Diatom-salinity relationships in 111 lakes from the Interior Plateau of British Columbia,Canada: the development of diatom-based models for paleosalinity reconstructions

Diatoms were identified and enumerated from the surface sediments of 111 lakes, 45 from the Kamloops region and 66 from the Cariboo/Chilcotin region, located on the southern Interior Plateau of British Columbia, Canada. This paper is an extension of another study which investigated the relationship of diatoms to salinity and ionic composition in 65 lakes from the Cariboo/Chilcotin region. The 111 lakes spanned a large gradient in salinity, ranging from fresh through hypersaline (late-summer salinity values ranged from 0.04 to 369 g l^–1), and included both carbonate- and sulphate-dominated lakes with sodium and magnesium as the dominant cations. The Kamloops region had more sulphate-dominated, hypersaline lakes and fewer carbonate-rich lakes than the Cariboo/Chilcotin region. Most lakes had higher salinities in the late-summer compared to the spring.Both salinity and brine-type were important variables that could explain the different diatom assemblages present in the lakes. The majority of diatom taxa had salinity optima in the freshwater to subsaline range (<3 g l^–1), and the taxa displayed a range of both narrow and broad tolerances along the salinity gradient. Weighted-averaging regression and calibration, and maximum likelihood techniques were used to develop salinity inference models from the diatom assemblages based on their relationship to the spring, late-summer and average lakewater salinity measurements. Simple weighted-averaging (WA) models generally produced the same or lower bootstrapped RMSEs of prediction than weighted-averaging with tolerance downweighting (WA(tol)) in the two regional and the combined datasets. Weighted averaging partial least squares (WA-PLS) showed little or no improvement in the predictive abilities of the datasets, as judged by the jackknifed RMSE of prediction. In all cases, the combined dataset of 102 lakes performed better than either of the smaller regional datasets, with relatively little difference between spring, average and late-summer salinity models. The maximum likelihood models gave lower apparent RMSEs of prediction in comparison to other methods; however, independent validation of this technique using methods such as bootstrapping were not undertaken because of the computer intensive nature of such analyses. These diatom-based salinity models are now available for reconstructing salinity and climatic trends from appropriately chosen closed-basin lakes in the Interior region of British Columbia.This is the second in a series of papers published in this issue on the paleolimnology of arid regions. These papers were presented at the Sixth International Palaeolimnology Symposium held 19–21 April, 1993 at the Australian National University, Canberra, Australia. Dr A. R. Chivas served as guest editor for these papers.