Preprocessing your own datasets
- Jonas Van Duijvenbode
- 1 Main Functions
- 2 Environment setup
- 3 Reprojecting a raster dataset
- 4 Selecting a single layer from a multi-layer raster
- 5 Reclassifying categorical data
- 6 Rasterizing a vector dataset (presence/absence)
- 7 Rasterizing a vector dataset with a specific attribute
- 8 Creating a distance raster from vector data
- 9 Writing the new dataset to disk
The main reason to want to use the ForestForesight package is to create your own datasets and use them to improve the model. Below we give instructions on how to do this.
Main Functions
The main functions that are required are:
Reprojection: Reprojection is the process of converting spatial data from one coordinate system to another. It allows you to align data from different sources or to display data in a desired projection for analysis or visualization.
Resampling: Resampling involves changing the cell size or resolution of a raster dataset. It can be used to increase or decrease the spatial resolution of data, often to match the resolution of other datasets or to reduce file size.
Reclassification: Reclassification is the process of reassigning values in a raster dataset based on specified criteria. It's commonly used to simplify complex data, create categorical maps from continuous data, or to recode values for analysis.
Filtering: Filtering in spatial analysis refers to selecting or highlighting specific data based on certain criteria. It can involve removing unwanted data or emphasizing particular features, often used to focus on areas of interest or to remove noise from datasets.
Clipping: Clipping is the process of extracting a portion of a spatial dataset that falls within a defined boundary. It's used to focus on a specific area of interest or to reduce the size of a dataset to a manageable extent.
Distance Calculation: Distance calculation in GIS involves computing the spatial distance between features or locations. It can be used to create buffer zones, analyze proximity relationships, or generate distance-based raster surfaces for various spatial analyses
Environment setup
library(ForestForesight)
# Make sure you have run ff_environment previously,
if not you have to assign the ff_folder yourself
ff_folder <- Sys.getenv("DATA_FOLDER")
template_folder <- list.files(file.path(ff_folder, "preprocessed", "input"),
pattern = "^[0-9]{2}[NS]_[0-9]{3}[EW]$", full.names = TRUE)[1]
template_raster <- rast(list.files(template_folder,
pattern = "\\.tif$", full.names = TRUE)[1]) For more information about the config file please refer to the configuration page here Open-Source Contribution .
Reprojecting a raster dataset
This is useful when your input raster is in a different coordinate reference system (CRS) than your template or has a different resolution or extent.
Below we give different example scripts on how to preprocess your own dataset. Please make sure you enter your own datasets when you load your dataset using rast or vect
Example: Reprojecting a land cover raster
# Load a land cover raster in a different CRS
land_cover <- rast("path/to/your/own/dataset.tif") #this is an example! Enter your own dataset
# Check CRS
print(crs(land_cover))
print(crs(template_raster))
# Reproject using nearest neighbor (best for categorical data)
land_cover_nearest <- project(land_cover, template_raster, method = "near")
# Reproject using cubic (often better for continuous data)
land_cover_cubic <- project(land_cover, template_raster, method = "cubic")
# Compare results
par(mfrow = c(1, 2))
plot(land_cover_nearest, main = "Nearest Neighbor")
plot(land_cover_cubic, main = "Cubic")Selecting a single layer from a multi-layer raster
This is common when working with satellite imagery or time series data.
Example: Selecting a single band from a Landsat image
# Load a multi-band Landsat image
landsat <- rast("landsat_image.tif") #this is an example! Enter your own dataset
# Check number of layers
nlyr(landsat)
# Select the Near-Infrared band (usually band 5 in Landsat 8)
nir_band <- landsat[[5]]
# Standardize to template
nir_standardized <- project(nir_band, template_raster, method = "cubic")
# Plot
plot(nir_standardized, main = "Near-Infrared Band")Reclassifying categorical data
This is useful for simplifying land cover classes or creating binary masks.
Example: Reclassifying a land cover raster
Rasterizing a vector dataset (presence/absence)
This is useful for creating binary masks from vector data.
Example: Creating a forest/non-forest mask from polygon data
Rasterizing a vector dataset with a specific attribute
This is useful when you want to preserve numerical or categorical information from the vector data.
Example: Rasterizing administrative boundaries with population data
Creating a distance raster from vector data
This is useful for proximity analysis, such as distance to roads or water bodies.
Example: Calculating distance to roads
Writing the new dataset to disk
The standard format is: {TILE_ID}_{DATE}_{FEATURE}.tif
Where:
{TILE_ID}is the geographic identifier (e.g., "00N_010E"){DATE}is the availability or creation date in YYYY-MM-01 format. The day numbers should always be 01{FEATURE}is a descriptive name of the raster's content. This should not contain underscores
Example: "00N_010E_2023-06-01_vegetationdensity.tif"
When creating new rasters, users should:
Use the same tile identifier as the template raster.
Choose an appropriate date that represents when the data becomes available or relevant.
Select a clear, concise name for the feature they've created.
Ensure the file is saved as a GeoTIFF (.tif extension).
The raster should be stored in the input/preprocessed/{TILE_ID} folder