---
title: "SpatStat"
format: gfm
editor: source
---

```{r setup, include=FALSE}
set.seed(02022025)
library(spatstat)
```

This demo will explore point process features in the `spatstat` package. Note the spatstat "package" actually contains a set of R packages.

For additional `spatstat` details run the following code `vignette('getstart')` in your R console.

#### 1. Generate Point Process Data

Note this will require both a $n$ and $\boldsymbol{s}$

```{r}
lambda <- 100
n <- rpois(1, lambda)

x <- rbeta(n, 1, 1)
y <- rbeta(n, 1, 1)

sim_pp <- ppp(x, y, window = owin(xrange = c(0,1), yrange = c(0,1)))
plot(sim_pp, main = 'simulated huckleberry map')
```

There are also a collection of methods in spatstat to simulation random point process, see `rpoispp` or many others. These can more closely link to our assumed generative models when considering, for example trend surfaces or the inclusion of raster data.

```{r}
CSR <- rpoispp(100)
plot(CSR)
```


__Activity 1:__ Generate a point process, where the intensity is a non-uniform function of x and y.

```{r}
pp <- rpoispp(lambda = function(x,y) { exp(4 * x + 3 *y)})
plot(pp)
```


#### 2. Summarize Point Process

```{r}
summary(sim_pp)
```

```{r}
summary(pp)
```


#### 3. Assess Spatial Structure - Does it exhibit CSP?

Remember, the $F$ function corresponds to empty space and $G$ corresponds to nearest neighbor space. The $K$ function explores the number of points in a specified radius.

```{r}
plot(envelope(sim_pp,Fest, verbose = F))
plot(envelope(sim_pp,Gest, verbose = F))
plot(envelope(sim_pp,Kest, verbose = F))
plot(envelope(CSR,Fest, verbose = F))
plot(envelope(CSR,Gest, verbose = F))
plot(envelope(CSR,Kest, verbose = F))

```

```{r}
plot(envelope(pp,Fest, verbose = F))
plot(envelope(pp,Gest, verbose = F))
plot(envelope(pp,Kest, verbose = F))
```

_Q:_ Think about the interpretation when observed lines are above and/or below the gray curves.

#### 4. How does surface differ from CSR? Smoothed density surfaces

```{r}
plot(density(sim_pp))
plot(density(CSR))
```

```{r}
plot(density(pp))

```


### Model Fitting

The `ppm` function can be used for model fitting with a point process. Consider an example that uses the location of *Beilschmiedia pendula* rainforest trees.

\vfill

```{r}
plot(bei)
```

\vfill

The `bei` dataset contains locations of trees in a tropical rain forest. The point pattern is clearly non-homogenous

\vfill
```{r}
plot(envelope(bei, Kest, verbose = F))
plot(envelope(bei, Fest, verbose = F))
plot(envelope(bei, Gest, verbose = F))

plot(density.ppp(bei))
```
\vfill


The pattern in the intensity of the trees may be related to elevation and the elevation gradient.

\vfill
```{r}
elev <- bei.extra$elev
grad <- bei.extra$grad
plot(elev)
plot(grad)
```

\newpage

The `ppm` function allows model fitting
\vfill
```{r}
tree.model <- ppm(bei ~ elev + grad);
tree.model
plot(tree.model)
```
\vfill

For more complicated models, `kppm` can be used for clustering behavior.

\vfill


## Now lets fit our simulated data

```{r}
sim_model <- ppm(pp ~ x + y);
sim_model
plot(sim_model)
```


### Marked Point Patterns
\vfill
Marked point process data contains meta data for each point. Rather than just $\boldsymbol{s}$, we have $(\boldsymbol{s},m)$.
\vfill

The marked information can either be categorical (multi-type) or continuous.

\vfill

The `lansing` data set contains locations of six types of trees.
```{r, fig.width=7, fig.height = 5}
plot(lansing, cols = 1:6)
```

```{r, fig.width=7, fig.height = 5}
plot(split(lansing))
```

\newpage

To analyze this data, consider the following model.
\vfill
```{r}
lansing.model <- ppm(lansing ~  marks - 1)
lansing.model
```

\vfill

In contrast with this model, we can also include 

```{r}
lansing.model2 <- ppm(lansing ~  marks * polynom(x,y,3))
#lansing.model2
plot(lansing.model2)
```
\vfill

Similarly continuous marked data can be included as a predictor i the `ppm` framework, potentially with interactions with spatially referenced data.

\vfill

Marked point process data can also be used for spatial-temporal point patterns, where the year corresponds to the mark.
\vfill


\newpage

### More advanced point pattern models

##### Cluster processes
Clustering is not well defined. In general the idea is that the point distances are shorter than expected. However, there "is a fundamental ambiguity between heterogeneity and clustering" (Diggle 2007).
\vfill

__Neyman-Scott Process__: This is a two stage process. 

\vfill
1. Generate parents 
\vfill
2. For each parent, generate a set of offspring
\vfill

__The shot noise processes__ are variations on the Neyman-Scott process, also with a two stage process.

\vfill

__Strauss Process__: contains a term that allows repulsion by adjusting the intensity in a vicinity of an existing point. The "hardcore" process will make the intensity 0 for any pair of points less than a specified distance $d_0$.
\vfill