BIRDIE DST: Model summary

Introduction

The species distributions module (DST) of the BIRDIE pipeline has four main steps: data preparation, model fitting, model diagnostics and model summary. See the BIRDIE: basics and BIRDIE: species distributions vignettes for general details about BIRDIE and about the DST module, respectively. In this vignette, we will go through the different tasks that are performed during the step of the DST module: model summary.

The main function used for summarising a model fit is ppl_summarise_occu(). This is a ppl_ function, and therefore it doesn’t do much processing itself (see BIRDIE: basics if this is confusing), but it does call the right functions to do the work.

This step is relatively simple. Here we need to conduct two tasks: predict occurrence and detection probabilities for all pentads in South Africa from a fitted model and summarise these predictions.

Predicting from a model fit

predictSpOccu() is used for predicting from a model fit and it is basically a wrapper around spOccupancy::predict.PGOcc(). However, it has a few data preparation steps to make sure the data we pass on to the predict function has the same variables and on the same scale as the data we used for fitting the model. Model are fitted using only those pentads that were visited in any given year, but we want to predict for all pentads in South Africa. To maintain the same scale, we make use of the covariate scale information that we stored in the model fit object (see BIRDIE DST: model fitting vignette).

The output of predictSpOccu() is a list with two elements containing posterior predictive samples for \(\psi\) (psi, probability of occurrence) and \(p\) (probability of detection) for each pentad.

Summarising predictions

In this step, we use the function summariseSpOccu() to extract the 0.025, 0.5 and 0.975 quantiles of the predictive samples for \(\psi\) and \(p\) obtained from predictSpOccu(). We will store these quantiles to display on the BIRDIE website rather than all of the posterior predictive samples.

We also compute the realized occupancy from the posterior predictive samples and the data. The realized occupancy is the probability of occurrence conditional on the observed data, such that

\[P[occu | obs = 1] = 1\] \[P[occu | obs = 0] = \frac{\psi q}{(1 - \psi + \psi q)}\] where \(obs = 1\) when the species was detected at a site on any visit, and \(obs = 0\) when the species was not detected at a site on any visit, \(\psi\) is the probability of occurrence estimated by our model and \(q = \prod_{i=1}^N (1 - p_i)\), with \(p_i\) being the probability of detection on visit \(i\) and \(N\) the total number of visits to a given site.

All these summaries are stored in a file called analysis/output/occu_pred_spOccupancy_YYYY_sp_code.csv, where YYYY is the year of the data we are fitting a model to and sp_code is the species code. These files will be exported to the database for storage and display on the BIRDIE website.

If no model could be fitted because of lack of enough data, then, we use the raw data to create the prediction file mentioned above The file still has the exact same structure, but now \(\psi\) and \(p\) will be NA for all pentads (because they could not be calculated). Raw data will be treated as realized occupancy and no confidence limits will be estimated. The function that performs these tasks is createPredFromAbap().