the means of values in short time series, weaken the occurrence probability of extreme values. Thus the distribution ranges of air temperature in January and July decrease, compared with daily average air temperature in corresponding seasons.

Key words: air temperature, spatial stationary, time scales, season, spatial interpolation