1. Subfossil chironomid head capsules were sampled from surficial sediments from 86 boreal shield lakes in south-central Ontario, Canada. Lake characteristics ranged from shallow to very deep (> 80 m), ultraoligotrophic to mesotrophic, and with end-of-summer hypolimnetic oxygen conditions ranging from near-saturation to anoxic.
2. Subfossil chironomid assemblages, comprising 44 taxa from 59 lakes, were analysed using multivariate ordination techniques such as redundancy analysis (RDA) and canonical correspondence analysis (CCA). Forward selection in RDA and CCA both showed that measures of oxygen, such as end-of-summer volume-weighted hypolimnetic oxygen concentration (VWHO) and bottom oxygen concentration (botO2), were the strongest explanatory variables for the chironomid data. Maximum depth and major ion chemistry were also important explanatory variables.
3. Oxygen inference models were developed using partial-least-squares regression (PLS), weighted-averaging partial-least-squares regression (WA-PLS), and weighted averaging regression (WA). Models were developed using both the full 44 taxa assemblage (which included littoral taxa) and using only 15 profundal-type taxa.
4. Cross-validated models (jackknifing) using full-assemblage or profundal-only taxa had similar statistical power (similar root mean squared error of prediction, RMSEP). The best models had moderate predictive power, with an r2jack as high as 0.56, and an RMSEP as low as 2.15 mg L–1 for [VWHO], and an r2jack of 0.49 and an RMSEP of 0.24 for log([botO2] + 1).
5. Reconstruction of [VWHO] and [botO2] using a previously published chironomid profile that showed strong lake response to land-clearance and logging suggests that oxygen inference models are reliable and accurate, reflecting the qualitative changes occurring in subfossil assemblages. However, the profundal-only models may be misleading in situations where the ratio of littoral-to-profundal subfossils changes drastically in response to lake disturbance.