I have a dataframe similar in structure to the one created below:
id <- rep(c("a", "b", "c", "d"), each = 3)
date <- seq(as.Date("2019-01-30"), as.Date("2019-02-10"), by="days")
lon <- c(-87.1234, -86.54980, -86.234059, -87.2568, -87.65468, -86.54980, -86.234059, -86.16486, -87.156546, -86.234059, -86.16486, -87.156546)
lat <- c(26.458, 26.156, 25.468, 25.157, 24.154, 24.689, 25.575, 25.468, 25.157, 24.154, 26.789, 26.456)
data <- data.frame(id, date, lon, lat)
data <- data %>% arrange(id, date)
I would like to measure the distance between consecutive points grouped by id. I do not want a distance matrix, which is why I refrain from using raster::pointDistance. I tried separating each unique id into its own sf dataframe (in reality I have ~400 ids so I kind of have to separate for the actual calculation due to the size) and using the following code:
#put rows for each id in their own dataframes
un1 <- unique(data$id)
for(i in seq_along(un1))
assign(paste0('id', i), subset(data, id == un1[i]))
#create point distance function
pt.dist <- function(dat){dat$pt.dist <- st_distance(dat, by_element=TRUE)
return(dat)}
#run function across every dataframe in working environment
e <- .GlobalEnv
nms <- ls(pattern = "id", envir = e)
for(nm in nms) e[[nm]] <- pt.dist(e[[nm]])
When I run this, all I get is a geometry column with lon and lat listed in a pair. I have also tried segclust2d::calc_distance like below:
distance <- function(dat){calc_dist(dat, coord.names = c("lon", "lat"), smoothed = FALSE)}
for(nm in nms) e[[nm]] <- distance(e[[nm]])
which returns a column where the distances are all 0 meters.
Any help would be greatly appreciated!
>Solution :
geosphere::dist* support this. The most-accurate is distVincentyEllipsoid (though it may be slower with larger data), followed by distVincentySphere and distHaversine. Its return value is in meters.
dplyr
library(dplyr)
data %>%
group_by(id) %>%
mutate(dist = c(NA, geosphere::distVincentyEllipsoid(cbind(lon, lat)))) %>%
ungroup()
# # A tibble: 12 x 5
# id date lon lat dist
# <chr> <date> <dbl> <dbl> <dbl>
# 1 a 2019-01-30 -87.1 26.5 NA
# 2 a 2019-01-31 -86.5 26.2 66334.
# 3 a 2019-02-01 -86.2 25.5 82534.
# 4 b 2019-02-02 -87.3 25.2 NA
# 5 b 2019-02-03 -87.7 24.2 118175.
# 6 b 2019-02-04 -86.5 24.7 126758.
# 7 c 2019-02-05 -86.2 25.6 NA
# 8 c 2019-02-06 -86.2 25.5 13744.
# 9 c 2019-02-07 -87.2 25.2 105632.
# 10 d 2019-02-08 -86.2 24.2 NA
# 11 d 2019-02-09 -86.2 26.8 291988.
# 12 d 2019-02-10 -87.2 26.5 105423.
base R
We can get to the same thing with ave. Because it only iterates over a single column, we pass row-indices as the first argument to be grouped. Because it coerces the return values to be the same class as the first argument, we convert the row-indices to numeric.
data$dist2 <- ave(
as.numeric(seq_len(nrow(data))), # values to use in calc
data$id, # grouping variable(s)
FUN = function(i) c(NA, geosphere::distVincentyEllipsoid(data[i, c("lon", "lat")]))
)
data
# id date lon lat dist2
# 1 a 2019-01-30 -87.12340 26.458 NA
# 2 a 2019-01-31 -86.54980 26.156 66334.13
# 3 a 2019-02-01 -86.23406 25.468 82534.47
# 4 b 2019-02-02 -87.25680 25.157 NA
# 5 b 2019-02-03 -87.65468 24.154 118175.40
# 6 b 2019-02-04 -86.54980 24.689 126757.93
# 7 c 2019-02-05 -86.23406 25.575 NA
# 8 c 2019-02-06 -86.16486 25.468 13743.74
# 9 c 2019-02-07 -87.15655 25.157 105631.82
# 10 d 2019-02-08 -86.23406 24.154 NA
# 11 d 2019-02-09 -86.16486 26.789 291988.42
# 12 d 2019-02-10 -87.15655 26.456 105422.87
Internally, the second call to the FUN function passed i=c(4,5,6) for the "b" id group. Those numbers do not need to be consecutive; in fact, one strength of ave over other group-processing functions is that it always returns in the same order as the input, so it is safe to reassign its value back to the original frame.