In dereksonderegger/NPS: Visualizing Long-term Monitoring for the NPS

This vignette shows us how to convert data into a format that is useful with the fda package.

library(devtools)
library(fda)
library(dplyr); library(stringr); library(ggplot2); library(tidyr);

install_github('dereksonderegger/NPSData') # Only need to do this when there are changes...
library(NPSData)
head(abcdPlotGpData)

data <- abcdPlotGpData %>% group_by() %>%
  mutate( PlotId = factor(PlotId),
          Park   = factor(str_sub(PlotId,1,4)) )
data$PlotId <- factor(data$PlotId)

Our first pass at visualizing the data. Notice there is no smoothing yet, just connecting the dots.

small.data <- data %>% filter( Park == 'WUPA' ) # 58 Plots in WUPA
ggplot(filter(small.data, yearV==2001),
       aes(x=doyN, y=NdviMed, group=PlotId)) + geom_line()

Now to import these into the fda package. First I want to make 58 curves.

# covariate matrix x is [n, 366]
wide.data <- small.data %>% 
  select(doyN, NdviMed, PlotId, Park, yearV) %>% 
  spread(key=doyN, value=NdviMed  ) %>%
  arrange(PlotId, yearV)

smooth.basis( 1:366,
              y = t(as.matrix(wide.data %>% select(-c(1:3)))),
              fdParobj = create.bspline.basis(rangeval=c(1,366, nbasis=20, norder=4)))

my.basis <- create.bspline.basis(rangeval=c(0,366), nbasis=20, norder=4)

X <- eval.basis(evalarg=small.data$doyN, basisobj = my.basis )
b <- solve(crossprod(X), crossprod(X,small.data$NdviMed))
my.fd <- fd( b, my.basis,  
             list('DOY','Plot' = list(unique(small.data$PlotId)), 'Median NDVI'))

plot(my.fd)

x <- small.matrix$doyN

model <- smooth.basis(small.data$doyN, )

daytime <- 1:365 - .5
JJindex <- c(182:365, 1:181)
tempmat <- daily$tempav[JJindex,]
tempbasis <- create.fourier.basis(c(0,365), 65)
tempfd    <- smooth.basis(daytime, tempmat, tempbasis)$fd