updates

AQP/aqp/genhz-distance-eval.Rmd

@@ -49,8 +49,8 @@ hzdesgnname(loafercreek) <- 'hzname'
 x <- loafercreek[20:40, ]
 
 # modify default arguments for some flair
-par(mar = c(0, 0, 0, 0))
-plotSPC(x, width = 0.3, name.style = 'center-center', depth.axis = FALSE, hz.depths = TRUE, cex.names = 0.65, print.id = FALSE)
+par(mar = c(0, 0, 0, 2))
+plotSPC(x, width = 0.33, name.style = 'center-center', cex.names = 0.65, print.id = FALSE)
 ```
 
 
@@ -88,8 +88,8 @@ table(y$genhz)
 # add short ID for checking our work
 site(y)$shortID <- 1:length(y)
 
-par(mar = c(0, 0, 3, 0))
-plotSPC(y, color = 'genhz', width = 0.3, name.style = 'center-center', depth.axis = FALSE, hz.depths = TRUE, cex.names = 0.65, fixLabelCollisions = FALSE, col.label = 'Generalized Horizon Label', label = 'shortID')
+par(mar = c(0, 0, 3, 2))
+plotSPC(y, color = 'genhz', width = 0.33, name.style = 'center-center', cex.names = 0.65, fixLabelCollisions = FALSE, col.label = 'Generalized Horizon Label', label = 'shortID')
 ```
 
 
@@ -99,6 +99,7 @@ Add empty horizons to bottom of the profiles with bottom-most horizon depth < 15
 z <- fillHzGaps(y, to_top = NA, to_bottom = 150)
 
 # label as 'R'
+# important to use one of the GHL specified in `genhz`
 z$hzname[is.na(z$genhz)] <- 'R'
 z$genhz[is.na(z$genhz)] <- 'R'
 
@@ -193,6 +194,44 @@ mtext("Median distance to all other profiles (not to scale)", side = 1, line = 2
 ```
 
 
+Another view of median distance to all other profiles, as a set of connected points. This helps demonstrate the scale of differences within the collection. Note that values in the distance matrix `d` range from 0 to 1. We have to use some tricks to get the vertical spacing right.
+```{r, fig.width=12, fig.height=6.5}
+# trick 1: adjust median distances to fit "above" profiles
+# shift down 125cm
+# rescale from 0 to 125cm
+.yoff <- 125
+med.dist.rescale <- .yoff - aqp:::.rescaleRange(med.dist, 0, .yoff)
+
+# trick 2: plot profiles, shifting "down" 150cm
+# use y.offset argument
+par(mar = c(0.25, 1.5, 3, 1.5))
+plotSPC(z, plot.order = idx, color = 'genhz', width = 0.33, name.style = 'center-center', cex.names = 0.65, col.label = 'Generalized Horizon Label', label = 'shortID', scaling.factor = 1, y.offset = 150)
+
+# add the scaled + shifted median distances as points connected by lines
+lines(x = 1:length(z), y = med.dist.rescale, type = 'b')
+
+# trick 3: make a special axis describing median distance
+# compute "pretty" intervals for the axis, on the shifted and scaled values
+.p <- pretty(med.dist.rescale)
+
+# use a linear model to convert between shifted/scaled and original values
+# slope ~ scale
+# intercept ~ shift
+.m <- lm(med.dist ~ med.dist.rescale)
+
+# convert pretty axis values back to original scale
+.lab <- predict(.m, data.frame(med.dist.rescale = .p))
+
+# add axis
+axis(side = 2, at = .p, labels = round(.lab, 2), line = -1.5, las = 1, cex.axis = 0.75)
+
+# annotate
+mtext('Median distance to all other profiles (max = 1)', side = 3, at = 1, adj = 0, line = -2)
+```
+
+
+
+
 Use `plotProfileDendrogram()` from the sharpshootR package to align profile sketches with the terminal leaves of a dendrogram. When applied to other data, you may have to tinker with the `scaling.factor` and `y.offset` arguments.
 ```{r, fig.width=12, fig.height=8}
 par(mar = c(1, 0, 3, 2))
@@ -232,7 +271,11 @@ axis(1, at = pos$relative.pos, labels = round(pos$grad, 3), line = 0, cex.axis =
 mtext("Distance from 1st Profile", side = 1, line = 2.25)
 ```
 
-Create an ordination from pair-wise distances using non-metric multidimensional scaling (`MASS::sammon`).
+Create an ordination from pair-wise distances using non-metric multidimensional scaling (`MASS::sammon`). It is necessary to add a small amount of distance back to the distance matrix before ordination for two reasons:
+
+  1. we chose to rescale distances to the interval of [0,1] (recall `rescaleResult = TRUE` argument to `NCSP()`)
+  2. there are duplicate soil profiles in the collection
+
 ```{r fig.width=8, fig.height=4.5}
 # add tiny amount for 0-distance pairs (duplicate profiles)
 d <- d + 0.001