Corrected vignettes + script to produce paper plots and results.

analysis/calibration_exploration.Rmd

@@ -659,6 +659,107 @@ plot_gpp_error +
 ```
 
 
+
+# Calibrate kphio params and kc_jmax 
+
+```{r}
+# Takes 5min to run
+set.seed(2023)
+
+# Define calibration settings
+settings_calib <- list(
+   method = "BayesianTools",
+   metric = rsofun::cost_likelihood_pmodel,
+   control = list(
+       sampler = "DEzs",
+      settings = list(
+            burnin = 3000,
+           iterations = 9000,
+           nrChains = 3,        # number of independent chains
+           startValue = 3       # number of internal chains to be sampled
+       )),
+   par = list(
+       kphio = list(lower = 0.03, upper = 0.15, init = 0.05),
+       kphio_par_a = list(lower = -0.004, upper = -0.001, init = -0.0025),
+       kphio_par_b = list(lower = 10, upper = 30, init =25),
+       kc_jmax = list(lower = 0.2, upper = 0.8, init = 0.41),
+       err_gpp = list(lower = 0.1, upper = 3, init = 0.8)
+   )
+)
+
+# Calibrate kphio-related parameters and err_gpp 
+par_calib <- calib_sofun(
+   drivers = p_model_drivers,
+   obs = p_model_validation,
+   settings = settings_calib,
+   par_fixed = list(
+       soilm_thetastar    = 0.6*240,
+       soilm_betao        = 0.2,
+       beta_unitcostratio = 146.0,
+       rd_to_vcmax        = 0.014,
+       tau_acclim         = 30.0),
+   targets = "gpp"
+)
+
+save(par_calib, file = "analysis/calibration_exploration_files/par_calib_kphio_kcjmax.rda")
+```
+
+```{r echo = TRUE, eval = FALSE}
+# Takes 1min to run
+# Evaluation of the uncertainty coming from the model parameters' uncertainty
+
+# Sample parameter values from the posterior distribution
+samples_par <- getSample(par_calib$mod, 
+                             thin = 100,              # get every 10th sample
+                             whichParameters = 1:5) |>  
+  as.data.frame() |>
+  dplyr::mutate(mcmc_id = 1:n()) |>
+  tidyr::nest(.by = mcmc_id, .key = "pars")
+
+run_pmodel <- function(sample_par){
+  # Function that runs the P-model for a sample of parameters
+  # and also adds the new observation error
+
+  out <- runread_pmodel_f(
+    drivers = p_model_drivers,
+    par =  list(                      # copied from par_fixed above
+      kphio = sample_par$kphio,
+      kphio_par_a = sample_par$kphio_par_a,
+      kphio_par_b = sample_par$kphio_par_b,
+      soilm_thetastar    = 0.6*240,
+      soilm_betao        = 0.2,
+      beta_unitcostratio = 146.0,
+      rd_to_vcmax        = 0.014,
+      tau_acclim         = 30.0,
+      kc_jmax            = sample_par$kc_jmax)       # value from posterior
+  )
+
+  # return modelled GPP and prediction for a new GPP observation
+  gpp <- out$data[[1]][, "gpp"]
+  data.frame(gpp = gpp, 
+             gpp_obs = gpp + rnorm(n = length(gpp), mean = 0, 
+                                   sd = sample_par$err_gpp),
+             date = out$data[[1]][, "date"])
+}
+
+set.seed(2023)
+# Run the P-model for each set of parameters
+pmodel_runs <- samples_par |>
+  dplyr::mutate(sim = purrr::map(pars, ~run_pmodel(.x))) |>
+  # format to obtain 90% credible intervals
+  dplyr::select(mcmc_id, sim) |>
+  tidyr::unnest(sim) |>
+  dplyr::group_by(date) |>
+  # compute quantiles for each day
+  dplyr::summarise(
+    gpp_q05 = quantile(gpp, 0.05, na.rm = TRUE),
+    gpp = quantile(gpp, 0.5, na.rm = TRUE),          # get median
+    gpp_q95 = quantile(gpp, 0.95, na.rm = TRUE),
+    gpp_obs_q05 = quantile(gpp_obs, 0.05, na.rm = TRUE),
+    gpp_obs_q95 = quantile(gpp_obs, 0.95, na.rm = TRUE)
+  )
+```
+
 # Old code:
 
 ```{r}