Introduction
Objective
Identify the most accurate, robust, and operationally useful methodology for predicting appropriate climate habitat for the forests of North America.
A need for decision support
There is mounting evidence that North America's forests are unable to migrate north at the velocity of climate change (Rustard et al., 2012; Fettig et al., 2013; Worrall et al., 2013; Sittaro et al., 2017; Zhu et al., 2012). In central North America there has been complete regeneration failure after wildfire as the boreal forest shifts to the drier prairie ecosystem to the south (Whitman, 2019; Coop et al., 2016; Kolden et al., 2017). Drought induced tree mortality in the boreal forest is reducing carbon sequestration and and is predicted to continue causing a potential climate feedback effect (Liu et al., 2023). Assisted migration (actively moving a southern seed sources northward and to higher elevations) is a key climate mitigation strategy for preserving North America's forests (Vitt et al., 2010; Jump and Penuelas, 2005). However, exactly how far to move seed sources for climate adaptation requires multivariate analysis of biologically important climate variables. Additionally there are limited resources so it is important to also predict what areas are at risk and what areas have the best likely hood of successful climate mitigation. Current known distributions of ecosystems and climate data can be utilized to create climate envelope models; with the aid of climate change projections, envelope models can be used to predict how habitat climates will shift under different climate scenarios as shown in figure 1 bellow (Maltby et al., 1999; Box, 2012). Inversely, envelope models can also be used to predict where future forests will be maladapted, and identify regions that will have novel climates helping decision makers allocate resources (Smith et al., 2022). Envelope models have been criticized for being a poor tool for predicting species presence or absence because they are based purely on climate data (Hewitt et al., 2011). However when planning assisted migration intervention it is already assumed that the species is not present and envelope models are an essential first step.
Figure 1. Envelope model design
Previous envelope models and what is missing
Researchers have created ecosystem level envelope models for British Columbia using a random forest (Wang et al., 2012) and Alberta (Schneider et al., 2009) by calculating climatic distances and disturbance rate. Additionally on the west coast tree species specific envelope models have been developed (Gray & Hamann, 2013). However there are no climatic ecosystem level prediction tools for all of north America. The only tool that predicts climate shifts for all of north America was created by Oregon State (St. Clair et al., 2022). However it provides a map of suitable seed sources with suitability scores at the pixel level. Operationally this tool is hard to use because seeds are collected from seed zones and the location is not mapped out at the pixel level. Additionally these tools only used random forest and climatic distances to provide climate shift predictions, additional machine learning techniques need to be applied. A tool that tests empirical climatic distance calculations and additional machine learning techniques to develop the best accuracy possible for forest shift while providing recommendations at the ecosystems or seed zone level is needed. Lastly the model prediction probability should also be able to indicate the quality of the prediction and site's suitability for assisted migration to provide decision support for all the forests in north America.