Exploring the variance in reference evapotranspiration (ET 0 ) and its dominant influencing factors is important for climate change, hydrological cycles and water management. Temperature (T), wind speed (U 2 ), net radiation (R n ) and actual vapour pressure (e a ) are the major climatic input variables in Penman-Monteith equation. Previous studies have successfully applied the total differential method to determine the relative contributions of changes in these variables to variation in ET 0 on different timescales. However, the interaction of climatic variables has not been incorporated into this method. Taking the inland river basin of Northwest China as the study area, we extended the total differential method to decompose the ET 0 variance into temporal variance and covariance of T, U 2 , R n and e a on intra-annual and annual scales during 1960-2017. The results indicated that the variance in ET 0 on the intra-annual scale was larger than that on the annual scale. Among the four single climatic variables, U 2 variance made larger contributions to intra-annual and annual ET 0 variance with relative contributions of 6.8% and 1.1%, respectively. However, the interaction of climatic variables played a dominant role in the variation in ET 0 . Specifically, on intra-annual scale, coupled U 2 and R n primarily controlled the ET 0 variance (76.1%), followed by coupled R n and e a (9.6%); but their positive effects were weakened by the negative effects of coupled T and e a (À2.3%). On annual scale, coupled U 2 and R n still governed the variance in ET 0 (97.3%), but the effects were also weakened by other groups of interaction effects. Furthermore, ET 0 was highly related to NDVI on an intra-annual scale (R 2 = 0.68, p = 0.00), indicating the strong effect of seasonal vegetation dynamics on ET 0 variance. This study provides new insights to quantitatively assess the interaction effects of environmental factors on key hydrometeorological variables.