What Sits in the Difference

A scrollytelling argument about a half-degree of warming, built for Funafuti Atoll, Tuvalu — Pacific Dataviz Challenge 2026

An interactive scrollytelling piece arguing that for Tuvalu, the difference between 1°C and 1.5°C of global warming is not a number — it’s the island. Built on an open, reproducible GLO-30 elevation pipeline rather than restricted flood rasters, with a bathtub inundation model across three sea-level scenarios. Submitted to the Pacific Dataviz Challenge 2026 (Climate Change theme), International Global Mention category.

NoteRead the interactive article

→ What Sits in the Difference

This page documents the analysis, editorial decisions, and implementation behind the project.

← Back to Steven Ponce’s Data Visualization Portfolio

Tuvalu isn’t the subject of this piece — the threshold is.

Project at a glance

Duration	June–August 2026
Role	Analysis, editorial design, spatial modeling, visualization, implementation
Deliverable	Interactive scrollytelling data story
Competition	Pacific Dataviz Challenge 2026 (Climate Change theme) — International Global Mention category
Tools	R, terra, sf, elevatr, osmextract, ggplot2, tidyterra, Quarto, Closeread
Data	Pacific Data Hub .Stat Explorer (Sea Level Anomalies, SST Anomalies); GLO-30 DEM via elevatr
Live article	what-sits-in-the-difference.netlify.app

---

For Tuvalu, the difference between 1°C and 1.5°C of warming is not a number — it’s the island.

---

1. Context, problem, and goal

Context

The Pacific Dataviz Challenge 2026 set climate change as its theme and opened its official dataset list — the Pacific Data Hub’s .Stat Explorer climate indicators — on June 1. Scoping ahead of that date meant working through roughly 217 available Pacific climate datasets, the majority of which were static infrastructure-exposure snapshots rather than anything with a narrative spine.

Problem

Sea-level rise is almost always visualized as a line trending upward. That framing is accurate and almost never lands — a rising line reads as abstract, not as stakes. The harder problem wasn’t finding data; it was finding one argument strong enough to carry an entire piece without padding it with additional metrics.

Goal

Build a single-spine argument, not a dashboard: for Funafuti Atoll, Tuvalu, the difference between a 1.0m and a 1.5m sea-level-rise scenario is not an incremental data point — it’s the difference between a diminished island and a fragmented one. Everything in the piece either serves that one threshold or was cut.

---

2. Why an open elevation pipeline, not the official flood rasters

Tuvalu’s flood-scenario rasters exist through SPC/NIWA but are access-restricted. Using them would have meant citing a result I couldn’t show the derivation of — a black box at the center of the piece’s central claim.

Instead, the inundation model is built from the ground up: a GLO-30 digital elevation model (via elevatr), corrected for a uniform canopy bias offset calibrated against published Funafuti ground-truth elevations, run through a bathtub inundation model at each scenario threshold. Every step is inspectable and rebuildable from public data.

That choice looked like it was trading away “official data” credibility. It turned out to be the opposite: full reproducibility — nothing taken on faith — is a stronger position for a climate argument than citing a restricted dataset, provided the methodology can stand on its own. It also meant the piece didn’t compete on the same crowded dataset most other entries were reaching for (see §7).

---

3. Approach: three scenes, one threshold

The piece steps through three sea-level scenarios for Funafuti — 0.0m baseline, 1.0m, and 1.5m — rendered as sticky maps against scrolling narrative text. An earlier version included a fourth scenario (0.5m), which was cut after the derived flood extents at 0.5m and 1.0m turned out to be visually indistinguishable — keeping a scenario that added no readable difference would have diluted the scroll pacing rather than strengthened it.

Coastline geometry comes from OpenStreetMap via osmextract’s reproducible PBF-based access, chosen over the Overpass API after rate-limiting made Overpass unreliable for repeated pipeline runs.

---

4. What the analysis finds

Modeled exposure. Land-area exposure nearly doubles between the 1.0m and 1.5m scenarios — the single figure the piece is built around. That jump, not the raw flood extent at either scenario individually, is the argument.

Observed warming signal. Sea surface temperature anomalies for Tuvalu (SST_ANOM, Pacific Data Hub, 1850–2025) show a clear warming trend accelerating in recent decades, reaching +0.9°C above baseline in 2024 — the warmest year in the observed record. This grounds the modeled scenarios in a measured local signal rather than leaving them purely hypothetical.

Sea-level anomaly, cited but not charted. The Pacific Data Hub’s sea-level anomaly series for Tuvalu is rounded to 0.1m resolution — too coarse to trend-chart honestly without fabricating precision the data doesn’t contain. It’s used as a cited figure instead (~+0.1m of observed rise since 1993; Funafuti tide gauge trend ~3.9 mm/yr), not plotted as a smooth line.

---

5. Piece structure: three scenes, not chapters

Scene 1 — Baseline

Establishes Funafuti as it stands today: the sticky map at 0.0m, narrative framing the threshold argument to come.

Scene 2 — 1.0m

The first scenario. Margins begin failing along the atoll’s narrowest points.

Scene 3 — 1.5m

The second scenario. Fragmentation begins — the same coastline, visibly broken rather than merely diminished. This is the scene the entire piece is built to arrive at.

A supporting slope chart (“A Half-Degree of Warming, Twice the Exposure”) and the observed-signals chart (SST trend) sit alongside the scroll narrative as evidence, not as additional scenes competing for attention.

---

6. Key design decisions

One threshold, not a framework
Feedback received after an early draft suggested adding a second axis — adaptation spending, a resilience score, something to “balance” the vulnerability story. It was declined. A piece built around one threshold doesn’t get stronger by bolting on a second dimension; it gets diluted into a different, more generic piece. The single spine was treated as the thing under protection, not a first draft waiting to be completed.

Provenance has to be visible in the artifact, not just true in the repo
Choosing an open, reproducible pipeline over restricted data is only a real advantage if a reader can see that choice without leaving the page. The methodology note and data-source footer are a deliberate part of the design, not an afterthought.

Match the chart to the data’s actual resolution
The sea-level anomaly series’ 0.1m rounding made a trend line dishonest — see §4. Citing it as a figure rather than forcing it into a chart it can’t support was the correct call, even though a plotted trend would have made for a “busier” piece.

Verify indicator codes against the data, not the filename
The temperature file is labeled “surface temperature” on download, but its indicator column reads SST_ANOM — sea surface temperature, not the broader global-mean-temperature indicator the filename implies. Every annotation citing that series was checked against the indicator code, not the file’s name.

---

7. Out of scope — by design

• A resilience/adaptation axis alongside the vulnerability story
• Additional scenario steps (0.5m) that didn’t produce a visually distinct flood extent from their neighbor
• Restricted SPC/NIWA flood rasters, in favor of a fully reproducible open pipeline
• A plotted trend line for the sea-level anomaly series, given its 0.1m-rounded resolution
• Framing the threshold as a eulogy for the island rather than a mitigation-stakes argument — the piece argues for closing the gap, not that the outcome is already settled

These were excluded to keep the piece saying exactly one thing, clearly — not because any of them lacked merit on their own.

---

8. Technical lessons from the build

theme: none breaks Closeread’s overlay layout

Closeread’s overlay-style sticky layouts depend on grid CSS that Quarto strips out under theme: none. Several overlay-layout attempts failed silently before the sidebar layout was accepted as the correct choice for this piece rather than a limitation to work around.

Terra SpatRaster objects don’t survive save()/load()

SpatRaster objects are pointers to external C++ objects — standard R serialization silently corrupts them across sessions. writeRaster() on save and rast() on load is the only reliable round-trip.

Closeread sticky visibility toggle

The default Closeread behavior is .sticky { opacity: 0 }, which is correct for production scroll-reveal but makes prototyping painful. Forcing .sticky { opacity: 1 !important } fixes visibility during development but breaks the reveal for every sticky element at once when left in. The correct pattern is a state-based toggle — .sticky.cr-active { opacity: 1 } — not a blanket override.

---

9. Outcome and reflection

Outcome

The piece is live and submitted to the Pacific Dataviz Challenge 2026, competing for the International Global Mention category as a non-Pacific entrant. Results aren’t out at time of writing, so this is a build reflection, not a victory lap.

Reflection

The hardest part of this project wasn’t the spatial pipeline — it was holding the line on a single argument once the piece was mostly built and every additional idea started sounding reasonable in isolation. The discipline to keep cutting rather than adding turned out to matter more than any individual technical decision documented above.

---

Project repository

• Interactive article: https://what-sits-in-the-difference.netlify.app/
• Source code: github.com/poncest/pacific-dataviz-2026 — private through the competition window, opening publicly after the submission period closes on August 31, 2026.

---

Appendix: Methodology & Build Notes

Data

• Pacific Data Hub .Stat Explorer, Climate Change Indicators: Sea Level Anomalies (SEA_LVL, Tuvalu, 1993–2023) and Sea Surface Temperature Anomalies (SST_ANOM, Tuvalu, 1850–2025)
• GLO-30 digital elevation model via elevatr
• Coastline geometry via OpenStreetMap (osmextract, PBF-based access)
• SST anomaly, 2024: +0.9°C above baseline (warmest year in the observed record)
• Sea-level anomaly: ~+0.1m observed since 1993; Funafuti gauge trend ~3.9 mm/yr (cited figure, not charted — 0.1m rounded resolution)

Technical Stack

• Language: R
• Spatial: terra, sf, elevatr, osmextract
• Visualization: ggplot2, tidyterra, showtext, ggtext
• Publishing: Quarto, Closeread v1.0.1
• Typography: Big Shoulders Display, DM Sans, JetBrains Mono
• Deployment: Netlify