Hierarchical Clustering

Hierarchical clustering by complete linkage to create a dendrogram showing multivariate clustering for water chemistry by site in Santa Barbara, CA.

code
analysis
hierarchical-clustering
Author

Elmera Azadpour

Published

March 7, 2021

To access data, html and Rmd/qmd files:

  • the stream chemistry data can be found in the data folder of this repo; the Rmd and html files, denoted as a4-task1, can be found in the src folder
  • the qmd format can be found here

Load packages

library(tidyverse)
library(here)
library(janitor)
# Packages for cluster analysis:
library(NbClust)
library(cluster)
library(factoextra)
library(dendextend)
library(ggdendro)

Introduction:

  • To perform hierarchical clustering by site, we’ll begin by making a data frame that has a single summary row per site (e.g. based on means from all observations at that site), then we will calculate the euclidean distance before performing complete linkage agglomerative hierarchical clustering.

Read in data

stream_chem <- read_csv(here("posts", "2021-03-07-hierarchical-clustering","sbc_lter_registered_stream_chemistry.csv")) %>% 
  clean_names() %>% 
  na_if(-999.0) %>% 
  group_by(site_code) %>% 
  summarise(across(nh4_u_m:spec_cond_u_spercm, mean, na.rm = TRUE)) %>% 
  drop_na()
Rows: 19390 Columns: 12
── Column specification ────────────────────────────────────────────────────────
Delimiter: ","
chr   (1): site_code
dbl  (10): nh4_uM, no3_uM, po4_uM, tdn_uM, tdp_uM, tpc_uM, tpn_uM, tpp_uM, t...
dttm  (1): timestamp_local

ℹ Use `spec()` to retrieve the full column specification for this data.
ℹ Specify the column types or set `show_col_types = FALSE` to quiet this message.

Scale the data

# scale the numeric variables
stream_chem_scaled <- stream_chem %>% 
  dplyr::select(2:11) %>% 
  scale()

# make rownames the site code names from original data set
rownames(stream_chem_scaled) <- stream_chem$site_code

# Compute dissimilarity values (Euclidean distances):
stream_euc_distance <- dist(stream_chem_scaled, method = "euclidean")

# Check out the output:
# View(euc_distance)

Perform hierarchical clustering by complete linkage with stats::hclust()

# Hierarchical clustering (complete linkage)
stream_hc_complete <- hclust(stream_euc_distance, method = "complete" )

# Plot it (base plot):
plot(stream_hc_complete, cex = 0.6, hang = -1)

Convert to class: dendogram

stream_dend_complete <- as.dendrogram(stream_hc_complete)

Plot using ggdendrogram(), a ggplot wrapper:

# ggplot 
ggdendrogram(stream_dend_complete, 
             rotate = TRUE)+
  labs(x = "Santa Barbara area watershed site code") +
  ylab(NULL)+ 
  theme_bw() + 
  theme(panel.border = element_blank(), panel.grid.major = element_blank(),
panel.grid.minor = element_blank(), axis.line = element_line(colour = "black"))

Data & Metadata Source:

  • SBC LTER: Stream chemistry in the Santa Barbara Coastal drainage area, ongoing since 2000 Creators: Santa Barbara Coastal LTER, & Melack, John M Citation: Santa Barbara Coastal LTER and J. Melack. 2019. SBC LTER: Land: Stream chemistry in the Santa Barbara Coastal drainage area, ongoing since 2000 ver 16. Environmental Data Initiative. https://doi.org/10.6073/pasta/67a558a24ceed9a0a5bf5e46ab841174.