In stineb/fvar:

devtools::load_all(".")
library(rsofun)
library(dplyr)
library(readr)
library(lubridate)
library(ggplot2)

Get data

usesite <- "CH-Lae"

getvars <- c( 
  "GPP_NT_VUT_REF", 
  "GPP_DT_VUT_REF",                
  "LE_F_MDS", 
  "LE_F_MDS_QC", 
  "NEE_VUT_REF", 
  "NEE_VUT_REF_QC", 
  "TA_F",
  "VPD_F",
  "P_F",
  "SW_IN_F",
  "NETRAD"
)

df_fluxnet <- get_obs_bysite_fluxnet2015( 
  sitename=usesite, 
  path_fluxnet2015 ="~/data/FLUXNET-2015_Tier1/20191024/HH/", 
  timescale="hh", 
  getvars=getvars,
  threshold_GPP = 0,      # take only measured data
  getswc = TRUE
  ) %>% 
  mutate(sitename = usesite) %>% 

  ## use daytime data only
  dplyr::filter(ppfd > 0.0) %>% 

  ## use growing season of CH-Lae only
  dplyr::filter(yday(date)>121 & yday(date)<121+161) %>% 

  ## additional cleaning (only positive VPD)
  dplyr::filter(vpd>0)

Prepare data for training

Prepare training data, removing NAs and outliers.

# ## use all observational soil moisture data
# varnams_soilm <- df_fluxnet %>% 
#   dplyr::select( starts_with("SWC_") ) %>% 
#   dplyr::select( -ends_with("QC") ) %>% 
#   names()

## define settings used by multiple functions as a list
settings <- list( 
  target        = "NEE_VUT_REF", 
  predictors    = c("temp","vpd", "ppfd"), 
  varnams_soilm = "wcont_splash",
  nneurons_good = 10,
  nneurons_all  = 10,
  nrep          = 3,
  package       = "nnet"
  )

df_train <- prepare_trainingdata_fvar( df_fluxnet, settings )

Train models and predict fvar

Train models for one set of soil moisture input data. In this case it's observational data.

df_nn_soilm_obs <- train_predict_fvar( 
  df_train,
  settings,
  soilm_threshold    = 0.6,
  weights            = NA, 
  verbose = TRUE
  )

stineb/fvar documentation built on Oct. 15, 2022, 12:06 p.m.