Forest reconstruction and past climatic estimates for a deforested region of south-eastern New Zealand

Predictions of species biomass from a forest simulation model were compared with pollen percentages for seven peatland sites in an area of Southland-Otago, New Zealand, now depleted of forest cover. Comparisons were made for the recent past (700–800 cal. yr BP) and for a period of the early Holocene (7000–8000 cal. yr BP). Satisfactory matches were obtained overall between predicted biomass and pollen for the recent dataset (r=0.73, P<0.001), in spite of expected poor correspondences for some pollen taxa known to be under-represented in the modern pollen rain. Nothofagus species tended to be over-represented by the simulation model, due most likely to dispersal limiting to their spread under actual conditions. Raising mean annual temperatures by 1 °C and lowering precipitation by up to 60% for the forest simulation produced a satisfactory match to the early Holocene site data (r=0.69, P<0.001). To test for consistency between recent and past periods, regressions for each period of modelled relative biomass against pollen percentages were compared, using all tree taxa from all sites. No discernible bias was found between the different climate regimes modelled. However, an examination of each site showed the dominant early-Holocene hardwood forests of Stewart Island were not reproduced by a simulation under the hypothesized past climate. These forests required a different set of conditions from those for the South Island sites, suggesting they grew under a different climatic regime. The low variation in climate among several of the sites tested the forest model's ability to reproduce the distinct forest communities identified from the pollen data. Comparisons with the pollen record improved confidence in the species attribute data used by the model, the completeness of the ecosystem processes explicitly modelled, and the disturbance regimes employed. A forest reconstruction of the region, under current climate conditions, indicated extensive areas of grassland and grassland-scrub vegetation could potentially be replaced by a range of podocarp, broadleaf, and beech forest types. Overall, the exercise suggested such approaches can improve our understanding of the processes required to restore forest in depleted landscapes and to model forest dynamics under changed climates.