One common type of task with spatio-temporal data is to match nearby -sites. For example, we may want to verify the location of an old list of -stations with current stations, or we may want to match the data from -different data sources. Some of these matches only concern the spatial -dimension, while others require temporal agreement.
-This vignette introduces how to spatially and spatio-temporally match -sites with the cubble structure with two examples. The first example -pairs traditional weather stations with nearby automated stations in New -South Wales, Australia. This exercise only concerns the matching based -on spherical distance between stations. The next example pairs the river -level recorded by the river gauges with the precipitation recorded by -the nearby weather station in Victoria, Australia.
-Spatial matching -
-Bureau of Meteorology collects water -data from river gauges and this includes variables: electrical -conductivity, turbidity, water course discharge, water course level, and -water temperature. In particular, water level will interactive with -precipitation from the climate data since rainfall will raise the water -level in the river. Here is the location of available weather station -and water gauges in Victoria:
-
In cubble, match_sites() houses
-match_spatial() and match_temporal(). For a
-spatial-only matching, you can use
-match_sites(temporal_matching = FALSE) or simply
-match_spatial().
Any matching requires two datasets in the cubble and we call them
-major and minor. Major and minor dataset
-differs from how distance is calculated. Spatial matching calculates the
-spherical distance using the Vincenty formula and this distance is
-calculated from each site in the major dataset is
-to every site in the minor dataset.
-res_sp <- match_spatial(climate_vic, river,
- spatial_n_group = 10, return_cubble = TRUE)
-(res_sp <- res_sp[-c(5, 8)] %>% bind_rows())
-#> # cubble: key: id [16], index: date, nested form, [sf]
-#> # spatial: [144.5203, -38.144913, 148.4667, -36.128657], WGS 84
-#> # temporal: date [date], prcp [dbl], tmax [dbl], tmin [dbl]
-#> id long lat elev name wmo_id ts type geometry
-#> <chr> <dbl> <dbl> <dbl> <chr> <dbl> <list> <chr> <POINT [°]>
-#> 1 ASN0… 145. -37.0 290 rede… 94859 <tibble> clim… (144.5203 -37.0194)
-#> 2 4062… 145. -37.0 NA CAMP… NA <tibble> rive… (144.5403 -37.01512)
-#> 3 ASN0… 148. -37.7 62.7 orbo… 95918 <tibble> clim… (148.4667 -37.6922)
-#> 4 2222… 148. -37.7 NA SNOW… NA <tibble> rive… (148.451 -37.70739)
-#> 5 ASN0… 147. -38.1 4.6 east… 94907 <tibble> clim… (147.1322 -38.1156)
-#> 6 2260… 147. -38.1 NA LA T… NA <tibble> rive… (147.1278 -38.14491)
-#> 7 ASN0… 145. -36.2 96 echu… 94861 <tibble> clim… (144.7642 -36.1647)
-#> 8 4067… 145. -36.1 NA DEAK… NA <tibble> rive… (144.7693 -36.12866)
-#> 9 ASN0… 146. -36.8 502 stra… 95843 <tibble> clim… (145.7308 -36.8472)
-#> 10 4052… 146. -36.9 NA SEVE… NA <tibble> rive… (145.6828 -36.88701)
-#> 11 ASN0… 145. -37.7 78.4 esse… 95866 <tibble> clim… (144.9066 -37.7276)
-#> 12 2302… 145. -37.7 NA MARI… NA <tibble> rive… (144.8365 -37.72771)
-#> 13 ASN0… 148. -37.9 49.4 bair… 94912 <tibble> clim… (147.5669 -37.8817)
-#> 14 2242… 148. -37.8 NA MITC… NA <tibble> rive… (147.5722 -37.815)
-#> 15 ASN0… 145. -36.3 105 kyab… 95833 <tibble> clim… (145.0638 -36.335)
-#> 16 4067… 145. -36.3 NA MOSQ… NA <tibble> rive… (144.9809 -36.32871)
-#> # ℹ 2 more variables: group <int>, dist [m]Once the distance is calculated, three arguments are available to -refine the matching results:
--
-
-
-
spatial_n_keep: Number of match each major site -receive
- -
-
spatial_dist_max: maximum distance allowed for a pair -of matching
- -
-
spatial_single_match: Whether each minor site can only -be matched to one major site
-
The order that these three arguments applied will slightly affect the
-results in cubble. spatial_n_keep, default to
-1, is first applied to keep n site(s) for each major site,
-spaital_dist_max, default to 10, is then applied to filter
-out the pairs with distance larger than this maximum distance.
-spatial_single_match is lastly applied to resolve the
-scenario where site a (minor) is the closest match for both
-site A and B (major) with distance 5km and
-8km. If spatial_single_match = TRUE, a will
-only be matched to the major site with the smaller distance, that is,
-site A here.
Temporal matching -
-Here we provide more details on how temporal matching works in
-cubble. Suppose two locations have been matched spatially
-and temporal matching will be conducted on variable A and
-a in the plot below: .
We first find the n peaks in each series (3 peaks here).
-A variable needs to be specified in temporal_independent
-for construct an interval. Here we pick variable A and
-construct an interval with a default length of 5. The peaks in variable
-a are then tested against whether they fall into the any of
-the intervals constructed from A. In this illustration,
-there are 2 matches for these two variable The available tuning
-parameter in temporal matches are:
-
-
-
-
temporal_n_highest: the number of peak used - 3 in the -example above
- -
-
temporal_window: the length of the interval - 5 in the -example above
- -
-
temporal_min_match: the minimum number of matched peak -for a valid matched pair. To return all the pairs of the match, set this -parameter to 0.
-
In the river level and precipitation example,
-Water_course_level in river will be matched to
-prcp in climate. This can be specified in
-temporal_by, an analogue to the by syntax in
-join. The goal in this example is to see if precipitation
-will be reflected by the water level in the river and this puts
-precipitation prcp, as the independent. Given there is one
-year worth of data, the number of peak (temporal_n_highest)
-to consider is slightly raised from a default 20 to 30 and
-temporal_min_match is raised accordingly.
-res_tm <- res_sp %>%
- match_temporal(
- data_id = type,
- match_id = group,
- return_cubble = TRUE,
- temporal_by = c("prcp" = "Water_course_level"))
-res_tm[[1]]
-#> # cubble: key: id [2], index: date, nested form, [sf]
-#> # spatial: [144.5203, -37.0194, 144.540295, -37.015122], WGS 84
-#> # temporal: date [date], matched [dbl]
-#> id long lat elev name wmo_id type geometry group
-#> <chr> <dbl> <dbl> <dbl> <chr> <dbl> <chr> <POINT [°]> <int>
-#> 1 ASN00088… 145. -37.0 290 rede… 94859 clim… (144.5203 -37.0194) 1
-#> 2 406213 145. -37.0 NA CAMP… NA rive… (144.5403 -37.01512) 1
-#> # ℹ 3 more variables: dist [m], ts <list>, match_res <dbl>The output from temporal matching is also a cubble with
-n_match for the number of matched temporal peaks (on top of
-the dist and group from spatial
-matching)：
We can look at the matched pair on the map:
-
-vic_map <- ozmaps::abs_ste |>
- filter(NAME == "Victoria")
-
-ggplot() +
- geom_sf(data = vic_map, fill = "grey95", color = "white") +
- geom_point(data = dplyr::bind_rows(river, climate_vic),
- aes(x = long, y = lat, color = type), alpha = 0.2, fill = 0.2) +
- geom_point(data = res_tm %>% as_tibble() ,
- aes(x = long, y = lat, color = type)) +
- ggrepel::geom_label_repel(
- data = res_tm |> filter(type == "river") %>% as_tibble(),
- aes(x = long, y = lat, label = group)) +
- scale_color_brewer(palette = "Dark2") +
- theme_void() +
- ggplot2::theme(legend.position = "bottom",
- legend.text = element_text(size = 15),
- legend.title = element_text(size = 15)) +
- ggplot2::labs(x = "Longitude", y = "Latitude") +
- guides(color = guide_legend(override.aes = list(size=5)))
or to look at the series:
-
-res_tm_long <- res_tm %>%
- face_temporal() %>%
- unfold(group, type) %>%
- rename(prcp = matched) %>%
- group_by(group, type) %>%
- mutate(prcp = (prcp - min(prcp, na.rm = TRUE))/
- (max(prcp, na.rm = TRUE) - min(prcp, na.rm = TRUE)))
-
-res_tm_long %>%
- ggplot(aes(x = date, y = prcp, group = type,color = type)) +
- geom_line() +
- facet_wrap(vars(group)) +
- scale_color_brewer(palette = "Dark2", guide = "none") +
- theme_bw() +
- labs(x= "date") +
- scale_x_date(date_labels = "%b") +
- labs(x = "Week", y = "Precipitation/ water level")
There are four pairs of matches - all locates in the middle Victoria -and we can observe concurrent increase of precipitation and water level -(precipitation and water level have been standardised between 0 and 1 to -be displayed on the same scale).
-
