NWT_phenocam

NWT ridge phenocam analysis code

Overview:

The basic workflow for running is to go to the scripts directory and run the scripts in order. Before running the scripts, be sure to make a /plots directory -- this houses some interim figures not used in the final analysis without it the code will have some errors trying to write to non-existing locations.

Processing details

• 0_download_data.R will download 3 data packages from EDI.

  • The phenocam data v2 (knb-lter-nwt.192.3)
    Elwood, K., J. Smith, S. Elmendorf, and Niwot Ridge LTER. 2022. Time-lapse camera (phenocam) imagery of sensor network plots, 2017 - ongoing. ver 3. Environmental Data Initiative. https://doi.org/10.6073/pasta/285918fbf5cc4bd2ed2c1241db9a1b2d (Accessed 2022-05-23).

  • The saddle sensor node met data v3 (knb-lter-nwt.210.4)
    Morse, J. and Niwot Ridge LTER. 2022. Climate data for saddle catchment sensor network, 2017 - ongoing. ver 4. Environmental Data Initiative. https://doi.org/10.6073/pasta/598894834ea3bae61d7550c30da06565 (Accessed 2022-05-23).

  • 5cm multispectral UAV imagery (knb-lter-nwt.67.1) Wigmore, O. and Niwot Ridge LTER. 2021. 5cm multispectral imagery from UAV campaign at Niwot Ridge, 2017 ver 1. Environmental Data Initiative. https://doi.org/10.6073/pasta/a4f57c82ad274aa2640e0a79649290ca (Accessed 2022-07-06). Select files from this need to be unzipped as well outside of the code.

• 1a_prep_data.R
  Does some initial preprocessing of the phenocam and met data, removing flagged values from the met data and averaging up to daily values. It also estimates snowmelt/snowfall doy where not recorded by the cameras as the first doy after doy 100 where the 5d running mean diurnal temperature range at 5cm exceeds the mean 5cm dtr at recorded snowmelt at that node. Note that snowfall occurred more synchronously than snowmelt across the nodes, so snowfall dates were assigned by inspection in some cases. Where node-specific met data were absent due to sensor failures, the snowmelt dates are not infilled, as the ability of the infilling to exactly replicate temperatures around the freezing point was unknown. This script also pads out missing data to 365 days per year. Generates files:

  • data/all_met_no_freezefill.csv
  • data/phen_clim_all.csv'

Note I had experimented with infilling moisture values when soil temperatures dropped below freezing, but this approach yielded some peculiar results. Some sources indicate that the soil moisture may be roughly accurate even a few degrees below 0C, so moisture data where soil temps were below freezing are left as they were recorded by the instruments. For the phenology analyses here, moisture values while soil is frozen are largely irrelevant but for applications where precise overwinter soil temperatures are needed an alternate treatment of overwinter soil moisture might be desirable.

• 1b_infill_met.R Infills the daily met data. Met data are infilled sequentially starting with air temperature, so that infilled values (e.g. air T at a node can be used to predict missing soilT at that node, etc). The algorithm uses MICE for infilling the air/soil temperatures, and mtdsi for infilling soil moisture. Soil moisture shows a high degree of temporal autocorrelation, so mtdsi gave much more sensible results for soil moisture than using MICE, which generated unrealistic discontinuities. In general, the algorithms try to infill based on the relationships in a narrow window (e.g. one month) around the datagaps, but the windows are extended in some cases where insufficient data to generate the relationships exist within that gap. This generates the file data/met_data_infilled.csv, which is used in subsequent analysis. It also generates 5 intermediary file all_met_infilled_seq_1...5_no_freeze_fill.rds. These are not used but can be helpful if you need to rerun this slow-running script in chunks, because each step takes ages (for future improvement this code could certainly be optimized to run more efficiently in parallel).

• 2_analyze_saddle_phenocams.R generates phenometrics from the phenocam data using a modified double-log beck workflow (sourced from file functions/FitDoubleLogBeck_modified.r. The basic code for the double-log Beck function was derived from the phenopix package and modified as follows:

  • 1. use the original Beck equation (a -1 is missing from the phenopix implementation - though this does not affect the results much)
  • 2. as well as allowing missing data in the time series.
  • 3. Weight the data with 0.15 weights given to any greenness values that were snow-covered at the time of measurement. These values are infilled based on the sensor-specific GCC values at snowmelt (random draws from a normal distribution based on the empirical mean +sd for that node).

Generates files data_deriv/phenometrics.csv

• 3_phen_surv.R runs survival models on the derived phenometrics, using AIC to select the best model for each phenophase and then doing leave-one-out cross-validation, sequentially dropping nodes and years. Generates some figures at the end.

• 4_plot_locations.R makes some nice plots of the locations of the study

• NOTE: plots used in the manuscript are rendered to /ms_plots; others to /plots