---
title: "Evaluation of example datasets from Attachment 1 to the US EPA SOP for the NAFTA guidance"
author: "Johannes Ranke"
date: "`r Sys.Date()`"
output:
  html_document:
    toc: true
    toc_float:
      collapsed: false
    mathjax: null
    fig_retina: null
references:
- id: usepa2015
  title: Standard Operating Procedure for Using the NAFTA
    Guidance to Calculate Representative Half-life Values and Characterizing
    Pesticide Degradation
  author:
    - family: US EPA
  type: report
  issued:
    year: 2015
  url: https://www.epa.gov/pesticide-science-and-assessing-pesticide-risks/standard-operating-procedure-using-nafta-guidance
vignette: >
  %\VignetteIndexEntry{Example evaluation of FOCUS Laboratory Data L1 to L3}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

# Introduction

In this document, the example evaluations provided in Attachment 1 to the SOP
of US EPA for using the NAFTA guidance [@usepa2015] are repeated using mkin.
The original evaluations reported in the attachment were performed using PestDF
in version 0.8.4. Note that PestDF 0.8.13 is the version distributed at the US
EPA website today (2019-02-26).

The datasets are now distributed with the mkin package.

```{r, include = FALSE}
library(knitr)
opts_chunk$set(tidy = FALSE, cache = FALSE, fig.height = 7)
library("mkin", quietly = TRUE)
```

# Examples where DFOP did not converge with PestDF 0.8.4

In attachment 1, it is reported that the DFOP model does not converge for these
datasets when PestDF 0.8.4 was used. For all four datasets, the DFOP model can
be fitted with mkin (see below). The negative half-life given by PestDF 0.8.4
for these fits appears to be the result of a bug. The results for the other
two models (SFO and IORE) are the same.

## Example on page 5, upper panel

```{r p5a}
p5a <- nafta(NAFTA_SOP_Attachment[["p5a"]])
plot(p5a)
print(p5a)
```

## Example on page 5, lower panel

```{r p5b}
p5b <- nafta(NAFTA_SOP_Attachment[["p5b"]])
plot(p5b)
print(p5b)
```

## Example on page 6

```{r p6}
p6 <- nafta(NAFTA_SOP_Attachment[["p6"]])
plot(p6)
print(p6)
```

## Example on page 7

```{r p7}
p7 <- nafta(NAFTA_SOP_Attachment[["p7"]])
plot(p7)
print(p7)
```

# Examples where the representative half-life deviates from the observed DT50

## Example on page 8

For this dataset, the IORE fit does not converge when the default starting values
used by mkin for the IORE model are used. Therefore, a lower value for the rate
constant is used here.

```{r p8}
p8 <- nafta(NAFTA_SOP_Attachment[["p8"]], parms.ini = c(k__iore_parent = 1e-3))
plot(p8)
print(p8)
```

# Examples where SFO was not selected for an abiotic study

## Example on page 9, upper panel

```{r p9a}
p9a <- nafta(NAFTA_SOP_Attachment[["p9a"]])
plot(p9a)
print(p9a)
```

In this example, the residuals of the SFO indicate a lack of fit of this model,
so even if it was an abiotic experiment, the data do not suggest a simple
exponential decline.

## Example on page 9, lower panel

```{r p9b}
p9b <- nafta(NAFTA_SOP_Attachment[["p9b"]])
plot(p9b)
print(p9b)
```

Here, mkin gives a longer slow DT50 for the DFOP model (17.8 days) than
PestDF (13.5 days). Presumably, this is related to the fact that
PestDF gives a negative value for the proportion of the fast degradation
which should be between 0 and 1, inclusive. This parameter is called
f in PestDF and g in mkin. In mkin, it is restricted to the interval from
0 to 1.

## Example on page 10

```{r p10}
p10 <- nafta(NAFTA_SOP_Attachment[["p10"]])
plot(p10)
print(p10)
```

Here, a value below N is given for the IORE model, because the data
suggests a faster decline towards the end of the experiment, which
appears physically rather unlikely in the case of a photolysis study.
It seems PestDF does not constrain N to values above zero, thus
the slight difference in IORE model parameters between PestDF and
mkin.

# The DT50 was not observed during the study

## Example on page 11

```{r p11}
p11 <- nafta(NAFTA_SOP_Attachment[["p11"]])
plot(p11)
print(p11)
```

In this case, the DFOP fit reported for PestDF resulted
in a negative value for the slower rate constant, which is
not possible in mkin. The other results are in agreement.

# N is less than 1 and the DFOP rate constants are like the SFO rate constant

In the following three examples, the same results are obtained with mkin as
reported for PestDF. As in the case on page 10, the N values below 1 are deemed
unrealistic and appear to be the result of an overparameterisation.


## Example on page 12, upper panel

```{r p12a}
p12a <- nafta(NAFTA_SOP_Attachment[["p12a"]])
plot(p12a)
print(p12a)
```

## Example on page 12, lower panel

```{r p12b}
p12b <- nafta(NAFTA_SOP_Attachment[["p12b"]])
plot(p12b)
print(p12b)
```

## Example on page 13

```{r p13}
p13 <- nafta(NAFTA_SOP_Attachment[["p13"]])
plot(p13)
print(p13)
```

# DT50 not observed in the study and DFOP problems in PestDF

```{r p14}
p14 <- nafta(NAFTA_SOP_Attachment[["p14"]])
plot(p14)
print(p14)
```

The slower rate constant reported by PestDF is negative, which is not
physically realistic, and not possible in mkin. The other fits give the same
results in mkin and PestDF.

# N is less than 1 and DFOP fraction parameter is below zero

```{r p15a}
p15a <- nafta(NAFTA_SOP_Attachment[["p15a"]])
plot(p15a)
print(p15a)
```

```{r p15b}
p15b <- nafta(NAFTA_SOP_Attachment[["p15b"]])
plot(p15b)
print(p15b)
```

In mkin, only the IORE fit is affected (deemed unrealistic), as the fraction
parameter of the DFOP model is restricted to the interval between 0 and 1 in
mkin. The SFO fits give the same results for both mkin and PestDF.

# The DFOP fraction parameter is greater than 1

```{r p16}
p16 <- nafta(NAFTA_SOP_Attachment[["p16"]])
plot(p16)
print(p16)
```

In PestDF, the DFOP fit seems to have stuck in a local minimum, as mkin finds
a solution with a much lower $\chi^2$ error level. As the half-life from the
slower rate constant of the DFOP model is larger than the IORE derived half-life,
the NAFTA recommendation obtained with mkin is to use the DFOP representative
half-life of 8.9 days.

# Conclusions

The results obtained with mkin deviate from the results obtained with PestDF
either in cases where one of the interpretive rules would apply, i.e. the
IORE parameter N is less than one or the DFOP k values obtained with PestDF are
equal to the SFO k values, or in cases where the DFOP model did not converge,
which often lead to negative rate constants returned by PestDF.

Therefore, mkin appears to suitable for kinetic evaluations according to the
NAFTA guidance.

# References