*3.3. Method of Reducing Experimental Error*

The isotope mixture model is an effective method for specific regional moisture recirculation research, but the inherent uncertainty of the model still needs to be considered. In this study, we focused on the correction of experimental errors in plant xylem moisture.

The water obtained in plant Xylem contained organic pollutants such as methanol and ethanol by the low-temperature vacuum distillation extraction technology, which caused deviations in the measurement of the laser isotope analyzer. This error has led to significant differences in the estimation of the amount of vegetation evapotranspiration. In different studies, some unreasonable calculation results of negative *fTr* value will appear.

In this study, deionized water was mixed with methanol and ethanol (pure chromatographical) of different concentrations. The spectral software (LWIA-Spectral Contamination Identifier v1.0, Los Gatos company) was used to determine the spectral measurement of the pollution degree of methanol (NB), and ethanol (BB), the δD and δ18O spectral pollution correction methods were established [38–40]. The configuration of the concentration of methanol and ethanol solutions during the calibration process is the same as the related experiments by Meng et al. (2012). For the calibration result of methanol, the logarithm of the broadband metric NB metric and Δ*δ*2*H* and Δ*δ*18*O* have significant quadratic curve relationships:

$$
\Delta \delta^2 H = 0.018 \text{ (ln.NB)}^3 + 0.092 \text{ (ln.NB)}^2 + 0.388 \text{ln.NB} + 0.785 \text{ (R}^2 = 0.991, p < 0.0001) \tag{21}
$$

$$
\Delta \delta^{18}O = 0.017 \text{ (In.NB)}^3 + 0.017 \text{ (In.NB)}^2 + 0.545 \text{In.NB} + 1.356 \text{ (R}^2 = 0.998, p < 0.0001) \text{ (22)}
$$

For the calibration results, the broadband metric BB metric has a quadratic curve and linear relationship with Δ*δ*2*H* and Δ*δ*18*O*, respectively:

$$
\Delta \delta^2 H = -85.67 B B + 93.664 \text{ (R}^2 = 0.7447 \text{ }, p = 0.026) \tag{23}
$$

$$
\Delta \delta^{18} O = -21.421 B B^2 + 39.935 B B - 19.089 \text{ (R}^2 = 0.769, p = 0.012) \tag{24}
$$

After correction, the calculation result of negative *fTr* value was eliminated. Of course, there are also other uncertainties in model research, such as driving data sampling and experimental errors, structural errors in the physical mechanism of the model, and parameter errors.
