# Exploratory assessment of differences between ESRI's flow direction

# output and GRASS Terraflow's flow direction output, with focus here on

# the SFD (D8) Terraflow output, which is (I think) basically the same

# as what ESRI uses.

#

# The input DEM in all cases is what I gather to be the Oregon case

# study DEM, which is this 3x3 degree, 3" resolution, smoothed (somehow?

# by someone?) SRTM:

#   jupiter:~organisms/flow/experimental/srtmv41_smth

# 

# The ESRI flowdirection raster used here is what I gather was produced by 

# layers.aml, which applies the GRID flowdirection function to the

# smoothed DEM to produce this output:

#   jupiter:~organisms/flow/experimental/flowdir

#

# I converted both of the above rasters to GeoTIFFs to bring them over

# to my machine to work with.

#

# I then generated the two Terraflow output flow direction grids (SFD

# and MFD) using the commands contained in grass-terraflow.sh.

#

# Other resources:

# * ESRI ArcGIS 9.3 Flow Direction documentation

#   http://webhelp.esri.com/arcgisdesktop/9.3/index.cfm?topicname=flow%20direction

# * GRASS terraflow documentation

#   http://grass.osgeo.org/grass64/manuals/html64_user/r.terraflow.html

#

# Jim Regetz

# NCEAS

# Created 16-Jun-2011

# data directory (local to xander)

datadir <- "~/media/temp/dem/new/flow"

# load data

# ESRI flowdirection output (see notes above)

or.flowdir.esri <- raster(file.path(datadir, "flowdir_OR_esri.tif"))

# GRASS terraflow (MFD) output 

or.flowdir.grass.mfd <- raster(file.path(datadir,

    "flowdir_OR_grass.tif"))

# GRASS terraflow (SFD) output 

or.flowdir.grass.sfd <- raster(file.path(datadir,

    "flowdir_OR_grass_sfd.tif"))

# create raster indicating where ESRI and terraflow(SFD) differ

flow.disagreement <- or.flowdir.esri

flow.disagreement[] <- ifelse(values(or.flowdir.esri !=

    or.flowdir.grass.sfd), 1, NA)

# illustrated differences

plot(or.hs, col=grey.colors(100))

mp <- prod(dim(flow.disagreement))

plot(flow.disagreement, add=TRUE, col="blue", legend=FALSE, maxpixels=mp)

# now isolate differences in cases where terraflow is not zero (i.e.,

# places calculated as pits?

flow.disagreement[or.flowdir.grass.sfd==0] <- NA

plot(flow.disagreement, add=TRUE, col="red", legend=FALSE, maxpixels=mp)

# notice what happens at the edges...

# in northernmost row, esri routes most cells east or west (i.e, along

# border) plus some north, and a few other misc directions 

table(as.matrix(or.flowdir.esri)[1,])

##    1    2    4    8   16   32   64 

## 1574    4    1    3 1600    1  425 

# whereas terraflow SFD routes all cells north

table(as.matrix(or.flowdir.grass.sfd)[1,])

##   64 

## 3608 

# (analogous patterns for the other 3 raster edges)

# quantify proportion in concordance, excluding the edges

edges <- c(1, 3608)

mean(as.matrix(or.flowdir.esri)[-edges, -edges] ==

    as.matrix(or.flowdir.grass.sfd)[-edges, -edges])

## [1] 0.9086855

# ...and what proportion are either in concordance or have 0 values

# associated with terraflow SFD? (means pits?)

mean((as.matrix(or.flowdir.esri)[-edges, -edges] ==

  as.matrix(or.flowdir.grass.sfd)[-edges, -edges]) |

  as.matrix(or.flowdir.grass.sfd)[-edges, -edges]==0)

## [1] 0.9755789