Combining data from multiple surveys

class: center, middle, inverse, title-slide

.title[
# Combining data from multiple surveys
]
.subtitle[
## IMR sdmTMB workshop
]
.author[
### 
]
.date[
### May 23–25 2023
]

---

# Combining surveys is a common challenge

* Ideally overlapping in time - space

* Otherwise some set of parameters need to be shared

---

# Combining surveys is a common challenge

---

# Combining surveys is a common challenge

.small[
* Gaps in time may be smoothed if:
* Assumption: equal catchability between surveys (scales intercept)
* Reasonable to use smooth or time-varying coefficients  
]

```r
fit <- sdmTMB(density ~ s(year), ...)

fit <- sdmTMB(density ~ 1, 
              time_varying = ~ 1,
              time="year",...)

fit <- sdmTMB(density ~ 1, 
              time="year",
              spatiotemporal = "ar1",
              ...)
```

---

# Combining surveys is a common challenge

* Ok if assumption of equal catchabilty is equal. 
<img src="09-combining-surveys_files/figure-html/unnamed-chunk-3-1.png" width="700px" style="display: block; margin: auto;" />

---

# Overlapping surveys and catchability

---

# Overlapping surveys and catchability

* Interactions between survey (or gears) and other fixed effects

```r
fit <- sdmTMB(density ~ fyear + survey, ...)

fit <- sdmTMB(density ~ fyear * survey, ...)

fit <- sdmTMB(density ~ s(year, by = survey), ...)
```

---

# Overlapping surveys and catchability

* Example: combining trawl and acoustic surveys for pelagic species

* Surveys overlap in space/time, but sample different portions of population

---

# Simulation testing

* Assume species has some depth affinity 
<img src="09-combining-surveys_files/figure-html/unnamed-chunk-6-1.png" width="700px" style="display: block; margin: auto;" />

---

# Simulation testing

.small[
* Layer on 2 surveys (acoustic, trawl) which might not overlap
* Can we recover depth relationship?? 
]
<img src="09-combining-surveys_files/figure-html/unnamed-chunk-7-1.png" width="700px" style="display: block; margin: auto;" />

---

# Model in sdmTMB

```r
fit <- sdmTMB(
  observed ~ log_depth + log_depth2+ survey,
  data = surveys, 
  mesh = mesh,
  spatiotemporal = "iid",
  time="year"
)
```

---

![](images/simulations.png)

---

# What sdmTMB cannot (yet) do

* Combine surveys using multiple responses (counts, densities)
  * Delta / hurdle model is exception 
  
* Link models with different meshes