40:null 41:["$","title",null,{"children":"Air Quality reports in South Portland | Sri Krishna Koushik Komanduru | Observable"}] 47:T6b0,{ const DIRS = ["N","NE","E","SE","S","SW","W","NW"]; const hue = d => d.dir === "S" ? "#b8322c" : "#bfb8ae"; const RMIN = 2.4, RMAX = 4.0; const wrap = (title, note, plot) => htl.html`

${title}

${note}

${plot}

`; const counts = Plot.plot({ width: 380, height: 290, marginLeft: 40, marginRight: 44, x: {label: null, grid: true}, y: {domain: DIRS, label: null}, marks: [ Plot.barX(byDirection, {y: "dir", x: "reports", fill: hue, tip: true}), Plot.text(byDirection, { y: "dir", x: "reports", text: d => d3.format(",")(d.reports), dx: 4, textAnchor: "start", fontSize: 10, fill: "#555" }), Plot.ruleX([0]) ] }); const rates = Plot.plot({ width: 380, height: 290, marginLeft: 40, marginRight: 44, x: {label: null, grid: true, domain: [RMIN, RMAX]}, y: {domain: DIRS, label: null}, marks: [ Plot.ruleY(byDirection, {y: "dir", x1: RMIN, x2: "rate", stroke: "#e2ded6", strokeWidth: 6}), Plot.dot(byDirection, { y: "dir", x: "rate", r: 5.5, fill: hue, stroke: "white", strokeWidth: 1, tip: true }), Plot.text(byDirection, { y: "dir", x: "rate", text: d => d.rate.toFixed(2), dx: 11, textAnchor: "start", fontSize: 10, fill: "#555" }) ] }); return htl.html`

${wrap("Total reports", "Raw count — south dominates", counts)} ${wrap("Reports per day of that wind direction", "Normalized — axis truncated at 2.4", rates)}

`; }48:T764,{ const total = d3.sum(byDirection, d => d.reports); const totalDays = d3.sum(byDirection, d => d.days); const slope = total / totalDays; const hue = d => d.dir === "S" ? "#b8322c" : "#6b6b6b"; const xmax = d3.max(byDirection, d => d.days) * 1.12; const ymax = d3.max(byDirection, d => d.reports) * 1.12; return htl.html`

Reports are explained by how often the wind blows

Each point is one wind direction. The dashed line is what you would expect if every day produced the same number of reports regardless of direction. All eight sit within about 20% of it, so the raw count mostly reflects wind frequency rather than any directional effect.

${Plot.plot({ width: 660, height: 420, marginLeft: 54, marginBottom: 44, x: {label: "Days with that wind direction →", domain: [0, xmax], grid: true}, y: {label: "↑ Total reports", domain: [0, ymax], grid: true}, marks: [ Plot.line([[0, 0], [xmax, xmax * slope]], { stroke: "#999", strokeDasharray: "4,4" }), Plot.dot(byDirection, { x: "days", y: "reports", r: 6, fill: hue, stroke: "white", strokeWidth: 1, tip: true, channels: { direction: "dir", days: "days", reports: "reports", "per day": d => d.rate.toFixed(2) } }), Plot.text(byDirection, { x: "days", y: "reports", text: "dir", dx: 13, dy: -6, textAnchor: "start", fontSize: 11, fill: "#444" }) ] })}

Dashed line: ${slope.toFixed(2)} reports per day, the overall average. South highlighted.

`; }49:T4c8,{ const byMonth = d3.rollups( reports, v => v.length, r => d3.utcMonth(r.datetime) ).map(([month, n]) => ({month, n})).sort((a, b) => a.month - b.month); const peak = byMonth.reduce((a, b) => b.n > a.n ? b : a); const overall = reports.length / new Set(reports.map(r => r.dayKey)).size; return htl.html`

Reports per month, 2019–2026

Reporting peaks at ${peak.n} in ${d3.utcFormat("%B %Y")(peak.month)} and declines steadily after. This cannot distinguish improving air quality from fading app engagement — the data records complaints, not conditions. Overall rate across the whole record: ${overall.toFixed(2)} reports per reporting day.

${Plot.plot({ width: 820, height: 270, marginLeft: 46, marginBottom: 34, x: {label: null}, y: {label: "↑ Reports", grid: true}, marks: [ Plot.rectY(byMonth, { x: "month", y: "n", interval: d3.utcMonth, fill: "#6b8f5e", fillOpacity: 0.85, tip: true }), Plot.ruleY([0]) ] })}

`; }4a:T14c5,{ const DIRS = ["N","NE","E","SE","S","SW","W","NW"]; const ord = n => n + ({1:"st", 2:"nd", 3:"rd"}[n] || "th"); let pickLocal = "S"; const geoFor = dir => ({ type: "FeatureCollection", features: reports.filter(r => r.compass === dir).map(r => ({ type: "Feature", geometry: {type: "Point", coordinates: [r.longitude, r.latitude]}, properties: {value: r.value} })) }); const radio = html`

`; const cards = html`

`; const story = html`

`; const caption = html`

`; const mapDiv = html`

`; DIRS.forEach(dir => { const b = html``; b.onclick = () => { pickLocal = dir; render(); }; b.dataset.dir = dir; radio.append(b); }); const map = new maplibregl.Map({ container: mapDiv, style: "https://basemaps.cartocdn.com/gl/positron-gl-style/style.json", center: [-70.272, 43.636], zoom: 12.2, attributionControl: true }); map.addControl(new maplibregl.NavigationControl(), "top-right"); requestAnimationFrame(() => map.resize()); const ro = new ResizeObserver(() => map.resize()); ro.observe(mapDiv); let ready = false; map.on("load", () => { map.addSource("reports", {type: "geojson", data: geoFor(pickLocal)}); map.addLayer({ id: "heat", type: "heatmap", source: "reports", maxzoom: 15, paint: { "heatmap-weight": ["interpolate", ["linear"], ["get","value"], 1, 0.2, 5, 0.8], "heatmap-intensity": ["interpolate", ["linear"], ["zoom"], 10, 0.35, 15, 1.1], "heatmap-radius": ["interpolate", ["linear"], ["zoom"], 10, 4, 12, 9, 15, 22], "heatmap-opacity": ["interpolate", ["linear"], ["zoom"], 13, 0.8, 15, 0.35], "heatmap-color": ["interpolate", ["linear"], ["heatmap-density"], 0, "rgba(0,0,0,0)", 0.08, "rgba(43,138,74,0.35)", 0.3, "#4fae4a", 0.55, "#a8cf3c", 0.8, "#e3e021", 1, "#fff45a"] } }); map.addLayer({ id: "pts", type: "circle", source: "reports", minzoom: 13, paint: {"circle-radius": 3.5, "circle-color": "#1f6b3a", "circle-opacity": 0.7} }); ready = true; render(); }); function render() { const d = byDirection.find(x => x.dir === pickLocal); const rank = key => [...byDirection].sort((a,b) => b[key]-a[key]) .findIndex(x => x.dir === pickLocal) + 1; const rCount = rank("reports"), rRate = rank("rate"); radio.querySelectorAll("button").forEach(b => { const on = b.dataset.dir === pickLocal; b.style.background = on ? "#b8322c" : "#fff"; b.style.color = on ? "#fff" : "#333"; b.style.borderColor = on ? "#b8322c" : "#d6d2c8"; }); const card = (label, value, r, note) => `

${label}

${value}

${ord(r)} of 8 directions

${note}

`; cards.innerHTML = card("Total reports", d3.format(",")(d.reports), rCount, `Filed on ${d.days} days with ${pickLocal} wind.`) + card("Reports per day", d.rate.toFixed(2), rRate, `Same reports, divided by how often ${pickLocal} wind actually blew.`); story.innerHTML = pickLocal === "S" ? `Southerly wind looks like the culprit — most reports of any direction. But southerly days are also the most common in the record, 362 of 1,428. Per day it drops to 7th of 8. The raw count was measuring the weather, not the smell.` : `${pickLocal} wind ranks ${ord(rCount)} by total reports and ${ord(rRate)} per day. ${ rCount !== rRate ? `The two disagree because ${pickLocal} wind occurred on ${d.days} of 1,428 reporting days — ranking by count is partly ranking by frequency.` : `Here the two happen to agree.`}`; caption.innerHTML = `Report density — ${pickLocal} wind. ${d3.format(",")(d.reports)} of ${d3.format(",")(reports.length)} reports. Zoom past level 13 for individual points.`; if (ready) map.getSource("reports").setData(geoFor(pickLocal)); } render(); invalidation.then(() => { ro.disconnect(); map.remove(); }); return html`

${radio}${cards}${story}${caption}${mapDiv}

`; }4b:T8d0,md`## Conclusions **Report volume by wind direction proves nothing.** Southerly wind leads the raw count by a wide margin — 1,048 against 817 for the next highest — which looks like evidence of a source to the south. It isn't. Southerly days are simply the most common in the record, 362 of 1,428 reporting days. Divide by that and south drops to 2.90 reports per day, seventh of eight. Any argument built on the raw count is really an argument about Portland's prevailing winds. **Once normalized, direction barely separates at all.** The full range across eight directions is 2.63 to 3.82 reports per day, against an overall average of 3.14 — roughly ±20%. SE is highest and NE lowest, but the spread is small enough that picking a new winner would be overreading. The honest conclusion is that this dataset does not isolate a directional effect, not that a different direction is guilty. **Reporting volume falls steadily after 2021.** Monthly counts peak in 2020–21 and decline through 2026. This cannot distinguish improving air quality from fading app engagement, and the file contains nothing that would let you choose between them. **Report locations barely shift between wind directions.** Selecting any direction on the map produces roughly the same cluster in inland South Portland. That is itself a finding: reports trace where people live, not where the smell is. A geographic argument about sources would need exposure data, not complaint data. **What this analysis cannot do.** It cannot identify a source. Wind is a daily average rather than a reading at the moment of each report, coordinates are deliberately fuzzed, and complaints are a proxy for exposure with unknown bias. The denominator in every rate above is *reporting days*, not all days, because the file has no wind record for days when nobody reported — so I cannot fully separate "southerly wind is common" from "southerly days are more likely to generate at least one report." **One discrepancy worth raising.** The original app states over 7,000 reports since 2020. Its own histogram begins in 2019. This CSV has 4,495 rows beginning May 2019. Three different figures for what should be the same collection — worth clarifying before anyone cites a count.`4c:T1b82,md`## Design notes (dataviz cheatsheet) ### Plan **Why am I making this?** Exploration, then communication. I opened the file with no hypothesis, plotted things until something moved, and found that the obvious answer was wrong. The final figure exists to show that. **Who for?** A technical reader who knows the domain but hasn't seen this dataset. Residents and city officials are the downstream audience for this kind of analysis, though I haven't designed for them specifically — they'd need the geography, not the methodology. **Constraints.** Lives on a web page, so it can be interactive — tooltips on every mark. Full color available. Rendered as vector SVG, so resolution is device-independent and it holds up projected or printed. **Software.** Observable Plot, with d3 for number formatting. **What story does it tell?** Southerly wind produces the most odor reports by a wide margin — 1,048, against 817 for the next highest. That looks like evidence of a source to the south. It isn't. Southerly days are simply the most common days in the record: 362 of 1,428. Divide by that and south drops to 2.90 reports per day, seventh of eight. The more careful conclusion is also the less exciting one — wind direction barely separates at all once normalized. The whole range is 2.63 to 3.82. **Expected caption.** "Southerly wind accounts for the most odor reports (left) only because southerly days are the most common. Per day of each wind direction (right), southerly wind ranks seventh of eight." **Where are the bodies?** Several, and they matter more than the charts. \`value\` is a 1–5 severity the reporter picks in the app. It's a judgment, not a measurement, and 61% of reports are a 4 or 5 — so it records how bad it was *given someone bothered to report*, which is not the distribution of odors in the air. \`compass\` and \`bearing\` come from NOAA, but they're the *prevailing* direction for the whole day. I checked this rather than assuming: zero of 1,438 reporting days contain more than one bearing. A report at 3am and one at 11pm on the same day carry identical wind. That's a 24-hour resolution on my main explanatory variable, and it will blur any real relationship. Coordinates are deliberately skewed within a short radius for privacy, so neighborhood patterns are fair game and street-level ones aren't. Repeat submitters can't be linked, so a spike might be one persistent resident rather than fifty different people. And 12 of 4,495 rows have no wind direction; I drop them. **Who does it affect? Who is left out?** Anyone without the app — no smartphone, never heard of it, language barrier. 83% of reports carry zipcode 04106, and nothing in this file can tell me whether that reflects worse air or just higher app adoption. The honest framing is that this measures who complains. Exposure is what you'd actually want, and complaints are a proxy for it with unknown bias. ### Identify **Data types.** Quantitative: \`latitude\`, \`longitude\`, \`gust\`, \`timestamp\`. Quantitative but circular: \`bearing\`, where 0° and 360° are the same direction — a linear scale would put them at opposite ends, which is why I binned to \`compass\` instead. Ordered qualitative: \`value\` (1–5) and \`compass\` (cyclic). Unordered qualitative: \`zipcode\`, \`description\`, \`symptoms\`. **Function of the graphic.** Comparing specific values across eight categories, and setting two quantities side by side so the reader can see one contradict the other. **Suitable plot types.** Horizontal bars for the counts, since they start at zero and length reads honestly. A dot plot for the rates, because the values sit in a narrow band and bars from zero would make eight different numbers look identical. ### Encode **Marks.** \`barX\` for total reports. \`dot\` on a light \`ruleY\` track for rates. \`text\` for a value at the end of every mark. **Channels.** Position on a common scale does the work in both panels — it's the channel people read most accurately. Direction sits on a shared categorical y axis, so a row is the same direction in both panels and the eye can track across. Color does exactly one job: marking south, the direction under examination. It encodes attention, not magnitude. **One deliberate asymmetry.** The rate axis starts at 2.4, not zero. That's defensible for dots, which encode position, and necessary because all eight rates fall between 2.63 and 3.82 — from zero they'd be indistinguishable. It would not be defensible for bars, where length implies magnitude. I say so on the chart rather than hoping nobody checks. ### Add information **Does the caption describe the how and what?** Each panel has a title stating what's measured and a subtitle stating how to read it — "Raw count — south dominates" and "Normalized — axis truncated at 2.4." The truncation is disclosed in the figure itself, not buried in body text where a skimmer would miss it. **Are the labels clear?** Values sit at the end of each bar and beside each dot, so nobody has to trace back to an axis. Compass abbreviations are standard and need no explanation. There's no color legend because color isn't encoding a variable — one category is highlighted, and the reader can see which. ### Revise **Is it expressive?** It shows the two quantities the argument needs and nothing more. No third variable smuggled in via size or hue. **Is it effective?** Now, yes — it took three attempts. The first faceted both measures onto one shared x scale, which squashed the rate panel into nothing next to counts reaching 1,048. The second gave rates their own scale but kept bars, so a 2.63-to-3.82 spread looked like eight bars of roughly equal length. The third separated the scales and switched rates to dots with a truncated axis. Worth recording, because each failure was a real encoding mistake and not just a styling preference. **Any chartjunk?** No shading, no 3D, no redundant legends. The gray rule behind each dot is a track connecting the label to its category, which is functional rather than decorative. **Are comparisons easy?** That's the entire design. Shared row ordering plus one highlighted direction means the reversal between panels reads in a single glance, with no legend and no reading. **Are the data faithfully represented?** With two disclosed caveats. The truncated rate axis, stated on the chart. And the denominator is *reporting days*, not all days — the file has no wind record for days when nobody reported, so I can't fully separate "southerly wind is common" from "southerly days are more likely to generate at least one report." Both are on the chart or in this section rather than papered over. **Does it tell a meaningful data story?** Yes, though a methodological one rather than an environmental one. A count that looks like evidence about a source direction turns out to be an artifact of how often that wind blows. The conclusion I ended up with is weaker than the one the raw chart suggested, and more useful for it.`4d:[{"id":"dc2d1af9291d0d4e60c1484ba145a5b7517658331b8883ed8eb02ac18cc76f231d0d7abc19aa6327e0ce6e39b6cefc4e3c92018ec69454e4a57eae3c4261c4c5","href":"https://static.observableusercontent.com/files/dc2d1af9291d0d4e60c1484ba145a5b7517658331b8883ed8eb02ac18cc76f231d0d7abc19aa6327e0ce6e39b6cefc4e3c92018ec69454e4a57eae3c4261c4c5","name":"stinky-reports.csv","type":"text/csv","lastModified":1789930067531,"size":505519}] 45:{"id":"d1d3cb73ed4d2838","slug":"air-quality-reports-in-south-portland","title":"Air Quality reports in South Portland","thumbnail":"0bff04610ad49cee777ae6b83c294a2434185dbcb466f5e8617d2efa15281d0c","access_level":"public","create_time":"$D2026-09-21T16:57:39.390Z","update_time":"$D2026-09-23T00:25:12.856Z","view_time":null,"first_public_time":"$D2026-09-23T00:04:12.709Z","has_unpublished_edits":null,"fork":false,"stars":0,"starred":false,"owner":{"type":"team","id":"b4c94c4c66391089","login":"skkoushik453","name":"Sri Krishna Koushik Komanduru","avatar_url":"https://avatars-next.observableusercontent.com/ada9707d11fdb3d7c43484f22852a01ece294f13ebec50e76979662aaa1f2f1c","tier":"starter_2024","bio":"","following":false},"version":139,"publish_version":139,"role":null,"authored":false,"authors":[{"type":"individual","id":"ba606d67d08a0ced","login":"skkoushik453","name":"Sri Krishna Koushik Komanduru","avatar_url":"https://avatars-next.observableusercontent.com/ada9707d11fdb3d7c43484f22852a01ece294f13ebec50e76979662aaa1f2f1c","tier":null,"bio":"","following":false}],"members":[{"type":"team","id":"b4c94c4c66391089","login":"skkoushik453","name":"Sri Krishna Koushik Komanduru","avatar_url":"https://avatars-next.observableusercontent.com/ada9707d11fdb3d7c43484f22852a01ece294f13ebec50e76979662aaa1f2f1c","tier":"starter_2024","role":"editor"}],"comments":[],"custom_thumbnail":false,"trashed":false,"lapsed":null,"subscribed":null,"forks":0,"collection":null,"type":"notebook","body":{"titling":"auto","title":"Air Quality reports in South Portland","cells":[{"id":1,"value":"md`# Air Quality reports in South Portland\n\n4,495 resident-submitted smell reports from May 2019 to September 2026, joined with daily NOAA wind data.\n\nThe obvious reading, that southerly wind brings the smell is an artifact of how often the wind blows from the south. But report *locations* do shift with wind direction, reversing around a fixed point, which is what a real source would produce.\n\nData from the 02-Story class notes, collected via SmellMyCity. Prior work: [the Stinky app](https://cs7290.github.io/stinky/) (CS7290, Fall 2022) and the [DS5110 project](https://github.com/ds5110/stinky).`","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","tool":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":2,"value":"data = FileAttachment(\"stinky-reports.csv\").csv({typed: true})\n","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","tool":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":3,"value":"Inputs.table(data)","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","tool":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":4,"value":"({rows: data.length, cols: Object.keys(data[0])})","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","tool":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":7,"value":"reports = data\n .filter(d => d.compass && Number.isFinite(d.value))\n .map(d => {\n const local = new Date(d.datetime.getTime() - 4 * 3600 * 1000);\n return {\n ...d,\n compass: d.compass.trim(),\n hour: local.getHours(),\n month: local.getMonth(),\n year: local.getFullYear(),\n dayKey: d.day instanceof Date ? d.day.toISOString().slice(0,10) : String(d.day)\n };\n })","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","tool":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":8,"value":"byDirection = {\n const DIRS = [\"N\",\"NE\",\"E\",\"SE\",\"S\",\"SW\",\"W\",\"NW\"];\n const dayDir = new Map();\n for (const r of reports) dayDir.set(r.dayKey, r.compass);\n\n const days = d3.rollup(Array.from(dayDir.values()), v => v.length, d => d);\n const rpts = d3.rollup(reports, v => v.length, d => d.compass);\n\n return DIRS.map(dir => ({\n dir,\n reports: rpts.get(dir) ?? 0,\n days: days.get(dir) ?? 0,\n rate: (rpts.get(dir) ?? 0) / (days.get(dir) ?? 1)\n }));\n}","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","tool":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":9,"value":"Inputs.table(byDirection, {format: {rate: d3.format(\".2f\")}})","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","tool":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":10,"value":"$47","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","tool":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":38,"value":"$48","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","tool":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":36,"value":"$49","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","tool":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":15,"value":"maplibregl = {\n const gl = await import(\"https://esm.sh/maplibre-gl@4\");\n const href = \"https://esm.sh/maplibre-gl@4/dist/maplibre-gl.css\";\n if (!document.querySelector(`link[href=\"${href}\"]`)) {\n const link = document.createElement(\"link\");\n link.rel = \"stylesheet\";\n link.href = href;\n document.head.appendChild(link);\n }\n return gl.default ?? gl;\n}","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","tool":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":18,"value":"$4a","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","tool":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":14,"value":"$4b","pinned":true,"hidden":false,"mode":"ts","author":"$undefined","summary":"$undefined","tool":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":11,"value":"$4c","pinned":true,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","tool":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"}],"files":"$W4d","stdlib":"2"}} 46:"$45:body"