Daily and monthly weather improve near-term population forecasts for the vector of Lyme disease (Ixodes scapularis) and it's primary host (Peromyscus leucopus)

Lyme disease is the most prevalent tick-borne disease in North America, and the number of people infected with Lyme disease each year has roughly doubled over the last decade, mostly attributed to an increase in tick populations over the same period. Indeed, previous studies show a strong correlation between tick abundance and human Lyme disease incidence, and that small-mammal host abundance can predict tick abundance. Therefore, future estimates of tick and small-mammal abundance could inform the public about the potential risk of encountering a tick and subsequently contracting the disease. To this end, forecasts for the vector of Lyme disease (black-legged ticks, Ixodes scapularis ), and their principle reservoir host (white-footed mice, Peromyscus leucopus ) were made for the summer of 2020 at the Cary Institute of Ecosystem Studies. Both forecasts were trained on a long-term dataset from the Cary Institute in a Bayesian state-space framework. The tick forecasts were made using a matrix model approach incorporating daily and monthly drivers, as well as mouse abundance. The mouse forecast stems from a Jolly-Seber model, also including daily drivers. Forecasts for both populations were made everyday for 16-days into the future. New tick observations were assimilated with a Bayesian forward filter, whereas new mouse observations were assimilated using a particle filter. As forecasts were probabilistic, they were compared against each other using the continuous ranked probability score and other strictly proper scoring rules. Forecasts that included daily weather covariates, such as relative humidity and vapor pressure deficit, to drive demographic rates improved forecast accuracy and reduced forecast uncertainty. Furthermore, daily minimum temperature improved detection probability of ticks. Forecasts using weather covariates outperformed the null (random walk) and static (no environmental driver(s)) models for both populations. For tick forecasts, models that included a month effect generally outperformed those without month effects. Lastly, the mouse forecasts outperformed tick forecasts. This work highlights the ability to forecast future populations of two very different species, and that including environmental covariates increases forecast accuracy and reduces overall uncertainty.