Ingest P-model drivers from coordinates

Pepa Aran

2023-03-23

Source: vignettes/get_drivers_coordinates.Rmd

Get all P-model drivers from a pair of coordinates

This vignette runs through the workflow to obtain an object like p_model_drivers (from the rsofun package) for arbitrary locations in the globe, given by their coordinates (lon, lat). These sites aren’t necessarily flux sites.

First, we compile all the necessary data (complete site information, meteorological data, climatic variables, etc.) from the raw data files. In this vignette, we use the paths corresponding to the GECO data archive (more information here). Then, we format the data to run the rsofun P-model implementation.

Site information and coordinates

As an example, we will use a subset of site locations from Atkin et al. 2017 (the original data is available via the TRY Plant Trait Database). All the columns in the siteinfo_globresp data.frame are necessary to retrieve the P-model drivers data: a site name for reference sitename, the coordinates of the site location in degrees lon and lat, its elevation in meters elv, and the first and last year of observations we want to collect year_start and year_end.

# Load siteinfo data
load("../data/siteinfo_globresp.rda")

Complement siteinfo with Water Holding Capacity

We use data from Stocker et al. 2021 to get the WHC for our sites. The data is in 0.05 degree resolution, which is appropriate for this use case.

# Define function to extract whc
extract_whc <- function(file, siteinfo){

  siteinfo <- siteinfo[, c("lon", "lat")] |>
    unique()                                   # Remove repeated coordinates
  
  rasta <- raster::brick(file)
  
  raster::extract(
    rasta,                                            # raster object
    sp::SpatialPoints(siteinfo[, c("lon", "lat")]),   # points
    sp = TRUE                                         # add raster values to siteinfo
  ) |>
    tibble::as_tibble() |>
    dplyr::rename(whc = layer)
}
# Get WHC data
path_whc <- "/data/archive/whc_stocker_2021/data"

df_whc <- extract_whc(file = paste0(path_whc, "/cwdx80.nc"),
            siteinfo = siteinfo_globresp)

# Merge whc with site information
siteinfo <- dplyr::left_join(
  siteinfo_globresp,
  df_whc,
  by = c("lon", "lat")
  )

# Fill gaps by median value (here there are no NAs)
siteinfo$whc[is.na(siteinfo$whc)] <- median(siteinfo$whc,
                                            na.rm = TRUE)

Get forcing datasets

Based on the siteinfo object, we collect all climate forcing data.

Meteorological forcing from WATCH-WFDEI

The following meteorological variables are obtained for the P-model forcing from the WATCH-WFDEI data: - temp: Daily temperature - prec: Daily precipitation - ppfd: Photosynthetic photon flux density - vpd: Vapor pressure deficit - patm: Atmospheric pressure

Loading this data can take a while, up to hours if you need several years of data for many sites.

# Get WATCH data
path_watch <- "/data/archive/wfdei_weedon_2014/data"
path_worldclim <- "/data/archive/worldclim_fick_2017/data" # for debiasing

df_watch <- ingestr::ingest(
  siteinfo = siteinfo,
  source = "watch_wfdei",
  getvars = c("temp", "prec", "ppfd", "vpd", "patm"),
  dir = path_watch,
  settings = list(
    correct_bias = "worldclim",    # 0.5 deg
    dir_bias = path_worldclim
  )
) |>
  suppressWarnings() |> suppressMessages()
# Variables tmin and tmax missing but not currently in use

# Memory intensive, purge memory
gc()

Cloud cover from CRU

Now, let’s complete the forcing with cloud cover ccov values from CRU data.

# Get CRU data
path_cru <- "/data/archive/cru_harris_2024/data/"

df_cru <- ingestr::ingest(
  siteinfo = siteinfo,
  source = "cru",
  getvars = c("ccov"),
  dir = path_cru,
  settings = list(
    correct_bias = NULL       # 0.5 deg resolution
  )
)

Merge meteorological drivers

Let’s put together the previous data, into a single data.frame:

df_meteo <- df_watch |>
  tidyr::unnest(data) |>
  left_join(
    df_cru |>
      tidyr::unnest(data),
    by = c("sitename", "date")
  ) |>
  dplyr::ungroup() |>                         # keep unnested
  dplyr::mutate(tmin = temp,                  # placeholders for variables tmin and tmax
                tmax = temp) |>               # (not used in P-model runs)
  dplyr::mutate(doy = lubridate::yday(date))  # add day of year

head(df_meteo)

Get CO2 data

The following chunk downloads the CO2 yearly average data from the Mauna Loa observatory and appends it to the meteorological drivers from above.

# Download CO2 data
df_co2 <- read.csv(
  url("https://gml.noaa.gov/webdata/ccgg/trends/co2/co2_annmean_mlo.csv"),
  skip = 59) |>
  dplyr::select(year, mean) |>
  dplyr::rename(co2 = mean)

# Merge with meteo drivers
df_meteo <- df_meteo |>
  dplyr::mutate(year = lubridate::year(date)) |>
  dplyr::left_join(df_co2, by = "year")            # keep unnested

Append fAPAR

Set fAPAR by default to 1 for all observations, which is the same as what ingest(siiteinfo_globresp, source = "fapar_unity") would do. This makes sense for leaf-level simulations.

# Add fapar column with value 1
df_meteo <- df_meteo |>
  dplyr::mutate(fapar = 1)

Put all data together into rsofun object

Up to here, we have merged all of the forcing data into df_meteo. Let’s put it in a nested format, such that there is a single data.frame per site:

# Nest forcing
df_meteo <- df_meteo |>
  dplyr::group_by(sitename) |>
  tidyr::nest() |>
  dplyr::rename(forcing = data) |>
  dplyr::ungroup()

# Add site information
df_siteinfo <- siteinfo |>
  dplyr::group_by(sitename) |>
  tidyr::nest() |>
  dplyr::rename(site_info = data) |>
  dplyr::ungroup()

We will use default simulation and soil texture parameters.

# Define default model parameters, soil data, etc
params_siml <- list(
    spinup             = TRUE,  # to bring soil moisture to steady state
    spinupyears        = 10,    # 10 is enough for soil moisture.
    recycle            = 1,     # number of years recycled during spinup 
    soilmstress        = FALSE, # soil moisture stress function is included
    tempstress         = FALSE, # temperature stress function is included
    calc_aet_fapar_vpd = FALSE, # set to FALSE - should be dropped again
    in_ppfd            = TRUE,  # if available from forcing files, set to TRUE
    in_netrad          = FALSE, # if available from forcing files, set to TRUE
    outdt              = 1,
    ltre               = FALSE,
    ltne               = FALSE,
    ltrd               = FALSE,
    ltnd               = FALSE,
    lgr3               = TRUE,
    lgn3               = FALSE,
    lgr4               = FALSE
  )

df_soiltexture <- bind_rows(
    top    = tibble(
      layer = "top",
      fsand = 0.4,
      fclay = 0.3,
      forg = 0.1,
      fgravel = 0.1),
    bottom = tibble(
      layer = "bottom",
      fsand = 0.4,
      fclay = 0.3,
      forg = 0.1,
      fgravel = 0.1)
  )

Finally, we can compile all the data together into a p_model_drivers object. Note that this object must be an ungrouped tibble object.

drivers <- df_meteo |>
  dplyr::left_join(df_siteinfo,
                   by = "sitename")

drivers$params_siml <- list(dplyr::as_tibble(params_siml))
drivers$params_soil <- list(df_soiltexture)

Run P-model example

We can now run the P-model with the rsofun implementation.

# Run P-model with parameters from previous optimizations
params_modl <- list(
    kphio           = 0.09423773,
    soilm_par_a     = 0.33349283,
    soilm_par_b     = 1.45602286,
    tau_acclim_tempstress = 10,
    par_shape_tempstress  = 0.0
  )

# run the model for these parameters
output <- rsofun::runread_pmodel_f(
  drivers = drivers,
  par = params_modl
)

And take a look at the results:

ggplot() +
    geom_line(
    data = output |> tidyr::unnest(),
    aes(
      x = date,
      y = vcmax25
    ),
    alpha = 0.8
  ) +
  facet_grid(sitename ~ .) +
  labs(
    x = "Date",
    y = "Vcmax25"
  )

ggplot() +
    geom_line(
    data = output |> tidyr::unnest(),
    aes(
      x = date,
      y = gpp
    ),
    alpha = 0.8
  ) +
  facet_grid(sitename ~ .) +
  labs(
    x = "Date",
    y = "GPP"
  )

Aggregate over 2001-2015 period to obtain across-years average forcing

To run the P-model for leaf trait calibration, we may want to have the average daily forcing over a span of years. This is because leaf traits are measured once and assumed constant over a span of time.

# Define function to aggregate forcing over day of year
aggregate_forcing_doy <- function(forcing){
  forcing |>
    dplyr::group_by(doy) |>
    summarise(date = first(date),                # 2001 as symbolic date
              temp = mean(temp, na.rm = TRUE),
              rain = mean(rain, na.rm = TRUE),
              vpd = mean(vpd, na.rm = TRUE),
              ppfd = mean(ppfd, na.rm = TRUE),
              snow = mean(snow, na.rm = TRUE),
              co2 = mean(co2, na.rm = TRUE),
              fapar = mean(fapar, na.rm = TRUE),
              patm = mean(patm, na.rm = TRUE),
              tmin = mean(tmin, na.rm = TRUE),
              tmax = mean(tmax, na.rm = TRUE),
              ccov = mean(ccov, na.rm = TRUE)
              ) |>                               # already ungrouped
    dplyr::slice(1:365)                          # ignore leap years
}

# Compute aggregated drivers
drivers_aggregated <- drivers |>
  dplyr::mutate(forcing = 
                  lapply(drivers$forcing, aggregate_forcing_doy))

Finally, let’s run the P-model on the aggregated data and visualise the results.

# run the model for these parameters
output_aggregated <- rsofun::runread_pmodel_f(
  drivers = drivers_aggregated,
  par = params_modl
)

ggplot() +
    geom_line(
    data = output_aggregated |> tidyr::unnest(),
    aes(
      x = date,
      y = vcmax25
    ),
    alpha = 0.8
  ) +
  facet_grid(sitename ~ .) +
  labs(
    x = "Date",
    y = "Vcmax25"
  )

ggplot() +
    geom_line(
    data = output_aggregated |> tidyr::unnest(),
    aes(
      x = date,
      y = gpp
    ),
    alpha = 0.8
  ) +
  facet_grid(sitename ~ .) +
  labs(
    x = "Date",
    y = "GPP"
  )