class: center, middle, inverse, title-slide # Leveraging Camera Data from Multiple Projects ## 📷<br><em>For Improved Precision</em> ### Marcus Becker ### David Huggard ### 2020-06-12 --- ## Outline 1. [Estimating Density](#density) 1. [Example Study Area](#area) 1. [Those CIs are pretty wide ...](#CI) 1. [Could we find more data?](#moar) 1. [Improved precision!](#precision) --- class: center, middle name: density # Estimating Density --- ## The ABMI uses (a version of) the REST model REST = <ins>**R**</ins>andom <ins>**E**</ins>ncounter <ins>**S**</ins>taying <ins>**T**</ins>ime -- ### Central Tenets: 1. **Images taken at discrete intervals are converted into a measure of continuous time, in order to calculate how long the animal was in the camera field of view.** --- ## The ABMI uses (a version of) the REST model .pull-left[ <img src="moose1.gif" height=300 width=500> ] .pull-right[ <img src="plot1.gif" height=265 width=300> ] --- ## The ABMI uses (a version of) the REST model ### Central Tenets: 1. Images taken at discrete intervals are converted into a measure of continuous time, in order to calculate how long the animal was in the camera field of view. 1. **Unlike quadrats, cameras do survey a fixed area. We estimate an *effective detection distance* (EDD) for each combination of species, habitat type, and season.** -- <div align="center"> <img src="fov.PNG" height=150> </div> --- ## The ABMI uses (a version of) the REST model ### Central Tenets: 1. Images taken at discrete intervals are converted into a measure of continuous time, in order to calculate how long the animal was in the camera field of view. 1. Unlike quadrats, cameras do survey a fixed area. We estimate an *effective detection distance* (EDD) for each combination of species, habitat type, and season. <br> `$$Density = \frac{\sum(N~*~T_f)}{A_f~*~T_0}$$` --- ### Resources -- ABMI report: [Animal Density from Camera Data](https://abmi.ca/home/publications/501-550/516) -- Code base: [github.com/ABbiodiversity/mammals-camera](https://github.com/ABbiodiversity/mammals-camera) -- Notable literature: + Rowcliffe et al (2008) - *Estimating Animal Density Using Camera Traps Without the Need for Individual Recognition* + Nakashima et al (2018) - *Estimating Animal Density Without Individual Recognition Using Information Exclusively Derived From Camera Traps* --- ## Oops, a small snag + We want to be able to report on differences in species abundance either between areas, or over time. -- + In order to make inferences about density in an area of interest, cameras are placed in a random systematic manner so that they are **surveying a representative sample** of the larger area. -- + However, **cameras sample a tiny area** relative to regions we're often interested in. Because of this, sampling distributions are often ... extremely skewed. -- + **More camera traps** (and/or longer sampling times) would improve the precision of density estimates for an area of interest. --- class: center, middle name: area # An Example Area --- ### Wildlife Management Units (WMUs) Here is a map of them in Alberta: -- .pull-right[ <!-- --> ] --- ### Wildlife Management Units (WMUs) Here is a map of them in Alberta: .pull-left[Let's pick out a subset: + Algar Lake (WMU 519) + Winefred Lake (WMU 517) + Crow Lake (WMU 512) + Heart Lake (WMU 515) ] .pull-right[ <!-- --> ] --- class: hide-logo ### Where are ABMI cameras? (...ish) <iframe src="map1.html" width="800" height="500"> </iframe> --- ### (Relative) Density Estimates .left-code[ ```r library(abmi.camera.extras) cameras <- ace_get_cam( aoi = ab_wmu_selection, group_id = WMUNIT_NAM, crs = 4326) density <- ace_join_dens( x = cameras, species = c( "Moose", "Black Bear")) ``` ] .right-plot[ <img src="bear.png" width="40%" style="display: block; margin: auto;" /> <img src="moose.png" width="40%" style="display: block; margin: auto;" /> ] --- ### (Relative) Density Estimates .left-code[ ```r library(abmi.camera.extras) cameras <- ace_get_cam( aoi = ab_wmu_selection, group_id = WMUNIT_NAM, crs = 4326) density <- ace_join_dens( x = cameras, species = c( "Moose", "Black Bear")) ``` ] .right-plot[  ] --- ### (Relative) Regional Density .left-code[ ```r summary <- ace_summarise_dens( x = density, group_id = WMUNIT_NAM, conflevel = 0.9 ) ``` ] .right-plot[  ] --- class: center, middle name: CI # Those CI's are pretty wide ... --- class: center, middle name: moar # Could we find more data? -- ### [WildTrax Data Discover](https://dev.wildtrax.ca/home/data-discover.html) --- class: hide-logo ### More Cameras! <iframe src="map1.html" width="800" height="500"> </iframe> --- class: hide-logo # More Cameras! We can incorporate the REST-derived density estimates from other camera projects as long as they were placed on the landscape in a **random (or systematic-random) sampling design**. -- <!-- --> --- class: hide-logo # More Cameras! We can incorporate the REST-derived density estimates from other camera projects as long as they were placed on the landscape in a **random (or systematic-random) sampling design**. <!-- --> --- class: hide-logo # More Cameras! We can incorporate the REST-derived density estimates from other camera projects as long as they were placed on the landscape in a **random (or systematic-random) sampling design**. <!-- --> --- class: center, middle name: precision # Improved Precision! --- ### (Recalculated) (Relative) Regional Density <img src="plot2.gif" height=400 width=475, align="center"> --- ### Takeaways 1. Camera deployments from different projects (with different research objectives) can sometimes be used for a common goal. - In this case, to contribute to a **joint regional density estimate**. 2. However, in this case, these projects must share the similar methodological underpinnings with respect to sampling strategy. --- ## Questions? <br> Feel free to email me anytime! [mabecker@ualberta.ca](mabecker@ualberta.ca) <br> ## Thanks!