Tried weighting MSEs

.github/workflows/run-r-test.yml

@@ -23,7 +23,7 @@ jobs:
 
     - name: Install R packages
       run: |
-        Rscript -e 'install.packages(c("here", "assertthat"))'  # Add any packages your script needs
+        Rscript -e 'install.packages(c("here", "assertthat", "dplyr"))'  # Add any packages your script needs
     - name: Run R script
       run: |
         Rscript tests/error_tests.R

results/initial_norm_exp.txt

@@ -0,0 +1,39 @@
+parameters	values
+beta	0.732
+q10_rh	2.64
+diff	2.4
+
+Objective Function Value: 0.00173
+
+CO2 MSE: 58.5
+T MSE:   0.0651
+RMSE:    7.65
+
+***Key Metrics***
+TCRE: 1.53
+TCR:  1.76
+
+***Historical Warming and ERF***
+GSAT Warming:              0.633
+Ocean Heat Content Change: 423
+Total Aerosol ERF:         -1.24
+WMGHG ERF:                 3.63
+Methane ERF:               0.538
+
+***Future Warming***
+scenario	start	end	GSAT
+ssp119	2021	2040	0.694
+ssp119	2041	2060	0.842
+ssp119	2081	2100	0.693
+ssp126	2021	2040	0.709
+ssp126	2041	2060	1.02
+ssp126	2081	2100	1.05
+ssp245	2021	2040	0.715
+ssp245	2041	2060	1.23
+ssp245	2081	2100	1.9
+ssp370	2021	2040	0.725
+ssp370	2041	2060	1.37
+ssp370	2081	2100	2.86
+ssp585	2021	2040	0.842
+ssp585	2041	2060	1.68
+ssp585	2081	2100	3.71

scripts/error_functions.R

@@ -33,6 +33,19 @@ rmse <- function(x, y) {
   return(sqrt(mse(x, y)))
 }
 
+# nmse - function to find normalized mean squared error between two vectors
+#        (NMSE = MSE / x^2)
+#
+# args: 
+#   x - observed values
+#   y - predicted values
+#
+# returns: numeric vector length 1 containing the NMSE between predicted and
+#          observed values
+nmse <- function(x, y) {
+  return(mse(x, y) / sum(x^2))
+}
+
 
 # get_var_mse: function to find MSE between observed and predicted data for
 #              a given variable
@@ -42,16 +55,17 @@ rmse <- function(x, y) {
 #   hector_data - data frame outputted by Hector
 #   var         - variable name
 #   yrs         - vector of years for finding MSE
+#   mse_fn      - function to calculate mse with (default = mse)
 #
 # Returns: MSE between predicted and observed data for var
-get_var_mse <- function(obs_data, hector_data, var, yrs) {
+get_var_mse <- function(obs_data, hector_data, var, yrs, mse_fn = mse) {
   obs_vec <- 
     filter(obs_data, year %in% yrs & variable == var)$value
 
   hector_vec <-
     filter(hector_data, year %in% yrs & variable == var)$value
 
-  return(mse(obs_vec, hector_vec))
+  return(mse_fn(obs_vec, hector_vec))
 }
 
 # mean_T_CO2_mse: function to find the mean of the temperature and CO2 MSEs 
@@ -75,6 +89,28 @@ mean_T_CO2_mse <- function(obs_data, hector_data) {
   return(mean(c(T_mse, CO2_mse)))
 }
 
+# mean_T_CO2_nmse: function to find the mean of the temperature and CO2 NMSEs 
+#                  between observed and predicted data for a given variable
+#
+# args: 
+#   obs_data    - data frame of observed data formatted like Hector data frame
+#   hector_data - data frame outputted by Hector
+#
+# Returns: NMSE between predicted and observed data for var
+mean_T_CO2_nmse <- function(obs_data, hector_data) {
+  T_mse <- get_var_mse(obs_data = obs_data, 
+                       hector_data = hector_data, 
+                       var = GMST(), 
+                       yrs = 1850:2014,
+                       mse_fn = nmse)
+  CO2_mse <- get_var_mse(obs_data = obs_data, 
+                         hector_data = hector_data, 
+                         var = CONCENTRATIONS_CO2(), 
+                         yrs = c(1750, 1850:2014),
+                         mse_fn = nmse)  
+  return(mean(c(T_mse, CO2_mse)))
+}
+
 # smooth_T_CO2_mse: function to find the mean of smoothed temperature and CO2
 #                   MSEs between observed & predicted data for a given variable
 #
@@ -97,7 +133,6 @@ smooth_T_CO2_mse <- function(obs_data, hector_data) {
   CO2_mse <- get_var_mse(obs_data = obs_data, 
                          hector_data = hector_data, 
                          var = CONCENTRATIONS_CO2(), 
-                         yrs = c(1750, 1850:2014))
-
+                         yrs = c(1750, 1850:2014))  
   return(mean(c(T_mse, CO2_mse)))
 }

scripts/graph_comparison_plots.R

@@ -0,0 +1,70 @@
+# Script to compare plots that use different hector variables
+# Author: Peter Scully
+# Date: 6/13/24
+
+### Constants and Imports ###
+
+# Importing libraries
+library(hector)
+library(ggplot2)
+
+# Setting up file paths
+COMP_DATA_DIR <- file.path(here::here(), "comparison_data")
+SCRIPTS_DIR <- file.path(here::here(), "scripts")
+RESULTS_DIR <- file.path(here::here(), "results")
+
+CO2_PATH <- file.path(COMP_DATA_DIR,
+                      "Supplementary_Table_UoM_GHGConcentrations-1-1-0_annualmeans_v23March2017.csv")
+TEMP_PATH <-
+  file.path(COMP_DATA_DIR,
+            "HadCRUT.5.0.2.0.analysis.summary_series.global.annual.csv")
+
+INI_FILE <- system.file("input/hector_ssp245.ini", package = "hector")
+PARAMS <- c(BETA(), Q10_RH(), DIFFUSIVITY())
+
+OUTPUT <- file.path(RESULTS_DIR, "comparison_plots.jpeg")
+
+
+source(file.path(SCRIPTS_DIR, "major_functions.R"))
+
+### Getting observational data ###
+co2_data <- get_co2_data(CO2_PATH)
+co2_data$lower <- co2_data$value
+co2_data$upper <- co2_data$value
+
+temp_data <- get_temp_data(TEMP_PATH, include_unc = T)
+temp_data <- filter(temp_data, year <= 2014)
+
+obs_data <- rbind(co2_data, temp_data)
+
+### Running Hector ###
+default_data <- run_hector(ini_file = INI_FILE, 
+                           params = NULL,
+                           vals = NULL,
+                           yrs = 1750:2014, 
+                           vars = c(GMST(), CONCENTRATIONS_CO2()))
+default_data$scenario <- "Hector - Default Fit"
+
+nmse_data <- run_hector(ini_file = INI_FILE,
+                        params = PARAMS,
+                        vals = c(0.732, 2.64, 2.4),
+                        yrs = 1750:2014,
+                        vars = c(GMST(), CONCENTRATIONS_CO2()))
+nmse_data$scenario <- "Hector - Fit to NMSEs"
+
+hector_data <- rbind(default_data, nmse_data)
+hector_data$lower <- hector_data$value
+hector_data$upper <- hector_data$value
+
+comb_data <- rbind(obs_data, hector_data)
+
+ggplot(data = comb_data, aes(x = year, y = value, color = scenario)) + 
+  geom_ribbon(data = 
+               filter(comb_data, scenario == "historical" & variable == GMST()),
+              aes(ymin = lower, ymax = upper),
+              fill = 'aquamarine1',
+              color = NA) +
+  geom_line() +
+  facet_wrap(~ variable, scales = "free") +
+  ggtitle("Comparing Parameterizations")
+ggsave(OUTPUT, width = 15)

scripts/initial_norm_exp.R

@@ -0,0 +1,69 @@
+# Script to try normalizing the T and CO2 MSEs
+# manuscript
+# Author: Peter Scully
+# Date: 6/13/24
+
+### Constants and Imports ###
+
+# Importing libraries
+library(hector)
+
+# Setting up file paths
+COMP_DATA_DIR <- file.path(here::here(), "comparison_data")
+SCRIPTS_DIR <- file.path(here::here(), "scripts")
+RESULTS_DIR <- file.path(here::here(), "results")
+
+CO2_PATH <- file.path(COMP_DATA_DIR,
+                      "Supplementary_Table_UoM_GHGConcentrations-1-1-0_annualmeans_v23March2017.csv")
+TEMP_PATH <-
+  file.path(COMP_DATA_DIR,
+            "HadCRUT.5.0.2.0.analysis.summary_series.global.annual.csv")
+
+INI_FILE <- system.file("input/hector_ssp245.ini", package = "hector")
+PARAMS <- c(BETA(), Q10_RH(), DIFFUSIVITY())
+
+OUTPUT <- file.path(RESULTS_DIR, "initial_norm_exp.txt")
+
+
+source(file.path(SCRIPTS_DIR, "major_functions.R"))
+
+### Getting observational data ###
+co2_data <- get_co2_data(CO2_PATH)
+temp_data <- get_temp_data(TEMP_PATH)
+obs_data <- rbind(co2_data, temp_data)
+
+### Calling optim ###
+best_pars <- run_optim(obs_data = obs_data,
+                       ini_file = INI_FILE,
+                       params = PARAMS,
+                       lower = c(0.5 - 0.232, 2.2 - 0.44, 2.3 - 0.1),
+                       upper = c(0.5 + 0.232, 2.2 + 0.44, 2.3 + 0.1),
+                       yrs = 1750:2014,
+                       vars = c(GMST(), CONCENTRATIONS_CO2()),
+                       error_fn = mean_T_CO2_nmse,
+                       method = "L-BFGS-B",
+                       output_file = OUTPUT)
+
+### Outputting individual MSEs ###
+hector_data <- run_hector(ini_file = INI_FILE, 
+                          params = PARAMS, 
+                          vals = best_pars, 
+                          yrs = 1750:2014, 
+                          vars = c(GMST(), CONCENTRATIONS_CO2()))
+
+T_mse <- get_var_mse(obs_data = obs_data, 
+                     hector_data = hector_data, 
+                     var = GMST(), 
+                     yrs = 1850:2014)
+CO2_mse <- get_var_mse(obs_data = obs_data, 
+                       hector_data = hector_data, 
+                       var = CONCENTRATIONS_CO2(), 
+                       yrs = c(1750, 1850:2014))
+
+write_metric("CO2 MSE:", CO2_mse, OUTPUT)
+write_metric("T MSE:  ", T_mse, OUTPUT)
+write_metric("RMSE:   ", sqrt(mean(CO2_mse, T_mse)), OUTPUT) # not 100% sure this is how we want to calculate this
+write("", OUTPUT, append = TRUE)
+
+### Outputting table metrics ###
+calc_table_metrics(PARAMS, best_pars, OUTPUT)

scripts/major_functions.R

@@ -53,12 +53,12 @@ get_co2_data <- function(file, scenario = "historical") {
   # Fixing table formatting
   co2_data <- na.omit(co2_data)
   colnames(co2_data) <- c("year", "value")
-
+  
   # Adding in new columns to match Hector data frames
   co2_data$scenario <- scenario
   co2_data$variable <- CONCENTRATIONS_CO2()
   co2_data$units <- " ppmv CO2"
-
+  
   return(co2_data)
 }
 
@@ -74,18 +74,18 @@ get_co2_data <- function(file, scenario = "historical") {
 get_temp_data <- function(file, scenario = "historical", include_unc = F) {
   temp_data <- read.csv(file)
   colnames(temp_data) <- c("year", "value", "lower", "upper")
-
+  
   # Adding in new columns to match Hector data frames
   temp_data$scenario <- scenario
   temp_data$variable <- GMST()
   temp_data$units <- "degC"
-
+  
   # Removing upper and lower bounds, if desired
   if (!include_unc) {
     temp_data$lower <- NULL
     temp_data$upper <- NULL
   }
-
+  
   return(temp_data)
 }
 
@@ -407,7 +407,7 @@ run_optim <- function(obs_data, ini_file, params, par = NULL, sd = NULL,
 #
 # returns: Nothing, but outputs table to given output file
 calc_table_metrics <- function(params, vals, output_file) {
-
+  
   ### KEY METRICS ###
 
   ## Finding TCRE ##
@@ -625,4 +625,4 @@ calc_table_metrics <- function(params, vals, output_file) {
               quote = FALSE,
               sep = "\t",
               row.names = FALSE)
-}
+}

tests/error_tests.R

@@ -4,6 +4,7 @@
 source(here::here("scripts", "error_functions.R"))
 
 library(assertthat)
+library(dplyr)
 
 # Confirming that mse works as expected
 mse_tests <- function() {
@@ -35,34 +36,29 @@ get_var_mse_tests <- function() {
   assert_that(get_var_mse(obs_data, 
                           hect_data, 
                           var = "CO2_concentrations", 
-                          start = 2000, 
-                          end = 2015) == 0)
+                          yrs = 2000:2015) == 0)
 
   assert_that(get_var_mse(obs_data, 
                           hect_data, 
                           var = "CO2_concentrations", 
-                          start = 2000, 
-                          end = 2000) == 0)
+                          yrs = 2000) == 0)
 
   assert_that(get_var_mse(obs_data, 
                           hect_data, 
                           var = "GMST", 
-                          start = 2000, 
-                          end = 2000) == 1)
+                          yrs = 2000) == 1)
 
   assert_that(get_var_mse(obs_data, 
                           hect_data, 
                           var = "GMST", 
-                          start = 2000, 
-                          end = 2001) == 0.5)
+                          yrs = 2000:2001) == 0.5)
 
   assert_that(get_var_mse(obs_data, 
                           hect_data, 
                           var = "GMST", 
-                          start = 2000, 
-                          end = 2004) == 3)
+                          yrs = 2000:2004) == 3)
 
 
 }
-
-mse_tests()
+mse_tests()
+get_var_mse_tests()