In joanaseg/mytestpkg: Demonstrates some concepts

knitr::opts_chunk$set(echo = TRUE)

First function

Here is my new function:

library(mytestpkg)
followers(3)

More Information can be found (need to look this up).

library(mytestpkg)
followers(5)

Wednesday

On Wednesday we started with a challenge to get familiar with work with all this. You can see the result in Figure \@ref(fig:cars) and Table \@ref(tab:tableone) Here is my table to practise using markdown files.

plot(cars, col="blue")

In Figure \@ref(fig:cars) you can clearly see that i managed to produce a plot. You can find interesting literature by an author called Cars here: [@madanipour2000social]. If you don't like figures, here is a table with numbers:

knitr::kable(tail(cars), caption = "My table with random cars")

I also created a new function.

library(mytestpkg)
m <- half(2)
n <- half(5)
o <- half(10)
print(c(m, n, o))

Useful for output description: paste("the value is", x) glue::glue("the value is", y)

Thursday

library(binford)
data(LRB)
knitr::kable(head(LRB))

Friday

Point pattern analyses - Spatial analysis

Interpolation methods

Point Patterns

harran <- read.table("../data/Sites_HarranPlain.csv", header = TRUE, sep = ",")
str(harran)

Spatstat

library(spatstat)
library(sp)
coordinates(harran) <- ~X+Y
proj4string(harran) <- CRS("+init=epsg:4326") #overrides projection information - be careful!!
harran <- spTransform(harran, CRSobj = CRS("+init=epsg:32637")) #326 stands for utm and 37 for 37N
str(harran)

harran_ppp <- ppp(x = harran@coords[,1], 
                  y = harran@coords[,2], 
                  window = owin(xrange = harran@bbox[1,], 
                                yrange = harran@bbox[2,]))
str(harran_ppp)
plot(harran_ppp)

harran_ppp <- ppp(x = harran@coords[,1], 
                  y = harran@coords[,2], 
                  window = owin(xrange = harran@bbox[1,], 
                                yrange = c(min(harran@coords[,2]), min(harran@coords[,2]+52000))))
# you get the warning that you throw out some points that will no longer be included in the analysis
# and some are duplicates (find out with duplicated.ppp and remove with unique.ppp)
# alternative: add "check = FALSE" to delete duplicates and points outside the area

anyDuplicated(harran_ppp)
harran_sing <- unique(harran_ppp)
# alternative: harran_unique <- harran[!is.TRUE(duplicated(harran_ppp))] #! means give me the opposite
# you could use this without "is.True" because it is a true/false answer

Spatstat has very nice descriptions, manual etc. It's the way to go if you want to work with spatial statistics and it's enough to have data in a table.

# Nearest neighbour
harran_sing_nn <- nndist(harran_sing)
str(harran_sing_nn) # distance is shown in m because we transformed the proj.
barplot(sort(harran_sing_nn))
hist(harran_sing_nn)

#create kernel density estimation
harran_dens <- density.ppp(harran_sing, sigma = mean(harran_sing_nn))
harran_dens_standard <- density(harran_sing, bw = mean(harran_sing_nn))
plot(harran_dens)
plot(harran_dens_standard)
# bw.ppl(harran_sing) or bw.diggle(harran_sing) to check which bw you should set
plot(bw.ppl(harran_sing), xlim = c(2000, 5000)) #you see the max value

#enter elevation as covariate

library(raster)
dem <- raster("../data/dem.tif") #rasterframe

#older alternative:
library(rgdal)
dem2 <- readGDAL("../data/dem.tif") #spatialgriddataframe
hist(dem2$band1)

im_dem <- as.im(as.image.SpatialGridDataFrame(as(dem, "SpatialGridDataFrame")))
# you need to transform because old spatstats can only work with this image format

harran_rhohat <- rhohat(object = harran_sing, covariate = im_dem, bw = 200)

plot(harran_rhohat)
# x-axis: elevation of our points, y-axis: intensity of points - the lower the elevation the higher
# the point density but the larger the sigma the less evident this relationship
# contrary to point correlation you have continuous datasets here because they are interpolated in the area

# compare theoretical rhohat function with original data
rho_dem <-predict(harran_rhohat)
plot(rho_dem)
diff_rho <- harran_dens - rho_dem
plot(diff_rho)
# first order effect map

challenge: produce random points that have the same density like the global density of our area (density = points per area)

#poisson process pattern - spatial randomness

set.seed(123)
harran_rpoispp2 <- rpoispp(lambda = harran_sing$n/area.owin(harran_sing$window), win = harran_sing$window)
set.seed(123)
harran_rpoispp3 <- rpoispp(intensity(harran_sing), win = Window(harran_sing))
set.seed(123)
harran_rpoispp4 <- rpoispp(ex = harran_sing)

plot(harran_sing)
points(harran_rpoispp2, col="red")
points(harran_rpoispp3, col="blue")
points(harran_rpoispp4, col="green")

#set.seed use random start number and restore data points for future use

To put it on github you need to change 4 files: - description (insert all imports), - travis.yml (addons - apt - packages, script) - docker (line 11-15) - README.Rmd (copied from Daniels repo)

Friday afternoon

Second Order effects

# g-Function compares NN-distance
harran_g <- Gest(harran_sing)
str(harran_g)
# r is distance radius, km is empirial data
plot(harran_g)
# blue line is poisson function, black line is real data, blue and green are data with edge correction
# at low distance we have more points than in theory so seems to be clustering, around 1000 m it seems random,
# in upper distance seems clustering again
harran_ge <- envelope(harran_sing, fun = "Gest")
# takes random points 99 times to calculate g-function
plot(harran_ge)
# shows that distribution of points in harran plain is completely random
harran_ge_large <- envelope(harran_sing, fun = "Gest", nsim = 999)
plot(harran_ge_large)
# the more values you use, the wider the gray area gets

harran_f <- Fest(harran_sing)
plot(harran_f)
harran_k <- Kest(harran_sing)
plot(harran_k)
# inhomogeneous data with covariate influence can be tested against Ginhom, Finhom, Kinhom
# e.g. if you know you have a first order effect, you predict the density image

harran_gi <- Ginhom(harran_sing, lambda = predict(harran_rhohat))
plot(harran_gi)