pspforecast

Shellfish toxicity forecast serving package

Requirements

• R v4+

• rlang

• dplyr

• readr

• tidyr

• httr

Installation

remotes::install_github("BigelowLab/pspforecast")

Reading the forecast database

Variables:

• version - the version/configuration of the model used to make the prediction

• ensemble_n - number of ensemble members used to generate prediction

• location - the sampling station the forecast is for

• date - the date the forecast was made on

• name - site name

• lat - latitude

• lon - longitude

• class_bins - the bins used to classify shellfish total toxicity (i.e. 0: 0-10, 1: 10-30, 2: 30-80, 3: >80)

• forecast_date - the date the forecast is valid for (i.e. one week ahead of when it was made)

• predicted_class - the predicted classification at the location listed on the forecast_date (in this case 0-3)

• p_0 - class 0 probability

• p_1 - class 1 probability

• p_2 - class 2 probability

• p_3 - class 3 probability

• p3_sd - class 3 probability standard deviation

• p_3_min - class 3 minimum probability (from ensemble run)

• p_3_max - class 3 maximum probability (from ensemble run)

• predicted_class - the predicted classification

predictions <- read_forecast(year = "2023") |>
  distinct()

glimpse(predictions)

## Rows: 474
## Columns: 19
## $ version             <chr> "v0.2.0", "v0.2.0", "v0.2.0", "v0.2.0", "v0.2.0", …
## $ ensemble_n          <dbl> 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10…
## $ location            <chr> "PSP12.13", "PSP10.11", "PSP10.33", "PSP12.01", "P…
## $ date                <date> 2023-04-11, 2023-05-02, 2023-05-03, 2023-05-01, 2…
## $ name                <chr> "Lumbos Hole", "Ogunquit River", "Spurwink River",…
## $ lat                 <dbl> 43.79553, 43.25030, 43.56632, 43.73848, 43.73064, …
## $ lon                 <dbl> -69.94557, -70.59540, -70.27305, -70.04343, -70.02…
## $ class_bins          <chr> "0,10,30,80", "0,10,30,80", "0,10,30,80", "0,10,30…
## $ forecast_start_date <date> 2023-04-15, 2023-05-06, 2023-05-07, 2023-05-05, 2…
## $ forecast_end_date   <date> 2023-04-21, 2023-05-12, 2023-05-13, 2023-05-11, 2…
## $ p_0                 <dbl> 89, 98, 99, 97, 98, 96, 94, 98, 99, 99, 99, 99, 99…
## $ p_1                 <dbl> 9, 2, 1, 3, 2, 4, 5, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1,…
## $ p_2                 <dbl> 2, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,…
## $ p_3                 <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,…
## $ p3_sd               <dbl> 4.849940e-02, 4.066616e-05, 4.983161e-06, 8.274184…
## $ p_3_min             <dbl> 1.354318e-01, 3.799968e-05, 2.470616e-06, 1.027468…
## $ p_3_max             <dbl> 2.972137e-01, 1.736575e-04, 1.842741e-05, 3.850234…
## $ predicted_class     <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,…
## $ f_id                <chr> "PSP12.13_2023-04-11", "PSP10.11_2023-05-02", "PSP…

2023 Season Results

Confusion Matrix

Probability of Closure-level Toxicity vs Measured Toxicity

Metrics

Season Accuracy:

## # A tibble: 1 × 1
##   accuracy
##      <dbl>
## 1    0.990

Closure-level (Class 3) Predictions

• tp - The model predicted class 3 and the following week’s measurement was class 3
• fp - The model predicted class 3 and the following week’s measurement was not class 3
• tn - The model predicted class 0,1,2 and the following week’s measurement was in class 0,1,2
• fn - The model predicted class 0,1,2 and the following week’s measurement was class 3
• precision - TP/(TP+FP)
• sensitivity - TP/(TP+FN)
• specificity - TN/(TN+FP)

## # A tibble: 1 × 7
##      tp    fp    tn    fn precision sensitivity specificity
##   <int> <int> <int> <int>     <dbl>       <dbl>       <dbl>
## 1     0     0   404     0       NaN         NaN           1

2022 Season Results

Confusion Matrix

Probability of Closure-level Toxicity vs Measured Toxicity

Metrics

Season Accuracy:

## # A tibble: 1 × 1
##   accuracy
##      <dbl>
## 1    0.799

Closure-level (Class 3) Predictions

• tp - The model predicted class 3 and the following week’s measurement was class 3
• fp - The model predicted class 3 and the following week’s measurement was not class 3
• tn - The model predicted class 0,1,2 and the following week’s measurement was in class 0,1,2
• fn - The model predicted class 0,1,2 and the following week’s measurement was class 3
• precision - TP/(TP+FP)
• sensitivity - TP/(TP+FN)
• specificity - TN/(TN+FP)

## # A tibble: 1 × 7
##      tp    fp    tn    fn precision sensitivity specificity
##   <int> <int> <int> <int>     <dbl>       <dbl>       <dbl>
## 1    16    20   603    12     0.444       0.571       0.968

Timing of initial closure-level predictions

2021 Season Results

Confusion Matrix

Probability of Closure-level Toxicity vs Measured Toxicity

Metrics

Season Accuracy:

## # A tibble: 1 × 1
##   accuracy
##      <dbl>
## 1    0.938

Closure-level (Class 3) Predictions

• tp - The model predicted class 3 and the following week’s measurement was class 3
• fp - The model predicted class 3 and the following week’s measurement was not class 3
• tn - The model predicted class 0,1,2 and the following week’s measurement was in class 0,1,2
• fn - The model predicted class 0,1,2 and the following week’s measurement was class 3
• precision - TP/(TP+FP)
• sensitivity - TP/(TP+FN)
• specificity - TN/(TN+FP)

## # A tibble: 1 × 7
##      tp    fp    tn    fn precision sensitivity specificity
##   <int> <int> <int> <int>     <dbl>       <dbl>       <dbl>
## 1     2     3   463     0       0.4           1       0.994

Closure-level accuracy

Timing of initial closure-level predictions

Possible manuscript plot(s)

Last Updated

## [1] "2024-03-08"