Species Distribution Models with GIS & Machine Learning in R

Build a species distribution model by mapping predicted distributions and selecting climate, topography, land cover, and population data from SRT, Aegeus, and WorldClim.

Access GBIF species occurrence data, filter for human observations with coordinates, download as CSV, and clean the dataset ready for building species distribution models with habitat layers.

Retrieve rhinoceros hornbill occurrence data from GBIF directly in R, filter by taxon key and country, and visualize latitude and longitude points across Malaysia and Southeast Asia.

Explore alternative sources of species geo-location data beyond GPA and GBR, including bird-related platforms like eBird and iNaturalist, to access observations, checklists, and hotspots for field study.

Extract species geo-location data from multiple databases in R using the spocc package, query species lists, and convert results into a data frame with latitude and longitude.

Access climate data and country outlines in R for species distribution modeling, pulling bio climatic variables (bio1–bio19) at 2.5-minute resolution and clipping to country or state boundaries.

Read in elevation data from RTM 59 and SRT 60 rosters, merge adjacent rosters into a single mosaic, and use the mean for overlapping areas for species distribution models.

Derive topographic products from elevation data by computing slope, aspect, and hill shade. Convert lat/long rasters to UTM for accurate analysis and use these products in species distribution modeling.

Conclude section 2 by outlining data sources for species distribution modeling, including occurrence data from the G-B website and GBI API, plus climate, topography, and optional land cover data.

Ensure raster predictors share the same extent, resolution, and coordinate reference system for reliable species distribution models in R, using a Peninsular Malaysia case study.

Learn how to ensure all geolocation data share the same coordinate reference system by converting rasters between UTM and lat long using projectRaster, preparing data for species distribution models.

Clip rasters to a defined extent and crop all rasters to Peninsular Malaysia using the crop function from the raster package and a Malaysia shape file.

Change the land use raster’s spatial resolution to match others by resampling with a common interpolation method in the raster package, producing a coarser, consistent raster for analysis.

Explore basic data visualizations for hornbill species distribution by building interactive leaflet maps and static ggplot maps in R, using mopper, map_leaflet, and ggplot2 to visualize locations across Southeast Asia.

Conclude section 3 with GIS operations on Droste data: CRS conversions between lat long and utm, clipping rasters via shapefiles, and resizing resolution to prepare for species distribution models.

Explore classical species distribution modeling techniques using presence data to map habitat suitability and ecological niches, including climate-based methods, the domain model, and maximum entropy (MaxEnt) approaches.

Discover bioclim, a climate envelope model using presence-only data to map a species' climate envelope and predict distribution under climate change scenarios.

Assess model performance by generating pseudo absences as background data and evaluating presence data with AUC, using a bio climatic model and predictors in R.

Implement the Maxent interface in R to perform species distribution modeling, including downloading Maxent, placing the jar in the right folder, and setting up Oddjob and Java for the interface.

Learn to use Maxent in R for presence-only species distribution modeling with raster predictors such as altitude and temperature. Evaluate performance (AUC 0.84) and predict habitat suitability using background points.

Use the red package to perform maxent analyses for species distribution modeling with multiple species, building predictor stacks from presence data. Evaluate with 20 percent test data and auc metrics.

Learn the domain sdm in r, using presence data and a predictor raster stack with background points, evaluated by cross-validation to assess suitability.

Conclude section 4 by reiterating bioclim and maxent, showing how to build models with our data, and preview domain-based modeling with Gayford data splitting and area under the curve evaluation.

Apply machine learning to build species distribution and habitat suitability maps, using logistic regression, k-nearest neighbors, support vector machines, random forests, and gradient boosting, with cross-validation and AUC evaluation.

Learn to prepare presence and absence data for machine learning based species distribution models by compiling raster predictors, extracting values, and generating pseudo absences with ecologically informed backgrounds.

Prepare presence/absence data with land cover, climate, and topography predictors; read and inspect it, then create a reproducible 75/25 train/test split in R using Garrett before building habitat suitability models.

Explore habitat suitability with logistic regression for species distribution maps, using GLMs in R, binary presence-absence data, and predictor importance, training/testing splits, and AUC evaluation.

Apply support vector machines in R for species distribution mapping, including data preparation, training and testing split, cross-validation, and kernel options (rbf and poly), with variable importance and auc assessment.

Apply the k nearest neighbor algorithm to build a species distribution model, using a 75/25 training/testing split and noting roughness and slope as key predictors for presence-absence.

Random forests provide robust classification for species distribution and habitat suitability maps, using presence and background data to reveal predictor importance and partial dependence of minimum temperature, roughness, and precipitation.

Explore the gradient boosting machine (gbm), a powerful tree-based method rivaling random forests, using tenfold cross-validation on presence–absence data. Assess predictor importance and partial dependence to map habitat suitability.

Evaluate species distribution models with binary classification metrics using an R package, applying AUC, confusion matrices, and entropy loss to compare random forest and other classifiers.

Conclude section five by summarizing pre-processing to derive absence or pseudo-absence data and the theory behind common algorithms such as random forests, GBMs, and SVMs for habitat suitability maps.

demonstrate obtaining digital elevation models within R using the elevator package, by defining an extent and zoom level to produce a merged elevation raster at 250 m or 2.5 km.

Identify the spatial distribution and concentration of geo locational points by applying 2D density mapping in R, visualizing forest fire locations in Australia with ggplot and world map overlays.

Introduces the leaflet package for interactive geographic visualizations, showing how to load base maps, add tiles, plot markers and circles from lat/long data, and customize popups.

Learn how to use GitHub and GitHub Desktop to turn local data projects into a portfolio, create repositories, and push code to GitHub for collaboration.

Explore a time-lapse view of the Brazilian Amazon showing deforestation driven by infrastructure development, including roads and hydroelectric dams, from 1984 to 2014 and the rapid land-use changes.

Convert a 12-class categorical map (0–11) in qgis to meaningful classes by using unique values, applying a color ramp, and editing labels for each category to produce a readable map.