8 Lab 2: Point Process Models and Landscape Metrics I

For this exercise, you will use a point dataset to explore a point process model alongside an environmental raster layer.

IMPORTANT NOTE: You may use either a GBIF call or the environmental raster code process from the tutorial, but not both! A 20 pt penalty on the whole exercise will be applied if both are used.

Learning Objectives:

  1. Pre-process spatial data, including points (species observations, events, etc.) and a covariate raster

  2. Generate a hypothesis-driven workflow for testing a point process model

  3. Execute the point process model and compare to a suite of models or the homogenous Poisson

8.1 Install Packages

library(landscapemetrics) # The landscape metrics package...duh
library(spatstat) # The point process modeling package
library(FedData) # Package for DEM and NLCD data import, among others
library(dplyr)
library(rgbif) 
library(ggplot2) 
library(raster)
library(sf)
library(terra)

8.2 Point Selection (15 pts)

Choose a point dataset with at least 300 points. In one to two paragraphs including code chunks as needed, describe your point data, including the source (3 pts), the spatial extent (3 pts), the number of points (3 pts), and some background on the biology of the points such as the species, the interactions occurring at the points, or the events (3 pts)

Plot your point data in two dimensions in a method that addresses overplotting (3 pts):

8.3 Environmental Raster Selection (15 pts)

Choose an environmental raster that you will treat categorically – this can either be one that is already in categories such as LULC or you can create your own categories from continuous data. In one to two paragraphs including code chunks as needed, describe your raster data, including the source (3 pts) and some background on why you choose this environmental layer to look at as driving the intensity of your points (3 pts).

Plot your raster data (6 pts) including x and y label axes that represent the spatial unit such as meters, degrees lat/long (3 pts):

8.4 landscapemetrics explortation (15 pts)

Using the landscapemetrics R package, explore your environmenmtal raster.

Explore one or more characteristics of the largest patch of one class of interest at the patch level (2 pts for identifying the largest patch; 3 pts for metric reporting and interpreation, no plot needed)

For the patch type of interest, calculate the percent of the raster that is that class (5 pts for correct answer):

At the landscape level create a plot showing one landscape level metric across the classes of your raster other than area (5 pts)

8.5 Hypothesis Generation (15 pts)

Based on your exploration, or your own biological insight into the system, describe a set of hypotheses that you will test.

Your competing hypotheses should include at least:

  1. The null hypothesis for your points (2 pts)

  2. A hypothesis about whether points will be more clustered or more regular than random (2 pts)

  3. A hypothesis about a nonhomogenous process not related to point interactions (2 pts)

  4. A hypothesis that includes point interactions (2 pts)

Next, describe how you will test your hypotheses 2-4 (7 pts) and your anticipated workflow. You may test them all simultaneously as a suite of candidate models or you may test each separately, up to you.

8.6 Point Process Modelling (35 pts)

Create a series of models to test your hypotheses, including: (1) Plots of predicted intensity for each hypothesis tested (15 pts, 5 pts ea.), (2) P-values or AIC comparison that properly interpret the rejection/acceptance of the null (12 pts, 4 pts ea.), and proper annotation of your process so that I can understand it (8 pts, 1 pt off for each code chunk without a description of what it is doing).

8.7 Recap (5 pts)

Write 2-3 sentences on the insights gained from your point modelling process. If you were to run a new point process model on these data, what would you do differently?