摘要： The analytical characteristics of D/H isotopes with a fs-laser filament are investigated via analyzing a set of D-enriched water samples with D concentrations ranging from 0.5 to 20%. The filament emission spectra feature a narrow peak width and near-zero continuum spectral component. The characteristics of Balmer lines (a, b and g) are evaluated, and the Balmer-a line is selected for isotope analysis. Isotopic information is extracted from filament emission spectra through four different approaches: spectral deconvolution least squares algorithm (SDA), partial least squares regression-internal validation (PLSR-IV), partial least squares regression-cross validation (PLSR-CV) and partial least squares regression-calibration free (PLSR-CF). A multivariate spectral fitting procedure is established in the SDA. Fine structure components (FSCs) of Ha and Da were integrated in the SDA, and it shows improved analytical performance compared to the conventional SDA which is carried out by fitting the experimental spectra with two Lorentzian or Voigt functions. It is also found that the SDA with FSCs gives more accurate results than PLSR-IV and PLSR-CV. Furthermore, the analytical performance is significantly improved by the use of PLSR-CF, in which the PLSR calibration matrix is constructed with a synthetic spectra set. The improvement of accuracy for the given sample set further allows a calibration curve exhibiting an R2 exceeding 0.998 and a slop of 1.009. In addition, the calibration procedure with isotopically enriched standard samples is not necessary in PLSR-CF, demonstrating its flexibility over classical chemometric approaches.