Uncertainties in Modelling Hawaii's Future Precipitation and What It Means for Endangered Forest Birds: A Review

Abstract

Aim

We aim to review present uncertainties in projecting fine-scale future precipitation in an area of high model disagreement, which is also data poor, topographically complex, and experiences climate-driven threats to endemic biodiversity.

Location

Hawaiian Islands.

Time Period

We primarily focused on downscaling studies from the past decade and studies comparing the most recent iterations of the Coupled Model Intercomparison Project.

Major Taxa Studied

Hawaiian honeycreepers.

Methods

We explored sources of uncertainties in two major categories: (1) downscaling general circulation models (GCMs) to islands and (2) systematic biases in the representation of the tropical Pacific climate. We framed this discussion in the context of management planning for endangered Hawaiian forest birds. We also explored a brief case study exploring the impact of differing precipitation projections on Hawaiian forest bird ranges. This involves the use of maximum entropy software to model suitable habitat for Kiwikiu (Pseudonestor xanthophrys) using baseline climate data and projecting that model to two different dynamically downscaled precipitation projections for Hawaii.

Results

The selection of downscaling methodology can affect as much as the sign of change for precipitation in areas of complex topography, especially forest bird habitat at higher elevations. We identified dynamical downscaling as the most used method for island climate predictions globally. Of statistical downscaling methods, machine learning proved to be the most common in recent island studies. The major sources of persistent uncertainty of GCM simulations in the tropical Pacific are the double Inter-Tropical Convergence Zone bias, the cold tongue bias, and westward-extended El Niño-Southern Oscillation sea surface temperature anomalies. These biases complicate the prediction of winter precipitation and future drought prevalence in Hawaii. The differences in precipitation projections from our case study show a large impact on range estimations of suitable habitat for Kiwikiu, especially on the leeward side of Maui.

Main Conclusions

Despite its limitations, dynamical downscaling may be better suited than statistical downscaling for simulating precipitation in Hawaii. Of statistical downscaling methods, perfect prognosis and machine learning show the most promise in accurate spatial representation of precipitation. Selected GCMs have recently achieved improved representations of the mean state tropical Pacific climate and more realistic El Niño –Southern Oscillation nonlinear feedbacks. To benefit from these improvements, future research could be dedicated to finding which models within the Coupled Model Intercomparison Project have the lowest precipitation bias over the northern central tropical Pacific. Future drought predictions in Hawaii will impact the planning of conservation actions such as predator control, conservation introductions, and novel disease management techniques.