Remote Sensing of Biodiversity

.title[
# Remote Sensing of Biodiversity
]
.subtitle[
## From Landscapes to Genes
]
.author[
### Jasper Slingsby
]
.author[
### & many others!
]
.date[
### MCB, UCT, 18 March 2026
]

---

### We want to measure biodiversity everywhere, all the time...

---

## It's a rapidly growing field

<img src="images/cavenderbares2020.png" alt="" width="75%" style="display: block; margin: auto auto auto 0;" />
.footnote[Cavender-Bares et al. 2020]

]

---

<img src="images/bioscape_new.png" alt="" width="100%" style="display: block; margin: auto;" />
.left[.footnote[...and the Cape is currently the epicentre of this endeavour | www.bioscape.io]]

---

## BioSCape: Biodiversity Survey of the Cape

- NASA's first biodiversity-focused field campaign
- `\(>\)` 150 scientists and conservation practitioners
- 19 teams (mixed US, RSA, other)
- fundamental and applied science
- terrestrial and aquatic
- 3 planes
- 6 instruments (2 x V-SWIR imaging spectrometers, hyperspectral thermal, multispectral (RGB + NIR) and 2 x LiDAR)

]

.pull-right[
<img src="images/bioscape_planes_portrait.png" alt="" width="95%" style="display: block; margin: auto;" />
.footnote[www.bioscape.io]
]

---

---

## The technology

---

text-color: white

---

## Sensor types

Active vs passive

Multispectral vs hyperspectral

Variation within each type around spectral range and resolution, spatial resolution, revisit time, etc

.right-column[
<img src="images/pettorelli2018_sensors.png" alt="" width="100%" style="display: block; margin: auto;" />
]

---

## Multispectral vs hyperspectral (imaging spectrometers)

---

Proteaceae shrubs (dark green) surrounded by low shrubs, forbs and graminoids at Silvermine, TMNP.

]

.pull-right[
<img src="images/purkis_klemas2011_lidar.png" alt="" width="100%" style="display: block; margin: auto;" />

Light detection and ranging (LiDAR) uses active remote sensing by firing a "laser beam" to measure topography and the vertical structure of vegetation.

---

### These are just numbers and not useful without ground observations

---

## But what are we trying to measure?

---

Each facet provides its own challenges and opportunities and require different methods of remote sensing.

]

.pull-right[
<img src="images/Noss_Biodiversity.png" alt="" width="100%" style="display: block; margin: auto;" />
.footnote[Noss 1990, _Conservation Biology_]
]

---

An advantage of remote sensing is that it can directly measure the structure, composition and function of biodiversity...

<img src="images/skidmore2021_fig1.png" alt="" width="120%" style="display: block; margin: auto;" />
.footnote[Skidmore et al. 2021]

]

...at least from the scale of individuals up...

]

---

##Multispectral remote sensing

---

##Productivity and Seasonality

---

##Land cover (and change)

---

##Land cover change detection

---

##Land cover change time series

---

## But there are challenges and limitations...

---

## What about hyperspectral data?

---

## Spectral unmixing can detect "spectral signatures"

Map species/types based on their reflectance of the electromagnetic spectrum!

Given a library of spectral signatures of different species and land cover types (endmembers), spectral unmixing infers the composition of each pixel from the possible mixes of endmembers. This gives a cover map of the majority endmember for each pixel (as here) and the fraction of each endmember for all pixels (next slide).

]

---

## Spectral unmixing can detect "spectral signatures"

---

Imaging spectroscopy ("hyperspectral" remote sensing) allows direct measurement of leaf traits.

]

.pull-right[
<img src="images/peninsula_hyperspec.png" alt="" width="100%" style="display: block; margin: auto;" />
]

---

---

## Potential uses of trait maps?

Using trait maps to infer the legacy impacts of invasive alien plants.

]

Using traits to map landscape flammability.

]

---

## Very high resolution multispectral data?

---

An iconic critically endangered species...

]

]

---

Manual mapping from 50cm RGB aerial imagery (2013)

.pull-right[
<img src="images/CederbergMap.jpeg" alt="" width="92%" style="display: block; margin: auto;" />

13,419 cedar tree localities!!!

---

## The Clanwilliam Cedar

Field validation:

- Typically only adult trees or clumps with >4m `\(^2\)` live canopy were reliably identified from aerial imagery, so many smaller individuals are missed.
- The majority of trees were dead, but these were not apparent in the imagery.
- Trees that had recently died (dead leaves still present) outnumbered live trees by a ratio of 2:1.

Useful baseline and suggestive of rapid change, but couldn't be used to investigate change over time.

---

Cedar decadal resurvey with 2022 imagery...

.pull-right[
<img src="images/cedar_points_plot_legend_inside.png" alt="" width="85%" style="display: block; margin: auto;" />

]

---

## The Clanwilliam Cedar

---

## The Clanwilliam Cedar - correlates of mortality?

Mortality ~ Environmental variables?

BUT WAIT!

This is the basis of species distribution modelling...

Since trees only die where they occur, all we're going to get is the distribution of trees, not the distribution of tree mortality...

---

## The Clanwilliam Cedar - correlates of mortality?

A Bayesian thinned spatial point process (SPP) modelling framework that couples occurrence with a mortality process to formally treat mortality events across the landscape as a spatial process.

Developed for modelling animal mortality within their known habitat use based on telemetry data.

- Joint model of occurrence and mortality
- 30m digital elevation model (DEM)
- CapeNature fire records
- CHELSA downscaled reanalysis climate data

]

---

## The Clanwilliam Cedar - correlates of mortality?

---

## The Clanwilliam Cedar - correlates of mortality?

]

---

## The Clanwilliam Cedar

### Conclusions and limitations

- Alarmingly high mortality (>17% in a decade)!
- Mortality is associated with greater numbers of fires and faster warming

BUT...

- Climate data were coarse (1km)
- Still to include field validation

Sadly, ±90% of the Cederberg burnt in the past year, so likely that many more trees have died...
]

]

---

## Mapping cedars with Lidar?

]

.pull-right[
<img src="images/bioscape_cedars_driehoek_lvis.png" alt="" width="100%" style="display: block; margin: auto;" />

]

---

## Mapping cedars with hyperspectral imagery?

---

## Mapping genotypes with hyperspectral imagery?

A bit of a holy grail for remote sensing of biodiversity is to be able to map genetic variation, which would be a game changer for conservation and evolutionary biology.

.pull-right[
<img src="images/cedarvariants.jpeg" alt="" width="100%" style="display: block; margin: auto;" />

---

## Mapping cedars with hyperspectral imagery?

.pull-right[
<img src="images/cedarunmixing.png" alt="" width="98%" style="display: block; margin: auto;" />

---

## Mapping _**milkwoods**_ with hyperspectral imagery

]

- Easier to measure (leaf and canopy)
- Larger distribution (and potential for variation)
- Extensive hyperspectral imagery from BioSCape

---
class: center, middle

# Thanks!

More at www.plantecolo.gy and www.bioscape.io