[GIS] I don’t find the mistake in the RandomForest-classification

errorrandom forestraster

I am trying to make a RF-classification of aerial-imagery to specify tree-species.
I extracted 3 bands and made a .csv file with columns:

band1  band2  band3  x  y  class

The band columns include the pixel-value of the point which I marked in ArcGIS and copied to the attribute-table.
x/y include the coordinates of the points.
class includes the class, such as forest, juniper…

The RF reads the .csv and classifies the bands it sees in the rasterstack.
The classifier gives back an image with the classes.

So I mark my code and run it:

A classifier-window shows up, and freezes.
5 Seconds later the console shows an error:

 Error in eval(expr, envir, enclos) : object 'band1' not found

The problem is unclear. The code is right, but the compiler reacts as if there was a typo when I run the code. But as far as I can see, there is no typo.
band1, like band2 and band3 are in the pointdata_sw_aug1.csv. It is the 1st column and has no space before or after it (" band1"). Therefore I uploaded the images.
Since my rasters are in tif-format, I recently changed the code because at first is was written by Aaron for img-files:

rlist=list.files(getwd(), pattern="img$", full.names=TRUE)

rlist=list.files(getwd(), pattern="tif$", full.names=TRUE)

The RandomForest-Package then opens a little "predict" window where it says that 31 steps are left.
It takes 5 seconds till it stops with an error-message:

Error in eval(expr, envir, enclos) : object 'band1' not found

It doesn't go on.

This should work, but gives me an error:

require(sp)
require(rgdal)
require(raster)
require(randomForest)

# Set the working directory

setwd("D:/BA-Workspace/DOP_10/orthophotos_abcd/R/test_run_R")

# Create list of rasters

rlist=list.files(getwd(), pattern="tif$", full.names=TRUE)

# Create raster-stack

rasters = stack(rlist)

# Read Raster Training-Table *.csv

intable = read.csv("D:/BA-Workspace/DOP_10/orthophotos_abcd/R/test_run_R/pointdata_sw_aug1.csv")

# Assign Training-points from Columns

myrf = randomForest(factor(class) ~ band1 + band2 + band3 , data = intable, ntree = 2500)

# Run RandomForest classifier

predict(rasters, myrf, filename="outFileName.img", type="response",
    index=1, na.rm=TRUE, progress="window", overwrite=TRUE)

my .csv:
enter image description here

enter image description here

As far as I can see there are no mistakes.
Whats wrong here?

Error-message is this (copied from console):

require(sp)
require(rgdal)
require(raster)
require(randomForest)


 >setwd("D:/BA-Workspace/DOP_10/orthophotos_abcd/R/test_run_R")


> rlist=list.files(getwd(), pattern="tif$", full.names=TRUE) 
> 

> rasters = stack(rlist)
> 
> intable = read.csv("D:/BA-Workspace/DOP_10/orthophotos_abcd/R/pointdata_sw_aug1.csv")
> 
> myrf = randomForest(factor(class) ~ band1 + band2 + band3 , data = intable, ntree =     2500)
> 
> predict(rasters, myrf, filename="outFileName.img", type="response",
+         index=1, na.rm=TRUE, progress="window", overwrite=TRUE)
**Error in eval(expr, envir, enclos) : object 'band1' not found**

Best Answer

I think your problem is related to the layer names of your raster stack ('rasters'). Make sure these are the same as in your .csv. You can get the layer names with names(rasters) and set them with names(rasters) <- c("band1", "band2", "band3")

Hope this helps TimSalabim

Related Solutions

[GIS] How to utilise single ‘randomForest’ model for several rasters in R

Welcome to SE-GIS! It is great that you have read the manual (as most people often forget to do it)! As it was said in the instruction - the names of the predictors must be the same that was used for training. The name of the predictors if you use 'raster' package are generated as 'raster_name'(without extension)+'.'+'band_number', e.g. 'imagery.tif' with 2 bands will generate names 'imagery.1' and 'imagery.2'. So if you want to use model generated for the one raster ('imagery.tif') to predict another ('another_imagery.tif') you need to supply raster with exact the same name that was used in the first place. Just rename the second raster (even temporary just for prediction) - that's all.

Alternatively you may do the same thing without 'raster' package at all. Training data have to be sampled separately and provided as a .csv-file. You will need to convert raster to data frame, rename its bands, make prediction and convert data frame into the raster again. The spatial reference information will be lost during transformation and you will need to recover it one way or another. I tried it once and I find it more tedious then renaming file for 'raster' package.

Here is the code that I used for myself (mostly taken from this article that is in Russian). SRS in this case was recovered externally (outside R). Here I had 5-band raster and training values as .csv-file (sampling of raster values was made in QGIS and results were saved as .csv). You may export to .csv the sampling results that were created using 'raster' package.

library(rgdal)
library(randomForest)

# import training data
train_new <- read.delim("train_new.csv")

#add class 2 (initial set contains only 0 and 1 values for $class)
train_new$class[ ((train_new$build_dist < 500) & (train_new$main_road < 300)) ] <- 2

train_new$class <- as.factor(train_new$class)
band_names <- names(train_new)

# start training
tt <- randomForest(class ~.,  data = train_new, ntree=1000, importance=T)


# import raster data
image <- "image.tif"
x <- new("GDALReadOnlyDataset", image)
width <- dim(x)[2]
height <- dim(x)[1]
imagedata <- data.frame(getRasterTable(x))
bnames <- band_names[1:5]

#rename raster bands to match names of attributes in .csv
names(imagedata) <- c("x", "y", bnames) 
imagedata <- imagedata[3:7] #delete unwanted 'x' and 'y' fields

# prediction
B <- predict(tt, imagedata, type="class")
BM <- as.matrix(B)

# export result to .tiff
tif_driver <- new("GDALDriver", "GTiff")
tif2 <- new("GDALTransientDataset", tif_driver, height, width, 1, 'Byte')
resmtx <- matrix(BM, width, height)
bnd1 <- putRasterData(tif2, resmtx)
tif_file <- "result.tif"
saveDataset(tif2, tif_file)
GDAL.close(tif2)
GDAL.close(tif_driver)

[GIS] Rf-Classification seems to work but gives an error 15 seconds later

You need to make sure that names(sdata@data[,3:ncol(sdata@data)]) and names(xvars) are exactly the same. Check this using

identical(names(sdata@data[,3:ncol(sdata@data)]), names(xvars))

If TRUE, your predict should run fine.

The edit related warnings/errors are irrelevant, they relate to you trying to display a SpatialPolygonsDataFrame (and S4 class object) as a standard data.frame in RStudio.

EDIT: It seems you have differences between your stack layer names and your sdata@data data frame. Make sure these are the same. If you would like to include x and y coordinates as layers to your stack (if this makes sense obviously depends on your objective) you could do it like this:

x <- coordinates(xvars)[, 1]
y <- coordinates(xvars)[, 2]

x_rst <- y_rst <- xvars[[1]]
x_rst[] <- x
y_rst[] <- y

Then you would need to add those to your stack at the appropriate position:

xvars <- stack(x_rst, y_rst, xvars)

Note also, that you have additional variables in your sdata@data data frame ("band1.1" etc). I don't know where these come from, maybe you are merging something earlier? Again, for predict() to work properly, the layers of your stack and the columns from your training data need to be identical (the names of these).

Best Answer

Related Solutions

[GIS] How to utilise single ‘randomForest’ model for several rasters in R

[GIS] Rf-Classification seems to work but gives an error 15 seconds later

Related Question