Added new episodic filter that examines the top 10 peaks each year an…

…d determines if they occur in winter, and what the mean SpC is vs the overall mean

Changed two of the filters
min_perc_peaks_winter -> 0.35
min_perc_winter_higher -> 0.65

4_EpisodicSalinization.R

@@ -25,9 +25,9 @@ p4_targets <- list(
   # our criteria for exhibiting "episodic" patterns in winter.
   tar_target(p4_ts_sc_peak_summary, 
              summarize_salt_peaks(p4_ts_sc_peaks, 
-                                  min_perc_peaks_winter = 0.40, 
-                                  min_perc_diff = 0.10,
-                                  min_perc_winter_higher = 0.75)),
+                                  min_perc_peaks_winter = 0.35, 
+                                  # min_perc_diff = 0.10,
+                                  min_perc_winter_higher = 0.65)),
   tar_target(p4_episodic_sites, filter(p4_ts_sc_peak_summary, is_salt_site)$site_no)
 
 )

4_EpisodicSalinization/src/summarize_sc_peaks.R

@@ -55,6 +55,23 @@ summarize_salt_peaks <- function(ts_peak_data, winter_months = c(12,1,2,3),
     mutate(perc_winter_max_higher = n_winter_higher / n_years) %>% 
     select(site_no, perc_winter_max_higher)
 
+  # Calculate of the 10 largest peaks each winter, how many occur in winter ('winterPeaks)
+  # Also calculate the different between the peakSpc and the mean Spc
+  salt_peaks = ts_peak_data %>% 
+    mutate(is_winter = month(dateTime) %in% winter_months) %>% 
+    group_by(site_no, year = year(dateTime)) %>% 
+    mutate(winterSum = sum(is_winter)) %>% 
+    filter(winterSum > 60) %>% # need sufficient winter data (>60 days)
+    arrange(desc(SpecCond)) %>% 
+    slice(1:10) %>% 
+    summarise(winterPeaks = sum(is_winter), peakSpc = mean(SpecCond, na.rm = TRUE)) %>% 
+    left_join(ts_peak_data %>% group_by(site_no, year = year(dateTime)) %>% 
+                summarise(meanSpc = mean(SpecCond, na.rm = TRUE))) %>% 
+    group_by(site_no) %>% 
+    summarise_all(mean) %>% 
+    select(-year)
+
+
   salt_sites_info <- ts_peak_data %>% 
     # Create a winter flag - if month in (12,1,2,3)
     mutate(is_winter = month(dateTime) %in% winter_months) %>% 
@@ -86,11 +103,14 @@ summarize_salt_peaks <- function(ts_peak_data, winter_months = c(12,1,2,3),
   salt_sites_info %>% 
     # Join in the data about winter maximums
     left_join(salt_sites_max_info, by = 'site_no') %>% 
+    left_join(salt_peaks, by = 'site_no') %>% 
     mutate(is_salt_site = 
              # More than `min_perc_peaks_winter` percent of global peaks must occur in winter
              peaks_perc_winterYes > min_perc_peaks_winter & 
              # The non-winter and winter mean SpC must be more than `min_perc_diff` percent different
-             sc_perc_diff > min_perc_diff &
+             # sc_perc_diff > min_perc_diff &
+             winterPeaks > 6 &
+             peakSpc >= 200 + meanSpc &
              # Maximum annual SpC should occur in winter more than `` percent of the years
              perc_winter_max_higher > min_perc_winter_higher)
 

6_DefineCharacteristics/src/apply_randomforest.R

@@ -19,12 +19,14 @@ apply_randomforest <- function(site_attr_data, mtry = NULL, ntree = NULL, seed =
   if(is.null(mtry)) mtry <- floor(sqrt(ncol(site_attr_data) - 1))
 
   set.seed(seed)
-  randomForest(
+  rfMod = randomForest(
     site_category_fact ~ .,
     data = site_attr_data,
     mtry = mtry, 
     importance = TRUE)
 
+  return(rfMod)
+
 }
 
 #' @title Tune random forest hyperparameters