In nabilabd/hybridSA: Implemnents Hybrid Source Apportionment

knitr::opts_chunk$set(echo = TRUE)

Introduction

In the course of the analysis for the hybridSA method, we need to incorporate measurements for both Elemental and Organic carbon (EC and OC, respectively). However, in the Chemical Speciation Network (CSN) of monitoring sites, many different parameters relating to these two chemical species, are measured. Moreover, the number of measurements taken for these various parameters, does not remain constant, as we originally thought that it did.

The goal, then, in this presentation, is to inspect the kinds of measurements that are available for EC and OC, and to present a rationale for selecting the particular parameters we chose to use for our analysis.

Obtain the data

suppressPackageStartupMessages({
  library(stringr)
  library(readr)
  library(devtools)
  library(tidyr)
  library(magrittr)
  library(lubridate)
  library(ggplot2)
  library(dplyr)
  library(knitr)
})

# code used to generatte this dataset in data-raw/
agg_obs <- readRDS("~/hsa_data/agg_observations_05_12.rds")

First we extract all observations from 2005-2012 for Elemental and Organic Carbon.

# complete list of parameter codes and names in the dataset
load("~/hybridSA/data/all_param_codenames.rda")
uniq_params <- all_param_codenames %>% 
  extract(, c("ParameterName", "ParameterCode")) %>% unique

# all EC/OC observations from the AQS dataset
ocec_obs <- agg_obs %>%
  semi_join(
    filter(uniq_params, str_detect(ParameterName, "EC|OC")) # EC/OC paramaters used
    ) %>%
  rename(lat = Latitude, long = Longitude)
rm(uniq_params, agg_obs)

A quick inspection of the data can be taken:

ocec_obs %>% glimpse

To inspect trends in numbers of measurements taken over time, we don't want to aggregate by year, since that seems too coarse initially. So we group by month and year:

# combine
ocec_obs <- ocec_obs %>%
  mutate(Species = str_sub(ParameterName, 1, 2), DateLocal = ymd(DateLocal))

# group observations by month and year
ocec_obs <- ocec_obs %>%
  mutate( Year = year(DateLocal), Month = month(DateLocal), Day = 1) %>%
  unite(Date, Year, Month, Day, sep = "-")

# table used for plot above
table_ocec <- ocec_obs %>%
  select(ParameterCode, ParameterName, Species) %>% unique %>%
  mutate(Type = ifelse(str_detect(ParameterName, "TOT"), "TOT",
                       ifelse(str_detect(ParameterName, "TOR"), "TOR", "LC")
  )) %>% as.data.frame

# here, we see data relating to the various parameters being measured
kable(table_ocec)

# monthly counts by species and type
counts_ocec <- ocec_obs %>%
  mutate(Type = ifelse(str_detect(ParameterName, "TOT"), "TOT",
                       ifelse(str_detect(ParameterName, "TOR"), "TOR", "LC")
                       )) %>%
  group_by(Date, Type, Species) %>%
  summarize(num_vals = length(ArithmeticMean)) %>%
  ungroup %>%
  mutate(Date = ymd(Date))



# TOT measurements clearly decreasing
counts_ocec %>%
  ggplot(aes(Date, num_vals, color = Type)) + geom_line() + # geom_point() +
  facet_grid(. ~ Species) + ggtitle("EC/OC Measurement Types") +
  ylab("Number of values")

There are a couple of noteworthy features from this graph. One is that there clearly is not a constant trend in the number of measurements being taken. So, we find that over time, the Thermal Optical Reflectance (TOR) method is gaining popularity over the Thermal Optical Transmittance (TOT) method. As for the LC category.

Instead of simply removing the "LC" category from the above graph, we can be more specific, and examine the temporal trends of the indiividual parameter codes over time (especially since for the earlier years, duplicate measurements aren't being accounted for in the above graph, and so actual usable values are much fewer than it seems):

ocec_data <- readRDS("~/hsa_data/ocec_data.rds")
ocec_data %>% 
  group_by(Year, ParamCode, Species) %>% summarize(counts = n()) %>%
  ggplot(aes(Year, counts, color = factor(ParamCode))) +
  geom_point() + geom_line() + ylab("Number of Observations") + 
  ggtitle("EC/OC Measurements")

Here we see that 2009 was a turning point.

That's why from 2010 and afterwards, this analysis uses the TOR measurements, and as for 2009 and before, TOT is used, with corrections performed. A potential area of further investigation is the extent to which the results would vary if we used the TOR values for 2009. However, the amount of Rj values obtained would be approximately the same either way.

nabilabd/hybridSA documentation built on May 23, 2019, 12:03 p.m.