GE585 Spring 2019: CDC Aedes mosquito forecast competition
-- Packages required: tidyverse, daymetr, zoo
- Run 1_Climate_data_import.R to pull in and plot county-level temperature, precipitation, and humidity data (years 2013-2017).
Taylor Perez ttperez@bu.edu 401-834-0343
Juliette Bateman jbateman@bu.edu 732-966-7586
John Foster fosterj@bu.edu 651-558-7362
Casey Kelly caseyk@bu.edu 908-547-7209
Carina Terry cterry@bu.edu 817-821-7115
Monthly abundance data from the CDC is downladed from the subset_aedes_data function in the County_subset.R script. Out of the 95 counties the CDC provided data, we chose the 10 counties that have at least two years of data (2016 and 2017) that aren't all zero or just presence/absence. The function is used in 1_Climate_data_import.R, and daily weather is downloaded from daymet for each county.
Within 1_Climate_data_import.R, daily precipitation, min and max temperature, and vapor pressure are downloaded. We then aggregate the daily observations to monthly mean, and calculate mean monthly relative humidity and total precipitaiton for each month to the data.
Currently we have two functions for fitting the training data, Random_Walk_Fit.R and GLM_Fit.R. The first is just a random walk, while the latter fits any combination of monthly weather covariates as an additive dynamic model. Both fits model the data with a Poisson distribution as we are dealing with counth data, and the process is modelled with a normal distribution. Process error is given an uninformative gamma prior to keep the zero bound on precision and for conjugacy.