El Niño Composites

You'll learn: branching graphs (one source feeding two processing chains), the enso node, filtering a stack by ENSO phase with selectFrames, collapsing each branch with timeReduce, and differencing the branches with math. The payoff is one of climate science's most famous pictures, built from nine nodes.

A composite answers "what does X typically look like during condition C?" by averaging every frame where C held. Composite the SST anomaly over El Niño months, do the same for La Niña months, subtract — and the equatorial Pacific's ENSO dipole should appear.

layer(anom) ──┬─► selectFrames(elnino) ─► timeReduce(mean) ─┐
              │        ▲ oni                                ├─► math(sub) ─► display + answer
              │        │                                    │
enso ─────────┴────────┴► selectFrames(lanina) ─► timeReduce(mean) ─┘

Build the trunk#

1. Add a layer node: layer: anom, start: 2016-01, stepMonths: 1. The anomaly layer (SST minus its 1971–2000 climatology for that place and month) is the right raw material — the seasonal cycle is already removed, so what's left is the unusual warmth or coolness.
2. Add an enso node (under sources). No params: it's the monthly Oceanic Niño Index, the standard ENSO record. Its output is teal — a series.

Branch one: El Niño months#

3. Add selectFrames (under transform). Connect the anomaly layer to value — and note its second, optional input oni?. Set phase: elnino. The node immediately shows a validation error: "phase needs the oni input (connect an enso node)". The phase filter can't classify months without the index — so connect enso → selectFrames.oni and watch the badge clear.
4. Add timeReduce with stat: mean and feed it the selectFrames output. This collapses "every El Niño month since 2016" into one average-anomaly map.

Branch two: La Niña months#

5. Repeat: another selectFrames (phase: lanina) fed by the same layer node and the same enso node — drag new edges from their output dots; one output can fan out to any number of inputs. Another timeReduce (stat: mean).

   If the canvas is getting crowded, drag node headers to arrange the two branches into rows, and pan/zoom as needed. Layout is yours; it never changes the program.

Combine#

6. Add math (under combine), fn: sub. Connect the El Niño mean to a and the La Niña mean to b. Field minus field, cell by cell — and the result is marked relative, so the display will pick a diverging colormap with a symmetric range automatically.
7. Add display (title: SST anomaly · El Niño − La Niña) and answer (label: El Niño − La Niña anomaly difference), both fed by the math node.

Run and read#

8. ▶ Run (done · ~33M cell-ops — the priciest tutorial yet, still under a second of compute once the data is cached).

   The map is the textbook: a broad red tongue along the equatorial eastern Pacific — during El Niño that band runs a couple of degrees warmer than during La Niña — flanked by a cooler horseshoe to the west and off-equator. You built the pattern that names the phenomenon, from raw monthly grids and an index, with no statistics beyond averaging and subtracting.

9. The honest print: the answer's global mean is small (~+0.3 °F) — ENSO redistributes heat more than it creates it — while p5/p95 (≈ −1 °F/+1.8 °F) show the swing where the action is. And a caveat worth knowing: selectFrames classifies phase by the instantaneous ONI ±0.5 threshold, a simplification of the official rule (which requires the threshold to hold for five consecutive seasons). Fine for composites; don't cite it as an event catalog.

The finished program#

{ "nodes": [
  { "id": "anom", "op": "layer", "params": { "layer": "anom", "start": "2016-01", "stepMonths": 1 } },
  { "id": "oni", "op": "enso" },
  { "id": "nino", "op": "selectFrames", "inputs": { "value": "anom", "oni": "oni" }, "params": { "phase": "elnino" } },
  { "id": "nina", "op": "selectFrames", "inputs": { "value": "anom", "oni": "oni" }, "params": { "phase": "lanina" } },
  { "id": "ninoMean", "op": "timeReduce", "inputs": { "value": "nino" }, "params": { "stat": "mean" } },
  { "id": "ninaMean", "op": "timeReduce", "inputs": { "value": "nina" }, "params": { "stat": "mean" } },
  { "id": "diff", "op": "math", "inputs": { "a": "ninoMean", "b": "ninaMean" }, "params": { "fn": "sub" } },
  { "id": "map", "op": "display", "inputs": { "value": "diff" }, "params": { "title": "SST anomaly · El Niño − La Niña" } },
  { "id": "ans", "op": "answer", "inputs": { "value": "diff" }, "params": { "label": "El Niño − La Niña anomaly difference" } }
] }

(It also ships as the built-in El Niño − La Niña pattern.)

Experiments: add months: "12,1,2" to both selectFrames nodes to composite winters only (ENSO peaks in boreal winter — the pattern sharpens); swap the layer to rain (coverage starts 2022-12, so shorten start) and see the precipitation shift instead; display one branch's mean directly to see a raw composite before differencing.

Next: Series, Charts & Lag — leaving maps for time series.