Assessing the impacts of multiple sources on statistical downscaling is challenged by uncertainty from global climate model (GCM), scenario and factor. In our study, by integrating stepwise cluster analysis (SCA), wavelet-based multiscale entropy (WME), and multi-level factorial analysis (MFA); a SCA-WME-MFA is developed to quantitatively analyze the diverse uncertainty (i.e., numerical fluctuation, and the complexity of the modes) of daily mean temperatures (Tmean) for Amu Darya River Basin (ADRB). The major results reveal that: (i) the most remarkable warming rate would be obtained (0.056 ± 0.015 ◦C/year) under SSP5-8.5; (ii) Compared to the base period (1979–2005), Tmean under SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5 would increase by 1.06 ± 1.26 ◦C,1.38 ± 1.39 ◦C, 1.741 ± 1.255 ◦C, and 2.05 ± 1.22 ◦C in the future (2022-2097); (iii) the secular mode of temperature projections is complex (WME values = 0.81 ± 0.15), while the short-term mode is relatively single (WME values = 0.14 ± 0.13); (iv), the uncertainty of temperature projections would increase under the resource and energy intensive development scenario SSP5-8.5; (v) the annual scales features of temperature projections has a marked impact on the relationships between Tmean and factors, and they can be identified by SCA model; (vi) air temperature at 850 hPa has dominant effect on the numerical fluctuation, and the interactions of geopotential height at 500 hPa on other factors have significant effects on downscaling processes; (vii) the ensemble downscaling based on multi-GCM datasets can reduce the diverse uncertainty of temperature projections.