Multivariate environmental time series including soil surface CO2 flux (FCO2) have non-stationarity and mutual interdependence, and thus the i.i.d assumption-based conventional regression techniques inevitably lead to spurious regression or lose the dynamic characteristics in the process of variable transformation. In this paper, we adopted a wavelet threshold technique for our newly developed wavelet-based multiresolution state-space model (MRSSM) to overcome such limitations and to quantitatively evaluate the environmental drivers (EDs) controlling the baseline of FCO2. First, the structural characteristics and the potential EDs (PEDs) of FCO2 were explored by wavelet denoised (threshold) SSM for complex environmental observation data. Then, the major EDs (MEDs) were identified using the scale localized correlation and the wavelet coherence analysis between PEDs and observation data. Next, the contribution of MEDs to FCO2 was quantitatively evaluated by calculating the effective dynamic efficiency using the wavelet energy ratio of the maximum-correlation time-frequency bands. Finally, the effectiveness of the wavelet threshold method for MRSSM was discussed. The proposed wavelet denoising method is expected to improve the performance of MRSSM which is effective to identify, evaluate and predict the main environmental factors inherent in the observation data from complex environmental systems where physicochemical and biological processes of various spatio-temporal scales occur simultaneously.