Spread of SARS-CoV-2 Coronavirus likely to be constrained by climate

Miguel B. Araujo; Babak Naimi

Decomposition of sources of variation We used a 3-way ANOVA to decompose variation in model projections from the 10 different modeling techniques used and 20 replications of the initial conditions (data on SARS-CoV-2 cases). We compared variation arising from data and models with variation associated with projected seasonal changes in climate suitability. The latter is the desired projection, rather than a methodological uncertainty. However, comparison of seasonal projected variation in climate suitability with variation arising from different partitions of data and model classes provides a benchmark against which to compare model variability. If data and model variability was greater than projected seasonal variability, then reliance on the models could be questioned. The analysis involved running a three-way Analysis of Variance (ANOVA) without replication(58, 59) for each cell, using climate suitability for SARS-CoV-2 as response variable and ecological niche models (ENM), bootstrapped samples (BS), and months (M) as factors. We then obtained the sum of squares to each of these sources and their interaction (ENM x BS, ENM x M, BS x M, ENM x BS x M). We estimated the variance components as the proportions of the sums of squares for the three sources of variation (and their interaction) with regards to the total sum of squares (20) . Analyses were performed for each cell in the world grid and we mapped each variance component separately (interaction term was not significant, hence not reported). Monthly variation in climate suitability is the expected outcome of the models. Variability associated with ENM and BS is expected to represent the variability associated with changing initial conditions (BS) and model classes (ENM). They are interpreted as uncertainty estimates and should ideally be much proportionally smaller than monthly variation.