stars3.Rmd

---
title: "3. stars tidyverse methods"
author: "Edzer Pebesma"
output:
  html_document:
    toc: true
    toc_float:
      collapsed: false
      smooth_scroll: false
    toc_depth: 2
vignette: >
  %\VignetteIndexEntry{3. stars tidyverse methods}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE, collapse = TRUE, dev = "png")
ev = TRUE
```

This vignette shows how some of the tidyverse verbs can be used on
`stars` objects.

The `stars` and `tidyverse` packages are loaded by
```{r}
library(stars)
library(dplyr)
```
Methods now available for class `stars` are
``` {r}
methods(class = "stars")
```

We will work with a three-band section of a landsat image:
```{r}
system.file("tif/L7_ETMs.tif", package = "stars") %>%
	read_stars -> x
x
```

## `slice`
`slice` slices a sub-array out of the cube; this is done by specifying
the dimension on which to act, and the slice number. 
```{r}
x %>% slice(band, 6) -> x6
x6
```
It returns a lower-dimensional array if a single element is selected
along the slice dimension.

## `filter`
Similar to `slice`, `filter` selects on dimensions but evaluates their values
rather than their index: in
```{r}
x %>% filter(x > 289000, x < 291000, band > 3) -> x7
x7
```
the subarray is created based on the x _coordinate_ values.

Note that `filter` converts the object to a `tbl_cube`, and uses
the `dplyr` `filter` method for `tbl_cube` objects. This has the
limitation that `stars` objects with rectilinear, curvilinear or
simple feature geometries cannot be handled. For such objects, using
regular `[` selection or using `st_crop` may be an alternative.

## `pull`

`pull` pulls out an array from a stars object:
```{r}
x %>% pull(1) -> x8
class(x8)
dim(x8)
```

## `mutate`
```{r}
x %>% mutate(band2 = 2 * L7_ETMs.tif) -> x2 
x2
```

## `select`

`select` selects an attribute, or a set of attributes:
```{r}
x2 %>% select(band2) -> x9
x9
```

## `geom_stars`

`geom_raster` is a `ggplot2` geom function that accepts `stars`
objects as its `data` argument and

* sets up the raster or vector spatial coordinates as plot dimensions, and
the first attribute as the fill variable
* allows for downsampling (without choosing a suitable downsampling level)
* chooses between using `geom_raster`, `geom_rect` and `geom_sf` depending on
whether the geometry is regular, rectilinear or has vector geometries

An example use is
```{r}
library(ggplot2)
library(viridis)
ggplot() + 
  geom_stars(data = x) +
  coord_equal() +
  facet_wrap(~band) +
  theme_void() +
  scale_fill_viridis() +
  scale_x_discrete(expand = c(0, 0)) +
  scale_y_discrete(expand = c(0, 0))
```