GetNearestCovariate.R issues (NAs, run time, projection)

[LeaDambly]

I've got a few issues with GetNearestCovariate.R. If I understand correctly, then the function is meant to get the spatially closest covariate value for a data (or integration) point. I am experiencing an issue where I end up with NAs assigned to the points. Sometimes as high as~20% of the points. There's also the issue of long run-time and the function only working for Lat/Long data. I have made some changes which would improve on this. Using the original function on 717 points with 3 covariates leads me to to get 10 points with NAs and a run time of approx. 50 seconds.
If I use the code below instead, I get 0 NAs and a run time of approx, 0.1 secs. Note that the nn2 function technically doesn't return NAs but instead "If there are no neighbours then nn.idx will contain 0 and nn.dists will contain 1.340781e+154 for that point". Not getting any of those either though.

` require(RANN)
require(dplyr)
if(class(points)=="SpatialPointsDataFrame") {
points <- points@coords
}
covnames <- names(covariates@data)
pointsXY <- as.data.frame(points)
names(pointsXY) <- c("X", "Y")
covXY <- cbind(covariates@coords, covariates@data)
covXY$ID <- seq(nrow(covXY))
closest <- nn2(covXY[c("X", "Y")], pointsXY[c("X", "Y")], k = 1)
closest <- as.data.frame(closest) %>%
rename(ID = nn.idx)
joined <- inner_join(closest, covXY, by = "ID") %>%
dplyr::select(-c(nn.dists, X, Y, ID))
allpts <- bind_cols(pointsXY, joined)

points.df <- SpatialPointsDataFrame(coords = allpts[c("X", "Y")],
data = allpts[covnames],
proj4string = CRS(proj4string(covs)))
return(points.df)`

I haven't tested this substantially (it works for my data so that doesn't necessarily mean anything) but I would definitely recommend using nn2 instead of spDistsN1. If only to save a lot of time.

[oharar]

Thanks. I'm not surprised there was something faster than my code, so I appreciate you taking the time to improve it.

I'll pass this on to Philip, who is working on the package (updating it to work with INLAbru).