In this project, we will attempt to predict the temperature in Knoxville, TN using the ts and forecast libraries.

First we’ll read in recent temperature data for Knoxville, TN. This was obtained for free from the National Centers for Environmental Information website. We’ll use data from 1/1/2010 to 5/30/2017. For simplicity we’ll just focus on daily maximum temperatures from one weather station. We’ll also use lubridate to turn the DATE column into a genuine date column.

Temperatures <- read_csv("983224.csv") %>%
  filter(STATION_NAME == "KNOXVILLE MCGHEE TYSON AIRPORT TN US") %>%
  select(DATE, TMAX) %>% mutate(DATE = ymd(DATE)) %>%
  filter(DATE >= "2010-1-1", DATE < "2017-5-31")

Next we’ll plot the last decade of temperature data. The yearly cycle is clearly visible.

Temperatures %>% ggplot(aes(x=DATE, y=TMAX)) + geom_line() + labs(y = "daily high temperature")

TempTS <- msts(Temperatures %>% select(-DATE), seasonal.periods=c(7,365.25), start = 2010)
fit <- tbats(TempTS)
fc <- forecast(fit)
plot(fc)

Predictions <- as.data.frame(fc$mean)
names(Predictions) <- "TMAX"

Predictions <- Predictions %>%
  mutate(DATE = ymd(20170530) + days(row_number())) %>%
  select(DATE, TMAX)

Predictions %>% head(10) %>% kable()