-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathuse-cases.Rmd
109 lines (71 loc) · 7.25 KB
/
use-cases.Rmd
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
---
title: "Model Use Cases"
subtitle: "Quantitative comparisons of animal diversity across space and time"
author: "Song, Xiao Ping"
date: "`r Sys.Date()`"
output: rmarkdown::html_vignette
vignette: >
%\VignetteIndexEntry{Model Use Cases}
%\VignetteEngine{knitr::rmarkdown}
%\VignetteEncoding{UTF-8}
---
```{r load biodivercity while in dev, include = FALSE}
devtools::load_all()
library("sf")
library("dplyr")
library("tmap")
```
## 1. Citywide monitoring
Models that are built using remotely sensed data can be used to predict the diversity of an animal group across time and geographical space. For example, satellite imagery processed as input predictors can be used to monitor biodiversity in the past and present.
The interactive map below shows an example of local (_Alpha_) diversity predicted for birds across [subzones](https://data.gov.sg/dataset/master-plan-2019-subzone-boundary-no-sea) in Singapore during the year 2020, at a pixel resolution of 500 m x 500 m. Depending on the level of detail required, the pixel resolution can be adjusted accordingly.
```{r echo = FALSE, warning = FALSE, message = FALSE, fig.width=2.6, fig.height = 2.0, dpi = 300, out.width="100%"}
sg_subzones <-
system.file("extdata", "sg_subzones.geojson", package="biodivercity") %>%
read_sf() %>%
relocate("subzone_n")
heatmap_dragons <-
system.file("extdata", "alpha-diversity_birds_delta-aic-2_subzones_500m.tif", package="biodivercity") %>%
terra::rast()
tmap_mode("view")
tm_basemap(c("CartoDB.Positron")) +
tm_shape(sg_subzones) +
tm_polygons(group = "Singapore Subzones",
alpha = 0) +
tm_shape(heatmap_dragons, raster.downsample = FALSE) +
tm_raster(group = "Local diversity: Birds",
title = "Number of bird species",
style = "fisher",
n = 8,
palette = "YlOrRd",
alpha = 0.6)
```
<center>
**Map: Spatial predictions of local (_Alpha_) diversity of birds across subzones in Singapore.** The pixel resolution used for predictions in this example is 500 km x 500 km.
</center>
<br>
Pixel values within the zones used in city planning can be summarised, to allow comparisons to be made between these planning units. The resulting distribution of the summarised values can subsequently be used to compare the ‘performance’ of each planning unit relative to others in the city, or to a set a benchmark/target for the desired level of 'performance' (Figure 1). For example, subzones in Singapore could be benchmarked against the mean of the distribution, as shown below:
```{r out.width = "60%", fig.align='center', dpi = 300, echo = FALSE, fig.cap = "**Figure 1: Histogram showing the distribution of values for the mean number of bird species per pixel within each of the 332 subzones in Singapore.** Subzones were assigned an arbitrary score of 0-4 based on standard deviations from the mean (i.e., performance of the 'average' subzone)."}
knitr::include_graphics("subzone-distribution-birds.png", dpi = 300)
```
<br>
If spatial predictions were made for multiple snapshots in time, benchmarking could be based on whether the average pixel value for a particular planning unit increases or decreases between two time periods. For example, if 'no net loss' in biodiversity is set as a target, a negative score could be assigned if the average pixel value is reduced, while a positive score could be assigned if the average pixel value increases.
Finally, it is worth noting that full customisation of both the pixel size and boundaries within which to summarise the pixel values can be done. This provides flexibility according to the level of analysis (e.g., geographical scale) required by the user. By summarising pixel values within zones used in city planning, animal diversity may be assessed alongside other indices also summarised at the level of these planning units, thus providing a more comprehensive view of components related to biodiversity and beyond.
<br>
---
## 2. Biodiversity in the future
Other than monitoring changes in biodiversity, there is also a need to assess future urban developments, for instance, to see if proposed designs can effectively mitigate the loss of biodiversity. But since such landscapes do not exist, snapshots of remotely sensed data can not be used. It is therefore important to carefully consider data compatibility between these different use cases when building and using the predictive models.
Urban design and planning involves the consideration of multiple design scenarios. Manually generated landscape elements (e.g., vector data for vegetation and water) may be produced from prospective designs, but the format and types of such data must be compatible with those used in the predictive models. For instance, when selecting landscape predictors to build the models, land cover classification as discrete rasters would be more compatible with manually generated data, compared to continuous rasters that cannot be feasibly calculated (e.g., spectral indices such as NDVI). Vegetation generated in design scenarios can be rasterized into discrete land cover-types (Figure 2), and used to replace the remotely sensed data within regions of interest (Figure 3). Such amendments to landscape data can be made across a site slated for urban development, and then used to make spatial predictions for that particular design scenario.
```{r out.width = "100%", fig.align='center', dpi = 300, echo = FALSE, fig.cap = "**Figure 2: Example showing how manually generated vector data of vegetation can be converted into a classified raster of vegetation types used in the predictive models.**"}
knitr::include_graphics("data-compatibility-rasterize.png", dpi = 300)
```
<br>
```{r out.width = "100%", fig.align='center', dpi = 300, echo = FALSE, fig.cap = "**Figure 3: Example showing how a classified raster of remotely sensed vegetation can be amended with the manually generated vector data (rasterised).**"}
knitr::include_graphics("data-compatibility-amend.png", dpi = 300)
```
<br>
While such data conversions may allow similar predictors (and hence models) to be used for different use cases, it should be noted that potential mismatches between different data sources may result in inaccurate predictions. Reducing the mismatch between remotely sensed and manually generated data is an important gap that should be addressed. For instance, the level of detail in design scenarios may not include the exact locations of planted trees, and their estimated canopy projection areas may vary greatly from reality after planting. Furthermore, the remotely sensed data represents a top-down view of the landscape, and the effect of multi-tiered planting is not accounted for within the landscape predictors. Collaboration between researchers and practitioners is needed to ensure that model workflows align with the data formats and outputs used in design practice, and that suitable methods are used to ensure that artificially generated datasets are both compatible and accurate to reality after implementation.
<br>
---
## References
Government of Singapore (2020). [Master Plan 2019 Subzone Boundary (No Sea)](https://data.gov.sg/dataset/master-plan-2019-subzone-boundary-no-sea). _data.gov.sg_
Released under the terms of the [Singapore Open Data Licence version 1.0](https://data.gov.sg/open-data-licence).