|dc.description.abstract||Western honey bees (Apis mellifera) have experienced high rates of colony loss over the
past decade. Honey bees are critical to agricultural production, providing 15 billion to the United
States (U.S.) economy each year via their pollination services. Due to their importance, studies
have attempted to determine what factors are behind these observed losses. However, relatively
few studies have considered the effects of weather in their analyses despite weather being a key
driver in ecological systems. Those studies that have considered the effects of weather have not
utilized spatial analysis despite studies in other fields finding that issues such as non-stationarity,
where relationships between variables differ across space, and the modifiable areal unit problem
(MAUP) may affect analysis results.
The objectives of Chapter 2 were to determine which weather variables best predict winter
colony loss rates at a national scale in the U.S. and then compare the results from a traditional non-linear approach, using a generalized linear regression (GLR), to results from a spatial approach,
using a geographically weighted regression (GWR) which takes into consideration data non-stationarity, thus helping to elucidate how changes may occur across space. The best supported
variables at the national scale were mean maximum temperature during the month of November,
mean precipitation during the month of February, mean windspeed, and mean elevation. The GWR
had an AIC score that was 328.80 points lower than the GLR and had an R-squared value of 20%
versus 13.5% for the GLR. Thus, these results show that a spatial approach is more statistically
robust than the traditional GLR and indicate that the effects of weather on colony loss are non stationary.
The objective of Chapter 3 was to determine how MAUP would change model results when
a linear regression was run at six different scales - zip code, county, state, and level one, two, and
three ecoregions. MAUP is the phenomenon of different data aggregation methods resulting in
different results. Results indicated that the effects of the two variables analyzed – mean
temperature and mean precipitation – differed substantially between the various scales of analysis.
Additionally, the R-squared values changed drastically, with of low of 8% at the zip code level
and a high of 73% for level two ecoregions. These results are consistent with findings from other
fields and indicate that MAUP should be considered when analyzing an aggregated honey bee
colony loss dataset.
The results of this thesis show that spatial phenomena such as non-stationarity and MAUP
can alter the results of honey bee colony loss analyses, sometimes to a great degree. This highlights
the need for more localized management strategies, as the effects of weather, and potentially other
variables, on colony loss vary by location. Additionally, this thesis shows the need for more
research involving spatial analysis in this field. Future studies may seek to analyze other variables,
such as varroa mite levels or pesticide presence, alongside weather variables to create a fuller
model that may further aid local management decisions.||en_US