First results SML sampling exercise
load("../output/first_test.RData")
stac500a = rast("~/serena/data/sml/europe/stack500.tiff")
sol <- rast("~/serena/data/sml/europe/soilRegion.tiff")
library(RColorBrewer)
JRC data
Here the maps of the input variables
The Soil Regions of the European Union and Adjacent Countries 1 : 5 000 000
nb.cols <- 20
mycolors <- colorRampPalette(brewer.pal(8, "Paired"))(nb.cols)
r2 = as.factor(sol)
plot(r2, col = mycolors)
## Warning: [barplot] a sample of 0.4% of the raster cells were used to estimate frequencies
To fill the JRC table
Here is the results of the different scenarios
cbind.data.frame(
scenario = c("A1","A2","A4","A5","A6",
"B1","B2","B3"),
nSites = c(testA1$NbSample,testA2$NbSample,testA4$NbSample,testA5$NbSample,
testA6$NbSample,
testB1$NbSample,testB2$NbSample,testB3$NbSample)
)
## scenario nSites
## 1 A1 4674
## 2 A2 3062
## 3 A4 627
## 4 A5 3932
## 5 A6 5517
## 6 B1 1820
## 7 B2 2818
## 8 B3 4936
comparison with 3 different sets of starting samples
Here we compare the computed required sample sizes using NUTS1 only as Domain and with different starting samples:
- A 5k sample selected by clhs based on 80k SI sample as proposed by JRC.
- A 80k SI sample
- A 20k SY sample
SI = random sampling SY = systematic sampling
The SY sample gave a larger required sample size.
There is a very large differences in computed required sample sizes.
## [1] "CLHS 5k sample" "879"
## [1] "A 80k SI sample" "1251"
## [1] "A 20k SY sample" "3038"
Here are the maps
For the 5K starting sample
For the 80K starting sample
For a grid as starting sample