Geoscience Australia’s Northern Australia Hydrogeochemistry Survey (NAHS) has been collecting groundwater samples across the north-eastern Northern Territory, investigating water-rock interaction to identify regional mineral prospectivity and establish geochemical baselines. Given the sensitivity of groundwater composition to a range of confounding variables, the program adopted an approach to ensure it minimised (and captured) as much uncertainty as possible in chemical results from the sampling, processing and analysis aspects of a groundwater survey through robust Quality Assurance and Quality Control (QA/QC) protocols. This presentation will share a systematic approach to adopt for future sampling campaigns, useful scripted methods for quickly visualising QA/QC data to make judgements on the quality, and examples from the NAHS of major problems caught through our QA/QC process.
QA/QC begins before the survey commences, with a plan (and budget) for additional samples that need to be collected. We follow a triplicate sampling approach with field and lab duplicates every 10 sites. Our field duplicates capture the errors introduced through the sampling process and field heterogeneity, while our lab duplicates capture variance in the laboratory analysis. Additionally water and filter blanks are collected on every sampling trip to measure any systematic contamination resulting from sampling, storage, transport and processing. For non-isotope systems, standards are included to assess accuracy of response as well as track batch effects. Overlapping samples are used to check for consistent performance when a new laboratory or method is trialled. Consideration is given both to how these blind QA/QC samples perform, as well as holistically whether the batch chemical results make sense using both charge balance and element ratios.
Using this range of QA/QC samples and semi-automated scripts, this project has been able to quickly calculate statistics and visualise performance of each new analysis batch. Worryingly, in several instances lab duplicates were found to have much poorer agreement than field duplicates. As a result, instrument-specific problems, changes in an instrument/calibration within and between batches, sample number mix-up, dilution errors, and systematic offsets attributed to instrument software errors were caught. By identifying these problems at an early stage, which is only possible with independent and blind QA/QC samples, an opportunity to work with the laboratory to deduce and resolve issues quickly was afforded. As a result greater confidence both in the true uncertainty in our datasets, and that interpretations are being made from a validated view of the groundwater system, exists.