pspforecast

Shellfish toxicity (PSP) forecast serving package

For the current 2025 Maine PSP predictions, click here

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

2025 Results

predictions <- read_forecast(year = "2025")

Metrics

• 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
• accuracy - Measure of how many correct classifications were predicted
• cl_accuracy - Considering predictions are those that correctly predicted toxicity above or below the closure limit or not
• precision - TP/(TP+FP)
• sensitivity - TP/(TP+FN)
• specificity - TN/(TN+FP)
• f_1

## # A tibble: 1 × 10
##      tp    fp    tn    fn accuracy cl_accuracy   f_1 precision sensitivity specificity
##   <int> <int> <int> <int>    <dbl>       <dbl> <dbl>     <dbl>       <dbl>       <dbl>
## 1    32    24   247    19    0.624       0.866 0.598     0.571       0.627       0.911

2024 Season Results

predictions <- read_forecast(year = "2024")

Metrics

• 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
• accuracy - Measure of how many correct classifications were predicted
• cl_accuracy - Considering predictions are those that correctly predicted toxicity above or below the closure limit or not
• precision - TP/(TP+FP)
• sensitivity - TP/(TP+FN)
• specificity - TN/(TN+FP)
• f_1

## # A tibble: 1 × 10
##      tp    fp    tn    fn accuracy cl_accuracy   f_1 precision sensitivity specificity
##   <int> <int> <int> <int>    <dbl>       <dbl> <dbl>     <dbl>       <dbl>       <dbl>
## 1     2     4   397     7    0.717       0.973 0.267     0.333       0.222       0.990

2023 Season Results

predictions <- read_forecast(year = "2023")

Confusion Matrix

Probability of Closure-level Toxicity vs Measured Toxicity

Metrics

## # A tibble: 1 × 10
##      tp    fp    tn    fn accuracy cl_accuracy   f_1 precision sensitivity specificity
##   <int> <int> <int> <int>    <dbl>       <dbl> <dbl>     <dbl>       <dbl>       <dbl>
## 1     0     0   550     0    0.993           1   NaN       NaN         NaN           1

2022 Season Results

Confusion Matrix

Probability of Closure-level Toxicity vs Measured Toxicity

Metrics

## # A tibble: 1 × 10
##      tp    fp    tn    fn accuracy cl_accuracy   f_1 precision sensitivity specificity
##   <int> <int> <int> <int>    <dbl>       <dbl> <dbl>     <dbl>       <dbl>       <dbl>
## 1    16    20   603    12    0.799       0.951   0.5     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

## # A tibble: 1 × 10
##      tp    fp    tn    fn accuracy cl_accuracy   f_1 precision sensitivity specificity
##   <int> <int> <int> <int>    <dbl>       <dbl> <dbl>     <dbl>       <dbl>       <dbl>
## 1     2     3   463     0    0.938       0.994 0.571       0.4           1       0.994

Closure-level accuracy

Timing of initial closure-level predictions

Possible manuscript plot(s)

Last Updated

## [1] "2025-07-01"

This site is open source. Improve this page.