In nabilabd/hybridSA: Implemnents Hybrid Source Apportionment

knitr::opts_chunk$set(echo = TRUE)

suppressPackageStartupMessages({
  library(plyr)
  library(tidyr)
  library(readr)
  library(scales)
  library(plotly)
  library(stringr)
  library(maps, warn=F)
  library(ggplot2, warn=FALSE)
  library(magrittr)
  library(devtools)
  library(lubridate)
  library(beepr)
  library(readr)
  library(pryr)
})

dev_mode()
# detach("package:dplyr", unload = T)
library(dplyr, warn = F)

Introduction

base_dir <- "/Volumes/My Passport for Mac/gpfs:pace1:/data_2005/Other_data/"

concen_year05 <- read_rds( str_c(base_dir, "concen_agg.rds"))
sa_mats05 <- read_rds( str_c(base_dir, "year_SA_matrices.rds") ) %>% 
  from_sitedatelist()

sa_mats05 <- sa_mats05 %>%  from_sitedatelist

Although the hybrid optimization has been only slightly refined from previous implementations, how the computation is performed is very different. So, to help bridge that gap, as well as introduce the method to those new to it, this document seeks to demonstrate how the functionality is actually used.

Much of the effort behind actually using this method has gone into just re-arranging the data into a suitable format, one that would facilitate the computations being performed on it.

Thus, the following is divided into two parts. The first part assumes that the data is already in a "proper" format, then shows how the method is used, as well as how the results would be interpreted. Then, the second part, for those daring enough to continue, is to illustrate how data is arranged into the format that's used.

(Of course, this assumes that all of the necessary data is already present. Otherwise, there was much code needed to wrangle and combine the data together from different locations and formats. That code, however, is likely less immediately relevant to understanding the method itself, and would only be useful to users if they obtain the raw data in the same format. Therefore, an explanation of that code used is not included here.)

This order for how the hybrid optimization is used, mirrors the thought process behind how the code ended up being organized, so hopefully it provides the reader some motivation.

This document does not aim to provide further details on what the method actually is or how it was derived. See References on papers which elaborate on it, which go into much more explanation than this introduction can.

Usage

The optimization is performed on each site-date (i.e., a given site on a given date) independently.

Two types of data are needed:

1. Concentration data
2. Sensitivity matrices

The concentration data includes: simulated concentrations, observed concentrations, and uncertainties of the observed concentrations. The sensitivity matrices, on the other hand, contain sensitivities of the chemical species to the sources which PM2.5 is being apportioned into.

There are two main functions in the hybridSA package: hybridsa, and get_optim. The former is an objective function, and is hidden from the user.

Single Site, Single Date

hyb_dir <- "/Volumes/My Passport for Mac/GRA work/hybridSA/"

source( str_c(hyb_dir, "R/hybridSA.R") )

# choose a site-date for which the Rj values converge
working_site_id <- "010730023"
working_date <- "2005-11-24"

concen_day <- concen_year05 %>% ungroup %>% 
  filter(SiteID == working_site_id, Date == working_date)
samat_day <- sa_mats05[[working_site_id]][[working_date]] %>% 
  set_names(c("Species"))

Single Site, Whole Year

All Sites, Whole Year

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