Building a prediction grid

class: center, middle, inverse, title-slide

.title[
# Building a prediction grid
]
.subtitle[
## IMR sdmTMB workshop
]
.author[
### 
]
.date[
### May 23–25 2023
]

---

# Why do we need a prediction grid?

* Generate some index of abundance

* Making pretty maps!

---

# Approach 1: at locations of data

```r
pred <- predict(fit)
```

* Why? evaluating residuals

---

# Approach 2: mesh locations

* Predict to locations of mesh vertices

* These are where random effects are estimated 
  * spatial fields
  * spatiotemporal fields
  * spatially varying coefficients

* Why? Diagnostics / understanding mesh & uncertainties

See [https://github.com/pbs-assess/sdmTMB/discussions/230](https://github.com/pbs-assess/sdmTMB/discussions/230)
---

# Approach 3: regular grid

* see `qcs_grid` as example (4km grid)
* generally built from survey design, excludes land
* Why? Standardized index of abundance across survey domain

See `?get_index`
<img src="08-prediction-grid_files/figure-html/unnamed-chunk-2-1.png" width="700px" style="display: block; margin: auto;" />

---

# Approach 4: grid based on the locations of observations

* Things to consider: spatial grid cell size?
* How many time steps should have observations?
* Why? Domain grid not available, applicable
  * e.g. rock / reef associated fishes

Code: [https://github.com/pbs-assess/sdmTMB-teaching/blob/main/imr-2023/exercises/coastal-survey-ex/coastal-survey-index-south.R#L205](https://github.com/pbs-assess/sdmTMB-teaching/blob/main/imr-2023/exercises/coastal-survey-ex/coastal-survey-index-south.R#L205)

---

# Tips and tricks

* Any prediction grid or dataframe will need to have values for all covariates
  * Excamples: year, depth, temperature
  * Random effects = exception, can be NA