Load the packages:
library(tidyverse)
library(tidypaleo)
Preparing the data:
alta_lake_geochem#> # A tibble: 192 × 9
#> location param depth age value stdev units n zone
#> <chr> <chr> <dbl> <dbl> <dbl> <dbl> <chr> <int> <chr>
#> 1 ALGC2 Cu 0.25 2015. 76 NA ppm 1 Zone 3
#> 2 ALGC2 Cu 0.75 2011. 108. 4.50 ppm 3 Zone 3
#> 3 ALGC2 Cu 1.25 2008. 158 NA ppm 1 Zone 3
#> 4 ALGC2 Cu 1.75 2003. 169 NA ppm 1 Zone 3
#> 5 ALGC2 Cu 2.5 1998. 161 NA ppm 1 Zone 3
#> 6 ALGC2 Cu 3.5 1982. 129 NA ppm 1 Zone 3
#> 7 ALGC2 Cu 4.5 1966. 88.7 3.86 ppm 3 Zone 2
#> 8 ALGC2 Cu 5.5 1947. 65 NA ppm 1 Zone 2
#> 9 ALGC2 Cu 6.5 1922. 62.3 9.53 ppm 3 Zone 2
#> 10 ALGC2 Cu 7.5 1896. 48 NA ppm 1 Zone 2
#> # … with 182 more rows
<- nested_data(
alta_nested
alta_lake_geochem,qualifiers = c(age, depth, zone),
key = param,
value = value,
trans = scale
)
alta_nested#> # A tibble: 1 × 4
#> discarded_columns discarded_rows qualifiers data
#> * <list> <list> <list> <list>
#> 1 <tibble [32 × 0]> <tibble [0 × 9]> <tibble [32 × 4]> <tibble [32 × 6]>
%>% unnested_data(data)
alta_nested #> # A tibble: 32 × 6
#> C[,1] `C/N`[,1] Cu[,1] Ti[,1] d13C[,1] d15N[,1]
#> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 -1.54 1.26 -0.794 0.807 1.03 -0.670
#> 2 -1.59 1.36 -0.559 1.33 1.19 0.0499
#> 3 -1.98 0.960 -0.721 0.682 1.10 -0.511
#> 4 -0.189 1.61 -0.749 0.233 0.836 -2.37
#> 5 0.993 2.48 -0.694 0.908 1.06 -2.55
#> 6 -0.157 1.76 -0.712 0.941 1.17 -1.52
#> 7 -0.642 1.36 -0.667 1.14 1.07 -1.39
#> 8 -1.07 0.924 -0.559 1.16 0.820 -0.439
#> 9 -0.722 0.696 -0.830 0.932 0.765 -0.929
#> 10 -0.631 0.309 -0.504 0.882 0.409 -0.166
#> # … with 22 more rows
%>% unnested_data(qualifiers, data)
alta_nested #> # A tibble: 32 × 10
#> age depth zone row_number C[,1] `C/N`[,1] Cu[,1] Ti[,1] d13C[,1] d15N[…¹
#> <dbl> <dbl> <chr> <int> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 1550 29.5 Zone 1 1 -1.54 1.26 -0.794 0.807 1.03 -0.670
#> 2 1566. 28.5 Zone 1 2 -1.59 1.36 -0.559 1.33 1.19 0.0499
#> 3 1581. 27.5 Zone 1 3 -1.98 0.960 -0.721 0.682 1.10 -0.511
#> 4 1597. 26.5 Zone 1 4 -0.189 1.61 -0.749 0.233 0.836 -2.37
#> 5 1613. 25.5 Zone 1 5 0.993 2.48 -0.694 0.908 1.06 -2.55
#> 6 1629. 24.5 Zone 1 6 -0.157 1.76 -0.712 0.941 1.17 -1.52
#> 7 1644. 23.5 Zone 1 7 -0.642 1.36 -0.667 1.14 1.07 -1.39
#> 8 1660. 22.5 Zone 1 8 -1.07 0.924 -0.559 1.16 0.820 -0.439
#> 9 1676. 21.5 Zone 1 9 -0.722 0.696 -0.830 0.932 0.765 -0.929
#> 10 1692. 20.5 Zone 1 10 -0.631 0.309 -0.504 0.882 0.409 -0.166
#> # … with 22 more rows, and abbreviated variable name ¹d15N[,1]
<- alta_nested %>% nested_prcomp()
pca
pca#> # A tibble: 1 × 8
#> discarded_col…¹ discar…² qualif…³ data model variance loadings scores
#> * <list> <list> <list> <list> <list> <list> <list> <list>
#> 1 <tibble> <tibble> <tibble> <tibble> <prcomp> <tibble> <tibble> <tibble>
#> # … with abbreviated variable names ¹discarded_columns, ²discarded_rows,
#> # ³qualifiers
plot(pca)
%>% unnested_data(qualifiers, scores)
pca #> # A tibble: 32 × 10
#> age depth zone row_number PC1 PC2 PC3 PC4 PC5 PC6
#> <dbl> <dbl> <chr> <int> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 1550 29.5 Zone 1 1 -2.48 -0.273 0.409 0.518 -0.316 0.0538
#> 2 1566. 28.5 Zone 1 2 -2.48 -0.675 0.902 0.561 0.364 0.190
#> 3 1581. 27.5 Zone 1 3 -2.41 -0.721 0.576 0.465 -0.619 0.184
#> 4 1597. 26.5 Zone 1 4 -2.39 1.73 -0.527 0.197 -0.742 -0.0656
#> 5 1613. 25.5 Zone 1 5 -2.73 2.82 -0.666 0.212 0.432 -0.0715
#> 6 1629. 24.5 Zone 1 6 -2.56 1.22 -0.155 0.235 0.208 0.105
#> 7 1644. 23.5 Zone 1 7 -2.57 0.708 0.123 -0.0210 0.0114 -0.0286
#> 8 1660. 22.5 Zone 1 8 -2.04 -0.243 0.490 0.101 0.157 -0.0834
#> 9 1676. 21.5 Zone 1 9 -1.98 0.0702 -0.0657 -0.147 -0.0813 -0.186
#> 10 1692. 20.5 Zone 1 10 -1.20 -0.376 0.202 -0.132 0.202 -0.220
#> # … with 22 more rows
%>% unnested_data(variance)
pca #> # A tibble: 6 × 6
#> component component_text standard_deviation variance variance_propor…¹ varia…²
#> <int> <chr> <dbl> <dbl> <dbl> <dbl>
#> 1 1 PC1 2.15 4.61 0.768 0.768
#> 2 2 PC2 0.884 0.781 0.130 0.899
#> 3 3 PC3 0.603 0.364 0.0607 0.959
#> 4 4 PC4 0.381 0.145 0.0242 0.984
#> 5 5 PC5 0.276 0.0761 0.0127 0.996
#> 6 6 PC6 0.151 0.0228 0.00380 1
#> # … with abbreviated variable names ¹variance_proportion,
#> # ²variance_proportion_cumulative
%>% unnested_data(loadings)
pca #> # A tibble: 6 × 7
#> variable PC1 PC2 PC3 PC4 PC5 PC6
#> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 C 0.380 0.540 -0.460 -0.162 0.567 0.0718
#> 2 C/N -0.401 0.451 0.207 0.735 0.223 -0.0408
#> 3 Cu 0.387 0.340 0.760 -0.173 -0.0609 0.352
#> 4 Ti -0.439 -0.0783 0.356 -0.483 0.565 -0.350
#> 5 d13C -0.458 -0.0890 -0.144 -0.145 0.126 0.851
#> 6 d15N 0.377 -0.613 0.144 0.386 0.539 0.149
<- keji_lakes_plottable %>%
keji_nested group_by(location) %>%
nested_data(qualifiers = depth, key = taxon, value = rel_abund)
%>% unnested_data(qualifiers, data)
keji_nested #> # A tibble: 37 × 9
#> location depth row_num…¹ Aster…² Aulac…³ Aulac…⁴ Cyclo…⁵ Tabel…⁶ Other
#> <chr> <dbl> <int> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 Beaverskin Lake 0.125 1 0 3.02 1.72 3.45 0 91.8
#> 2 Beaverskin Lake 0.375 2 0 3.25 2.03 5.69 1.63 87.4
#> 3 Beaverskin Lake 0.825 3 1.32 0 2.32 4.30 1.32 90.7
#> 4 Beaverskin Lake 2.12 4 0.333 0 2.67 3 1.67 92.3
#> 5 Beaverskin Lake 3.12 5 0 0 5.32 6.98 1.33 86.4
#> 6 Beaverskin Lake 4.12 6 0 0 3.54 13.2 0.643 82.6
#> 7 Beaverskin Lake 5.38 7 0.987 0 10.2 12.8 2.63 73.4
#> 8 Beaverskin Lake 6.38 8 0.993 0 8.94 17.5 3.97 68.5
#> 9 Beaverskin Lake 7.62 9 1.63 0 8.82 20.3 2.29 67.0
#> 10 Beaverskin Lake 9.12 10 0.328 0 10.8 23.9 2.95 62.0
#> # … with 27 more rows, and abbreviated variable names ¹row_number,
#> # ²`Asterionella ralfsii var. americana (large)`, ³`Aulacoseira distans`,
#> # ⁴`Aulacoseira lirata`, ⁵`Cyclotella stelligera`,
#> # ⁶`Tabellaria flocculosa (strain III)`
<- keji_nested %>%
coniss nested_chclust_coniss()
plot(coniss, main = location)
plot(coniss, main = location, xvar = qualifiers$depth, labels = "")
%>% select(location, zone_info) %>% unnest(zone_info)
coniss #> # A tibble: 4 × 12
#> location hclus…¹ min_d…² max_d…³ first…⁴ last_…⁵ min_r…⁶ max_r…⁷ first…⁸
#> <chr> <int> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 Beaverskin La… 1 0.125 4.12 0.125 4.12 1 6 1
#> 2 Beaverskin La… 2 5.38 23.4 5.38 23.4 7 17 7
#> 3 Peskawa Lake 1 0.125 5.12 0.125 5.12 1 5 1
#> 4 Peskawa Lake 2 6.38 35.1 6.38 35.1 6 20 6
#> # … with 3 more variables: last_row_number <dbl>, boundary_depth <dbl>,
#> # boundary_row_number <dbl>, and abbreviated variable names ¹hclust_zone,
#> # ²min_depth, ³max_depth, ⁴first_depth, ⁵last_depth, ⁶min_row_number,
#> # ⁷max_row_number, ⁸first_row_number
%>%
keji_nested nested_chclust_coniss(n_groups = c(3, 2)) %>%
select(location, zone_info) %>%
unnested_data(zone_info)
#> # A tibble: 5 × 12
#> location hclus…¹ min_d…² max_d…³ first…⁴ last_…⁵ min_r…⁶ max_r…⁷ first…⁸
#> <chr> <int> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 Beaverskin La… 1 0.125 4.12 0.125 4.12 1 6 1
#> 2 Beaverskin La… 2 5.38 13.6 5.38 13.6 7 13 7
#> 3 Beaverskin La… 3 15.1 23.4 15.1 23.4 14 17 14
#> 4 Peskawa Lake 1 0.125 5.12 0.125 5.12 1 5 1
#> 5 Peskawa Lake 2 6.38 35.1 6.38 35.1 6 20 6
#> # … with 3 more variables: last_row_number <dbl>, boundary_depth <dbl>,
#> # boundary_row_number <dbl>, and abbreviated variable names ¹hclust_zone,
#> # ²min_depth, ³max_depth, ⁴first_depth, ⁵last_depth, ⁶min_row_number,
#> # ⁷max_row_number, ⁸first_row_number
<- halifax_lakes_plottable %>%
halifax_nested nested_data(c(location, sample_type), taxon, rel_abund, fill = 0)
%>% unnested_data(qualifiers, data)
halifax_nested #> # A tibble: 20 × 9
#> location sampl…¹ row_n…² Aulac…³ Eunot…⁴ Fragi…⁵ Tabel…⁶ Tabel…⁷ Other
#> <chr> <chr> <int> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 Anderson Lake bottom 1 4.65 2.42 1.16 4.11 3.76 83.9
#> 2 Anderson Lake top 2 1.87 0 0.330 5.71 5.71 86.4
#> 3 Bayers bottom 3 11.6 5.71 4.50 4.84 0 73.4
#> 4 Bayers top 4 0.993 6.81 3.40 3.40 0 85.4
#> 5 Bell Lake bottom 5 0.476 2.62 0.833 2.14 6.90 87.0
#> 6 Bell Lake top 6 9.22 0.432 2.59 6.20 1.01 80.5
#> 7 Cranberry Lake bottom 7 0 9.17 11.6 3.06 9.39 66.8
#> 8 Cranberry Lake top 8 0 7.72 8.94 0.203 8.54 74.6
#> 9 Frasers Lake bottom 9 6.42 0.714 0.624 1.87 2.59 87.8
#> 10 Frasers Lake top 10 4.85 0 0 8.58 10.6 76.0
#> 11 Kinsac lake bottom 11 11.3 8.33 2.98 12.7 0.992 63.7
#> 12 Kinsac lake top 12 0 3.85 2.75 3.30 0 90.1
#> 13 Little Albro L… bottom 13 2.34 2.34 1.91 1.70 6.69 85.0
#> 14 Little Albro L… top 14 5.78 2.61 1.90 0 5.78 83.9
#> 15 Little Springf… bottom 15 12.1 2.64 6.23 10 0 69.1
#> 16 Little Springf… top 16 0 19.8 14.1 11.4 0 54.6
#> 17 Maynard Lake bottom 17 9.75 4 1.5 11.6 7.88 65.2
#> 18 Maynard Lake top 18 2.98 1.23 2.26 7.51 10.1 75.9
#> 19 Miller Lake bottom 19 1.79 2.19 2.19 0.299 15.6 77.9
#> 20 Miller Lake top 20 0.816 4.35 2.90 1.27 1.18 89.5
#> # … with abbreviated variable names ¹sample_type, ²row_number,
#> # ³`Aulacoseira distans`, ⁴`Eunotia exigua`, ⁵`Fragilariforma exigua`,
#> # ⁶`Tabellaria fenestrata`, ⁷`Tabellaria flocculosa (strain IV)`
<- halifax_nested %>%
hclust nested_hclust(method = "average")
plot(
hclust, labels = sprintf(
"%s (%s)",
$location,
qualifiers$sample_type
qualifiers
) )
%>%
alta_nested nested_analysis(vegan::rda, data) %>%
plot()
biplot(pca)