Methodology

Data collection

Niño regions. NOAA CPC publishes weekly sea surface temperatures. We check every six hours.

La Jolla. Scripps / SCCOOS measures pier water temperatures about every four minutes. We fetch quality-passed daily means hourly.

Data normalization

Both feeds become dated Celsius observations. Missing La Jolla days are linearly interpolated between valid readings before averaging complete Sunday–Saturday weeks, dated Wednesday. Its headline keeps the latest daily mean; today’s UTC value is partial.

Chart values are interpolated onto a shared 365-day calendar. Historical means weight years equally, excluding the oldest and newest years; shading shows one and two standard deviations.

Similarity

We compare this year’s available temperatures with the same calendar days in complete historical years using root mean square error (RMSE). Lower error means a closer match.

Projection

Using the analog method, we start at the latest chart value and follow the average remaining-year temperature change of the five closest historical years. This is a historical estimate.

News

Each day we read feeds from KQED, Los Angeles Times, KPBS, Times of San Diego, Civil Beat, NASA Earth Observatory, BBC, The Guardian, Al Jazeera, RNZ.

Jev (typesafe/jev-1.13) classifies El Niño and Pacific weather relevance, regional fit, and tone, then checks duplicate events. The small LLM GPT-6 Luna (openai/gpt-6-luna) writes short summaries; publisher excerpts are kept when summarization is unavailable.