A garden of glyphs: edible plants as flowers

TidyTuesday 2026-02-03 · The Edible Plant Database (GROW Observatory)

Published

June 9, 2026

Session 1 · co-developed

This page came out of a live, turn-by-turn conversation between Jon Minton and Claude (Fable 5) — questions, plots and interpretation built together in real time. The Session 2 pages were instead produced by Claude working autonomously.

The Edible Plant Database describes 146 food crops by their growing requirements — sunlight, water, feeding, soil pH, time to harvest. That’s k-dimensional data crying out for a multivariate display.

The classic answer is a Chernoff face: map each variable to a facial feature and let human face-perception do the pattern-matching. Faces are powerful but faintly absurd for vegetables. So this page borrows Chernoff’s idea and the shape of Fisher’s iris measurements (petal length, petal width, counts) but grows flower glyphs instead — each crop rendered as the bloom its growing conditions imply.

Code

source("R/flower_glyphs.R")
plants <- load_plants()

How to read a flower

Every glyph is a faithful encoding of five growing variables. Nothing is decorative:

Flower feature	Variable	Meaning
Number of petals	Sunlight	5 = shade-tolerant · 6 = sun/part-shade · 8 = full sun
Petal length	Days to harvest	longer petals = slower crop
Petal width	Water need	fatter petals = thirstier
Centre size	Feeding (nutrients)	bigger golden disc = hungry feeder
Petal colour	Soil pH	acid-blue → neutral-violet → alkaline-pink (a hydrangea’s logic)

Code

decoder <- tibble(
  common_name = c("Quick · sun · thirsty · hungry",
                  "Slow · sun · dry · lean",
                  "Shade · medium · acid soil",
                  "Slow · part-shade · alkaline"),
  sun_lvl = c("Full sun", "Full sun", "Shade-tolerant", "Sun or part shade"),
  harvest_days = c(30, 160, 90, 150),
  water_lvl = c(5, 1, 3, 2),
  nutrient_lvl = c(3, 1, 2, 2),
  ph_mid = c(6.4, 6.4, 5.3, 7.6)
)

draw_garden(build_garden(decoder, ncol = 4), label_size = 3) +
  labs(title = "Four reference blooms",
       subtitle = "Read every real flower below against these")

The kitchen garden

A selection of common kitchen-garden crops, sorted left-to-right, top-to-bottom by time to harvest — so the bloom-size visibly swells as you read down the plot, from the radish that’s ready in three weeks to the cucumber and leek that make you wait.

Code

wanted <- c("Radish", "Spinach", "Lettuce (Headed)", "Rocket", "Beetroot",
  "Pea", "Mangetout", "Carrot", "Kale", "Broccoli", "Cauliflower",
  "Bell Pepper", "Chilli Pepper", "Tomato (Cherry)", "Garlic", "Fennel",
  "Sweetcorn", "Celery", "Beans (Runner)", "Pumpkin", "Onion", "Parsnip",
  "Cucumber", "Leek")

garden <- plants |>
  filter(common_name %in% wanted,
         !is.na(harvest_days), !is.na(ph_mid),
         !is.na(water_lvl), !is.na(nutrient_lvl)) |>
  arrange(harvest_days)

draw_garden(build_garden(garden, ncol = 6)) +
  labs(
    title = "A kitchen garden, encoded",
    subtitle = "Petals = sunlight · length = days to harvest · width = water · centre = feeding · colour = pH"
  )

Each flower is one crop; its shape is its growing requirements. Sorted by days to harvest.

What the glyphs reveal at a glance

The flower form does what Chernoff intended — it turns rows of numbers into gestalts you can sort by eye:

The blueberry-blue end is nearly empty. Almost every kitchen crop wants near-neutral soil; the violet-to-pink cast across the garden shows how narrow the edible-pH band really is. The blue acid-lovers (cranberry, blueberry) are the specialists that live in the missing corner of this selection.
Big golden centres cluster with fat petals. Hungry feeders (pumpkin, celery, sweetcorn) tend also to be thirsty — the glyph couples nutrient and water demand visually, hinting at the real correlation below.
Small tidy flowers are the quick wins. Radish, rocket and salad leaves read as compact five- and six-petal blooms: shade-tolerant, fast, low-input — the crops to interplant between the slow giants.

Code

plants |>
  filter(!is.na(water_lvl), !is.na(nutrient_lvl)) |>
  count(water_lvl, nutrient_lvl) |>
  ggplot(aes(water_lvl, nutrient_lvl, size = n)) +
  geom_point(colour = "#7a9a4e", alpha = 0.8) +
  scale_size_area(max_size = 16, name = "species") +
  scale_x_continuous(breaks = 1:5,
    labels = c("very low", "low", "medium", "high", "very high")) +
  scale_y_continuous(breaks = 1:3, labels = c("low", "medium", "high")) +
  labs(
    title = "Thirsty crops do tend to be hungry crops",
    subtitle = "The glyphs' coupling of petal width and centre size reflects a genuine association",
    x = "Water requirement", y = "Nutrient requirement"
  ) +
  theme_minimal(base_size = 12)

Is the visual coupling of thirst and hunger real? Water vs nutrient demand across all 146 species (jittered).

The glyph display isn’t just decoration: the eye’s impression that fat-petalled flowers have big centres turns out to be a real positive association between water and nutrient demand. That is exactly the Chernoff bet — that a well-chosen visual mapping lets pattern-recognition find structure faster than a correlation table would.

Honest limits of the form

Glyph displays trade precision for gestalt. Petal count is the strongest visual signal but encodes only a 3-level variable (sunlight), while the subtle pH colour does heavy lifting that’s easy to miss. As with Chernoff faces, which feature carries which variable changes what pops — assign the loudest channel to the variable you most want compared.

--- title: "A garden of glyphs: edible plants as flowers" subtitle: "TidyTuesday 2026-02-03 · The Edible Plant Database (GROW Observatory)" date: 2026-06-09 --- ::: {.callout-tip icon=false} ## Session 1 · co-developed This page came out of a live, turn-by-turn conversation between Jon Minton and Claude (Fable 5) — questions, plots and interpretation built together in real time. The [Session 2 pages](index.qmd) were instead produced by Claude working autonomously. ::: The [Edible Plant Database](https://growobservatory.org/) describes 146 food crops by their growing requirements — sunlight, water, feeding, soil pH, time to harvest. That's *k*-dimensional data crying out for a multivariate display. The classic answer is a **Chernoff face**: map each variable to a facial feature and let human face-perception do the pattern-matching. Faces are powerful but faintly absurd for vegetables. So this page borrows Chernoff's idea and the *shape* of Fisher's iris measurements (petal length, petal width, counts) but grows **flower glyphs** instead — each crop rendered as the bloom its growing conditions imply. ```{r setup} source("R/flower_glyphs.R") plants <- load_plants() ``` ## How to read a flower Every glyph is a faithful encoding of five growing variables. Nothing is decorative: | Flower feature | Variable | Meaning | |---|---|---| | **Number of petals** | Sunlight | 5 = shade-tolerant · 6 = sun/part-shade · 8 = full sun | | **Petal length** | Days to harvest | longer petals = slower crop | | **Petal width** | Water need | fatter petals = thirstier | | **Centre size** | Feeding (nutrients) | bigger golden disc = hungry feeder | | **Petal colour** | Soil pH | acid-blue → neutral-violet → alkaline-pink (a hydrangea's logic) | ```{r decoder} #| fig-height: 3.6 #| fig-width: 10 #| fig-cap: "The decoder: four reference blooms spanning the extremes of each axis." decoder <- tibble( common_name = c("Quick · sun · thirsty · hungry", "Slow · sun · dry · lean", "Shade · medium · acid soil", "Slow · part-shade · alkaline"), sun_lvl = c("Full sun", "Full sun", "Shade-tolerant", "Sun or part shade"), harvest_days = c(30, 160, 90, 150), water_lvl = c(5, 1, 3, 2), nutrient_lvl = c(3, 1, 2, 2), ph_mid = c(6.4, 6.4, 5.3, 7.6) ) draw_garden(build_garden(decoder, ncol = 4), label_size = 3) + labs(title = "Four reference blooms", subtitle = "Read every real flower below against these") ``` ## The kitchen garden A selection of common kitchen-garden crops, sorted left-to-right, top-to-bottom by time to harvest — so the bloom-size visibly swells as you read down the plot, from the radish that's ready in three weeks to the cucumber and leek that make you wait. ```{r garden} #| fig-height: 9.5 #| fig-width: 10.5 #| fig-cap: "Each flower is one crop; its shape is its growing requirements. Sorted by days to harvest." wanted <- c("Radish", "Spinach", "Lettuce (Headed)", "Rocket", "Beetroot", "Pea", "Mangetout", "Carrot", "Kale", "Broccoli", "Cauliflower", "Bell Pepper", "Chilli Pepper", "Tomato (Cherry)", "Garlic", "Fennel", "Sweetcorn", "Celery", "Beans (Runner)", "Pumpkin", "Onion", "Parsnip", "Cucumber", "Leek") garden <- plants |> filter(common_name %in% wanted, !is.na(harvest_days), !is.na(ph_mid), !is.na(water_lvl), !is.na(nutrient_lvl)) |> arrange(harvest_days) draw_garden(build_garden(garden, ncol = 6)) + labs( title = "A kitchen garden, encoded", subtitle = "Petals = sunlight · length = days to harvest · width = water · centre = feeding · colour = pH" ) ``` ### What the glyphs reveal at a glance The flower form does what Chernoff intended — it turns rows of numbers into *gestalts* you can sort by eye: - **The blueberry-blue end is nearly empty.** Almost every kitchen crop wants near-neutral soil; the violet-to-pink cast across the garden shows how narrow the edible-pH band really is. The blue acid-lovers (cranberry, blueberry) are the specialists that live in the missing corner of this selection. - **Big golden centres cluster with fat petals.** Hungry feeders (pumpkin, celery, sweetcorn) tend also to be thirsty — the glyph couples nutrient and water demand visually, hinting at the real correlation below. - **Small tidy flowers are the quick wins.** Radish, rocket and salad leaves read as compact five- and six-petal blooms: shade-tolerant, fast, low-input — the crops to interplant between the slow giants. ```{r water-nutrient-check} #| fig-height: 4 #| fig-cap: "Is the visual coupling of thirst and hunger real? Water vs nutrient demand across all 146 species (jittered)." plants |> filter(!is.na(water_lvl), !is.na(nutrient_lvl)) |> count(water_lvl, nutrient_lvl) |> ggplot(aes(water_lvl, nutrient_lvl, size = n)) + geom_point(colour = "#7a9a4e", alpha = 0.8) + scale_size_area(max_size = 16, name = "species") + scale_x_continuous(breaks = 1:5, labels = c("very low", "low", "medium", "high", "very high")) + scale_y_continuous(breaks = 1:3, labels = c("low", "medium", "high")) + labs( title = "Thirsty crops do tend to be hungry crops", subtitle = "The glyphs' coupling of petal width and centre size reflects a genuine association", x = "Water requirement", y = "Nutrient requirement" ) + theme_minimal(base_size = 12) ``` The glyph display isn't just decoration: the eye's impression that fat-petalled flowers have big centres turns out to be a real positive association between water and nutrient demand. That is exactly the Chernoff bet — that a well-chosen visual mapping lets pattern-recognition find structure faster than a correlation table would. ::: {.callout-note} ## Honest limits of the form Glyph displays trade precision for gestalt. Petal *count* is the strongest visual signal but encodes only a 3-level variable (sunlight), while the subtle pH colour does heavy lifting that's easy to miss. As with Chernoff faces, which feature carries which variable changes what *pops* — assign the loudest channel to the variable you most want compared. :::