• Boosted regression tree;
  • fire;
  • forest loss;
  • machine-learning methods;
  • Māori;
  • New Zealand;
  • settlement history


Aim  Despite small and transient populations, early Māori transformed large areas of New Zealand's forest landscapes. We sought to isolate the biophysical predictors that explain forest loss in the pre-historic (i.e. pre-European) period in New Zealand.

Location  New Zealand.

Methods  We used resampled boosted regression trees to isolate the key predictors of forest loss from a suite of 19 topographic, climatic, soil-related and archaeological predictors at a 1-km spatial resolution across New Zealand.

Results  The key predictors of fire-driven forest loss during New Zealand's pre-history relate to moisture and elevation gradients, with sites characterized by low moisture levels and gentle slopes being most vulnerable. Proxies for human activity were important in the North Island, where Māori population densities were higher, but not the South Island. The predicted pattern of forest loss and its relationship with biophysical variables suggest that early Māori neither deliberately protected fire-prone regions nor systematically burnt less fire-prone ones.

Main conclusions  Before Māori settlement of New Zealand fire was naturally rare, despite biophysical conditions being conducive to fire spread. The introduction of an ignition source by humans made widespread forest loss inevitable, even in the absence of sustained and deliberate use of fire. Rapid forest loss at the time of human settlement is recurrent across eastern Polynesia, so understanding this dynamic in New Zealand has implications for the region as a whole.