---
title: "Upscaling your data"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Upscaling your data}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r, include = FALSE}
knitr::opts_chunk$set(
  collapse = TRUE,
  comment  = "#>"
)
```


`funbiogeo` provides an easy way to upscale your site data to a coarser resolution.
The idea is that you have any type of data at the site level (diversity metrics, 
environmental data, as well as site-species data) that you would like to work or
visualize at a coarser scale. The aggregation process can look daunting at first
and be quite difficult to run.
We explain in this vignette, how to do so through the `fb_aggregate_site_data()`
function. We'll detail three use cases: the first through aggregating arbitrary
site-level data, the second focused on aggregating the site x species object,
and the last aggregating functional diversity metrics.


## Aggregating arbitrary site-data


```{r setup}
library("funbiogeo")
```

Let's import our site by locations object, which describes the geographical locations of sites.


```{r import-site-locations}
data("site_locations")

site_locations
```

These sites are a collection of regular spatial polygons at a resolution of 0.5° over Western Europe.

For each site, we want to compute the species richness. 

```{r compute-richness}
# Import site x species data ----
data("site_species")

# Compute species richness ----
species_richness <- fb_count_species_by_site(site_species)

head(species_richness)
```

Before going any further let's map these original values with the function
`fb_map_site_data()`.


```{r map-original-richness, dev='png'}
fb_map_site_data(site_locations, species_richness, "n_species")
```

Now, let's say that our next analyses require to work at a coarser resolution. We need to
aggregate site data on a new spatial grid (object `SpatRaster` from the 
[`terra`](http://cran.r-project.org/package=terra) package). Let's import this coarser raster.


```{r import-grid}
# Import study area grid ----
coarser_grid <- system.file("extdata", "grid_area.tif", package = "funbiogeo")
coarser_grid <- terra::rast(coarser_grid)

coarser_grid
```

We will aggregate the site data (resolution of 0.5°) to this new coarser raster (resolution of 0.83°) 
with the function `fb_aggregate_site_data()`. This function requires the following arguments:

- `site_locations`: the site x locations object
- `site_data`: a `matrix` or `data.frame` containing values per sites to aggregate on the provided grid `agg_grid`. Can have one or several columns (variables to aggregate). The first column must contain sites names as provided in the example dataset `site_locations`
- `agg_grid`: a `SpatRaster` object (package `terra`). A raster of one single layer, that defines the grid along which to aggregate
- `fun`: the function used to aggregate sites values when there are multiple sites in one cell

Let's aggregate our species richness values on this grid.


```{r upscaled-richness}
# Upscale to grid ----
upscaled_richness <- fb_aggregate_site_data(
  site_locations = site_locations,
  site_data      = species_richness[ , 1:2],
  agg_grid       = coarser_grid,
  fun            = mean
)

upscaled_richness
```

The result of this function is a `SpatRaster` where the cells contain the
average values of species richness aggregated over the coarser grid.


```{r map-upscaled-richness, dev='png'}
fb_map_raster(upscaled_richness)
```


## Aggregating site-species data to a coarser spatial scale

Now that we've learned how to aggregate arbitrary data at the site scale over a
spatial scale. We're going to use our provided example named `site_species` on
Western European mammals at a resolution of 0.5° to get new sites from a grid
with pixels of 0.83° of resolution.

As shown in the part above, we'll need three objects: `site_species`, which
describes the species present across sites; `site_locations`, which gives the
spatial locations of sites; and `agg_grid` which is a `SpatRaster` object
defining the coarser grid.

We'll use the previously defined object to run our example. To aggregate 
the presence-absence of species within each pixel of the new grid, we'll use
the `max()` function (as the `fun` argument). As such, coarser pixels which
contains a mix of presence and absence of certain species, we'll be considered
as having the species present.

```{r upscale-site-species}
site_species_agg <- fb_aggregate_site_data(
  site_locations,
  site_species,
  agg_grid = coarser_grid,
  fun = max
)
```

The return object is a `SpatRaster` as well but can be transformed easily in a
data.frame to follow back with the regular analyses provided in `funbiogeo`.
The new object contains one layer for each aggregated variable, i.e. here, one
per species.

```{r show-upscale-site-species}
site_species_agg
```

We can visualize both maps for a single species to see the difference:

```{r plot-map-upsacale-site-species}
library("ggplot2")

single_species <- merge(
  site_locations, site_species[, 1:2], by = "site", all = TRUE
)

finer_map <- ggplot(single_species) +
  geom_sf(aes(fill = as.factor(sp_001))) +
  labs(fill = "Presence of sp_001", title = "Original resolution (0.5°)")

coarser_map <- fb_map_raster(site_species_agg[[1]]) +
  scale_fill_binned(breaks = c(0, 0.5, 1)) +
  labs(title = "Coarser resolution (0.83°)")

patchwork::wrap_plots(finer_map, coarser_map, nrow = 1)
```


### Obtaining back a site x species `data.frame`

Now we obtained a raster of aggregated site-species presences. However, the 
other functions of `funbiogeo` don't play well with raster data. They need
data.frames to work well. We can do this through the specific function
`as.data.frame()` in `terra` (make sure to check the dedicated help page that
specifies all the additional arguments with `?terra::as.data.frame`).

```{r upscaled-site-species-df}
# Use the 'cells = TRUE' argument to index results with a new cell column
# corresponding to the ID of the coarser grid pixels
site_species_agg_df <- terra::as.data.frame(site_species_agg, cells = TRUE)

site_species_agg_df[1:4, 1:4]

colnames(site_species_agg_df)[1] <- "site"
```

With this, we're ready to reuse all of `funbiogeo` functions to work on these
coarser data. You can proceed similarly to aggregate the ancillary site-related
data, to use them in the rest of the analyses.


## Aggregating site-functional diversity data

Because `funbiogeo` focuses on the functional biogeography workflow, we'll
explore in this section how to aggregate the results for a functional
biogeography function. First, we'll show the example with the CWM of body mass
then we'll show an example with the functional diversity using the `fundiversity`
package.

### Coarser CWM of body mass

To compute the CWM we'll use the internal function `fb_cwm()`.

```{r cwm-body-mass}
site_cwm <- fb_cwm(site_species, species_traits[, 1:2])

head(site_cwm)
```

Now we can aggregate the CWM of body mass at coarser scale using
`fb_aggregate_site_data()` as done in the previous section:

```{r upscaled-cwm}
colnames(site_cwm)[3] <- "adult_body_mass"

upscaled_cwm <- fb_aggregate_site_data(
  site_locations,
  site_cwm[, c(1, 3)],
  coarser_grid
)

upscaled_cwm
```

We can then map the CWM using the `fb_map_raster()` function:

```{r map-upscaled-cwm}
fb_map_raster(upscaled_cwm) +
  scale_fill_continuous(trans = "log10")
```


### Coarser FRic through `fundiversity`

In a similar fashion as in the
[introduction vignette to `funbiogeo`](funbiogeo.Rmd) in this section we'll
compute the Functional Richness using two traits across our example dataset.

```{r compute-fric, eval = require("fundiversity")}
# Get all species for which we have both adult body mass and litter size
subset_traits <- species_traits[
  , c("species", "adult_body_mass", "litter_size")
]
subset_traits <- subset(
  subset_traits, !is.na(adult_body_mass) & !is.na(litter_size)
)

# Transform trait data
subset_traits[["adult_body_mass"]] <- as.numeric(
  scale(log10(subset_traits[["adult_body_mass"]]))
)

subset_traits[["litter_size"]] <- as.numeric(
  scale(subset_traits[["litter_size"]])
)

# Filter site for which we have trait information for than 80% of species
subset_site <- fb_filter_sites_by_trait_coverage(
  site_species, subset_traits, 0.8
)
subset_site <- subset_site[, c("site", subset_traits$species)]

# Remove first column and convert in rownames
rownames(subset_traits) <- subset_traits[["species"]]
subset_traits <- subset_traits[, -1]
rownames(subset_site) <- subset_site[["site"]]
subset_site <- subset_site[, -1]

# Compute FRic
site_fric <- fundiversity::fd_fric(
  subset_traits, subset_site
)

head(site_fric)
```

We can now follow a similar upscaling process as in the previous sections

```{r upscale-fric, eval = require("fundiversity")}
agg_fric <- fb_aggregate_site_data(site_locations, site_fric, coarser_grid)

fb_map_raster(agg_fric)
```