Improved methods for signal processing in measurements of mercury by Tekran^® 2537A and 2537B instruments

Abstract. Atmospheric Hg measurements are commonly carried out using Tekran^® Instruments Corporation's model 2537 Hg vapor analyzers, which employ gold amalgamation preconcentration sampling and detection by thermal desorption (TD) and atomic fluorescence spectrometry (AFS). A generally overlooked and poorly characterized source of analytical uncertainty in those measurements is the method by which the raw Hg atomic fluorescence (AF) signal is processed. Here I describe new software-based methods for processing the raw signal from the Tekran^® 2537 instruments, and I evaluate the performances of those methods together with the standard Tekran^® internal signal processing method. For test datasets from two Tekran^® instruments (one 2537A and one 2537B), I estimate that signal processing uncertainties in Hg loadings determined with the Tekran^® method are within ±[1 % +  1.2 pg] and ±[6 % + 0.21 pg], respectively. I demonstrate that the Tekran^® method can produce significant low biases (≥  5 %) not only at low Hg sample loadings (<  5 pg) but also at tropospheric background concentrations of gaseous elemental mercury (GEM) and total mercury (THg) (∼  1 to 2 ng m⁻³) under typical operating conditions (sample loadings of 5–10 pg). Signal processing uncertainties associated with the Tekran^® method can therefore represent a significant unaccounted for addition to the overall  ∼  10 to 15 % uncertainty previously estimated for Tekran^®-based GEM and THg measurements. Signal processing bias can also add significantly to uncertainties in Tekran^®-based gaseous oxidized mercury (GOM) and particle-bound mercury (PBM) measurements, which often derive from Hg sample loadings < 5 pg. In comparison, estimated signal processing uncertainties associated with the new methods described herein are low, ranging from within ±0.053 pg, when the Hg thermal desorption peaks are defined manually, to within ±[2 % + 0.080 pg] when peak definition is automated. Mercury limits of detection (LODs) decrease by 31 to 88 % when the new methods are used in place of the Tekran^® method. I recommend that signal processing uncertainties be quantified in future applications of the Tekran^® 2537 instruments.