predict_rasterEngine: Model predictions (including Raster* objects)

In spatial.tools: R Functions for Working with Spatial Data

Description

Model predictions (including Raster* objects)

Usage

predict_rasterEngine(
  object,
  filename = NULL,
  na.rm.mode = TRUE,
  gain = 1,
  offset = 0,
  ncores = 1,
  debugmode = FALSE,
  verbose = F,
  ...
)

Arguments

object	a model object for which prediction is desired.
filename	Character. Output filename. If unset, will default to a temporary file.
na.rm.mode	Logical. Attempt to fix missing data, even if the model object doesn't support na.rm? Default is TRUE.
gain	Numeric. A multiplier on the outputs (default = 1).
offset	Numeric. An additive to the output (default = 0).
ncores	Numeric. Number of cores to use when predicting. Only used with randomForestSRC for now. Do not combine with foreach registered parallel engines (e.g. sfQuickInit())
debugmode	Logical. Internal debugging for the code, will be removed eventually. Default is FALSE.
verbose	Logical. Enable verbose execution? Default is FALSE.
...	additional arguments affecting the predictions produced.

Details

predict will operate normally, unless a parameter named "newdata" is found and it is of class Raster*. If this occurs, predict will use rasterEngine to perform a prediction. Currently, this works for predict.* statements in which the data to predict on is called by the parameter "newdata", the input data is in the form of a data.frame, and the output is a vector or matrix of numbers or factors.

predict will run in parallel if a cluster is registered with foreach via a do* statement, or if the user uses sfQuickInit().

Author(s)

Jonathan A. Greenberg ([e-mail])

See Also

predict

Examples

library("raster")
# This example creates a linear model relating a vegetation
# index (NDVI) to vegetation height, and applies it to a raster
# of NDVI.

# Load up a 3-band image:
tahoe_highrez <- setMinMax(
		brick(system.file("external/tahoe_highrez.tif", package="spatial.tools")))

# Determine NDVI
ndvi_nodrop <- function(GRNIR_image)
{
	red_band <- GRNIR_image[,,2,drop=FALSE]
	nir_band <- GRNIR_image[,,3,drop=FALSE]	
	ndvi <- (nir_band-red_band)/(nir_band + red_band)
	return(ndvi)
}

tahoe_ndvi <- rasterEngine(GRNIR_image=tahoe_highrez,fun=ndvi_nodrop)
names(tahoe_ndvi) <- "ndvi"

# Load up Lidar files
tahoe_lidar_highesthit <- setMinMax(
		raster(system.file("external/tahoe_lidar_highesthit.tif", package="spatial.tools")))

tahoe_lidar_bareearth <- setMinMax(
		raster(system.file("external/tahoe_lidar_bareearth.tif", package="spatial.tools")))

# Determine vegetation height:
LIDAR_height <- function(bareearth,firstreturn)
{
	height <- firstreturn-bareearth
	return(height)
}

tahoe_height <- rasterEngine(
		bareearth=tahoe_lidar_bareearth,
		firstreturn=tahoe_lidar_highesthit,
		fun=LIDAR_height)
names(tahoe_height) <- "vegetation_height"

# Stack them:
tahoe_analysis_stack <- stack(tahoe_ndvi,tahoe_height)

# Pick some random points from the stack
randomly_extracted_data <- as.data.frame(sampleRandom(tahoe_analysis_stack,size=100))

# Generate a linear model from these points:
height_from_ndvi_model <- lm(vegetation_height~ndvi,data=randomly_extracted_data)

# Apply model to NDVI image:
# Enable parallel engine to run larger images faster:
# sfQuickInit()
height_from_ndvi_raster <- predict_rasterEngine(object=height_from_ndvi_model,newdata=tahoe_ndvi)
# sfQuickStop()

spatial.tools documentation built on Feb. 14, 2020, 1:07 a.m.