Depth distribution of chironomids and an evaluation of site-specific and regional lake-depth inference models: a good model gone bad?

We used 39 surface samples from Marcella Lake, Yukon Territory, to examine the distribution of chironomid head capsules in relation to depth and to develop a site-specific (intralake) inference model for reconstructing past lake levels. Ten of the 34 most-frequently occurring taxa encountered in the surface-sediment samples are significantly related to depth. We then applied the site-specific inference model and a previously developed regional model to samples from deep- and shallow-water cores from Marcella Lake. The inferences were compared to an independent Digerfeldt-type reconstruction of lake level history and to moisture inferences drawn from pollen data. Although the site-specific model was good in having better performance statistics than the regional model, it was bad at producing depth reconstructions because most samples from the long cores lacked suitable analogues in the site-specific training set. None of the chironomid-based reconstructions was a good match to the Digerfeldt-type reconstruction. Inconsistencies remain between the paleohydrological inferences derived from the chironomid depth models, the Digerfeldt-type reconstruction and pollen-inferred reconstructions of past moisture regimes.