A Statistical Assessment of Outlier Removal Methods for Defining Groundwater Background Levels from Historical Data in Coal Mining Regions: A Southern Brazil Case Study

Groundwater quality in coal-bearing regions is often compromised by natural and anthropogenic sulfide mineral oxidation, challenging the application of standard environmental limits. This study’s main objective was to evaluate the effectiveness of different statistical outlier removal methods for establishing geochemically coherent natural background levels (NBLs) from historical databases, using the Transgressive Sand sequence aquifer in the South Santa Catarina Coalfield, Brazil, as a representative case study. Using a historical dataset of 66 analyses from 19 unimpacted wells, we compared three techniques: the interquartile range, the modified Z-score, and the iterative Grubbs’ test. The resulting datasets were used to estimate NBLs for pH, Eh, electrical conductivity, sulfate, Fe, Al, and Mn, and were further assessed using Pearson correlation and principal component analysis (PCA), as combining statistical filtering with hydrochemical validation is essential for ensuring robust interpretation. Results show the Grubbs’ test was the most robust method, preserving natural data heterogeneity and providing the most geochemically consistent dataset. The NBLs estimated from the Grubbs’ dataset revealed naturally acidic conditions (pH 4.4–6.3) and elevated concentrations of Al, Fe, and Mn, frequently exceeding Brazilian drinking water standards. PCA of the Grubbs’ data explained 60% of the total variance, effectively separating variables associated with sulfide oxidation. This study demonstrates that the Grubbs’ test is a superior tool for processing complex historical datasets and highlights that generic water quality legislation may be inadequate for regions with significant mineralization. Thus, this methodological approach can be used to define realistic recovery targets in other coal-bearing regions worldwide.