issues
search
weecology
/
MATSS-forecasting
Forecasting Analysis Comparison for Ecological Time Series
https://weecology.github.io/MATSS-forecasting/
Other
5
stars
1
forks
source link
Roadmap (2019 September)
#17
Open
ha0ye
opened
5 years ago
ha0ye
commented
5 years ago
Status Check-in
methods
all the forecast methods in Ward et al. 2014 are implemented
Steve will be contributing code for the Gaussian Process
permutation entropy methods (@garlanjt has cython code up at
https://github.com/garlanjt/Ordinal_TSA
; @pennekampster has an R implementation at
https://github.com/pennekampster/intrinsic_predictability_eco
)
mitchell o'hara-wild, rob hyndman, and earo wang have a package for extracting features from time series (see
https://github.com/tidyverts/feasts
)
data
we have about 80k time series
need to check for duplicates (pulling from multiple databases that could be drawing from the same source)
assemble metadata, and then check for duplicates
Primary Goals
provide guidance to researchers on selecting forecasting methods in ecology
brief discussion of alternative frameworks
Basic approach for the analysis
measure complexity (weighted permutation entropy, others?) and forecast skill across our collection of time series
identify what properties affect forecast skill; how these properties can guide selection of forecast models
Action Items
build metadata for time series datasets
taxon species-level if appropriate
sampling period (start and end times)
sampling frequency (annual, etc.)
geographical location (don't worry about preprocessing for now - keep in whatever form the original data records it as)
life history characteristics if they're readily available
lifespan
age at first reproduction
temperature variation at location of sampling
Frank will look at LPI and GPDD #18 #19
Hao will look at BioTime #20
get prototype of GP-EDM working #21
get an update from Steve about this
check R vs. cython outputs for WPE #22
Hao will write an initial script for this, check with Josh
code for generating synthetic time series #23
Josh will look into doing this in R
Meeting Times
maintain previous schedule
every two weeks on Fridays
alternate between 1300 UTC and 1600 UTC
next call is Friday, October 4 @ 1600 UTC (1200 EST)
Status Check-in
Primary Goals
Basic approach for the analysis
Action Items
Meeting Times
maintain previous schedule