modelNDVI() function from phenex package, other package for smoothing and analysing irregular multi-year time series?

Question

I have multiple time series as *.csv files. They all have a first column for date, and a second column for (ndvi) values. The time series is describes irregular, a few years are missing in between the 30 year time span and also the values are not spaced evenly within each year (which excludes working with class "ts" and a few other from the beginning). Each year in my series has some kind of seasonal amplitude that i would like to fit a function to.
For those familiar with phenology and remote sensing, i would like to basically retrieve the same metrics as what TIMESAT does (for the years containing enough values). I've found the phenex package, including the function modelNDVI(), which I think would do exactly what I am looking for - correcting my values using bise (best index slope extraction) and smoothing with Savitzky-Golay or Assymetric Gaussian. My data looks basically like this:

head(test)
     system.time_start         X1
3  1985-01-17 01:00:00 0.04546319
6  1985-02-10 01:00:00 0.05106874
7  1985-02-18 01:00:00 0.10060238
8  1985-02-26 01:00:00 0.04757872
9  1985-03-06 01:00:00 0.03484827
11 1985-03-22 01:00:00 0.02705866

...
2357 2015-11-17 01:00:00 0.04524300
2358 2015-11-17 01:00:00 0.04476613
2359 2015-11-25 01:00:00 0.03424461
2360 2015-11-25 01:00:00 0.05062974
2364 2015-12-11 01:00:00 0.09578227
2368 2015-12-27 01:00:00 0.09661864

Clearly modelNDVI() needs a vector, so:
test.vec <- as.vector(test$X1) .
I've also found out it must be a vector of length [1:365] which is my first problem. So I figured i have to convert my vector to another vector that is of the desired length, containing my values and a lot of NA's to be handled by the function.
Second problem:
I would like to model my whole time series, not only a single year (modelNDVI() only accpects year.int as a single observation year as an argument). So my question is, does anyone know if I can get modelNDVI() to work with my dataset? And if not, can anyone suggest a different way to handle my problem with modeling my irregular multi-year time series? Any packages, tutorials? I'm still relatively new to R and while I was looking for a solution I came accross numerous ways of handling irregular timeseries but none really could solve my problem of ending up with correcting and modeling my data. Thanks for any advice!

Answer 1

After puzzling over a similar problem, I realized that the NDVI vector needs to be 365/366 days long (as you discovered), but also that it needs to be an NDVI class object. Check out ?bise1 for example code.

As for multi-year applications, I can't comment.

Answer 2

Ran into this same problem. Here is my fix which creates a vector 365 days long and adds the ndvi value for each day i had data based on the ndvi.tiff file (eg 20160131_NDVI.tiff is 31/01/2016 or 31st day of the year) but you can easily modify this if you already have the ndvi values and their respective date of acquisition.

library(raster)
library(plyr)
library(rgdal)
library(rgeos)
library(phenex)
library(lubridate)

ndvi.list <- list.files(path = ".", pattern = "*ndvi.tif$", ignore.case = TRUE, 
full.names = TRUE, recursive = TRUE)
ndvi.stack <- stack(ndvi.list)
ndvi_values <- matrix(0,0,366)

# find mean ndvi (e.g. within a shapefile boundaries)
ndvi <- extract(ndvi.stack, shapefile, fun=mean, df=TRUE)

#use filenames to retrieve data information from ndvi.tiff files (e.g. mine were 
called 20160210_NDVI.tif)
ndvicols <- names(ndvi.stack)  
ndvicols <- sub("X","",ndvicols) 
ndvicols <- sub("_NDVI","",ndvicols)
dates <- as.Date(ndvicols, format = "%Y%m%d")
doy <- lubridate::yday(dates)  

#create ndvi vector
vec <- rep(0, 365)
ndvi.vals <- as.vector(t(ndvi[,2:ncol(ndvi)]))
vec[doy] <- ndvi.vals

# create NDVI object, apply correction (e.g. BISE) and model values (e.g. Linear interpolation)
ndvi.list <- modelNDVI(ndvi.values=vec,
                    year.int=2016, multipleSeasons=FALSE, correction="bise",
                    method="LinIP", MARGIN=2, doParallel=FALSE, slidingperiod=10)

# create vector of modelled values
vec.mod <- ndvi.list[[1]]@modelledValues

NOTE: for multiple years you'll have to run each year separately and append the data for each subsequent year