A comparison of the performance of regression models of Amblyomma americanum (L.) (Ixodidae) using life cycle or landscape data from administrative divisions

Abstract

Listen

Translate article

The distribution of Amblyomma americanum (L.) in the continental United States has been modelled using the reported distribution in counties as “established” (six or more ticks or two or more life stages were recorded during a specified time period), “reported” (fewer than six ticks of a single life stage were recorded), or “absent” (neither of the above categories were met) and a set of environmental and biotic explanatory variables. Categorical (vegetation, climate, and habitat suitability for the main host of the tick) or continuous variables (raw data on temperature, vegetation, and habitat fragmentation), as well as the processes of the tick's life cycle, were tested to build models using multiple logistic regression. The best results were obtained when the life cycle processes were used as descriptors of regressions. Better models derived from life cycle processes were obtained after inclusion of habitat suitability for the white tailed deer (the main host of the tick) and landscape fragmentation. Using this approach, 86% of “absent” and 83% of “established” counties were classified correctly, but all “reported” counties were erroneously classified. Modelling life cycle processes with descriptions of host abundance and habitat fragmentation produced the best outcome when coordinates were missing. When only standard categorical descriptors of climate or vegetation were included in models, results produced poor outcomes. Models were improved with remotely sensed information on temperature and vegetation but produced high rates of misclassification for 14% of “absent”, 37% of “established”, and 100% of “reported” counties. Modelling produced poor results in the absence of point distributions (coordinates). Therefore, “reported” counties cannot be handled adequately by modelling procedures, probably because this category does not reflect the true status of the tick distribution. We conclude that standard categorical classifications of the distribution of an organism can not be reliably used as input for modelling procedures.