My wife said to me the other day, “The Celtics broke AI.” At first I had to wonder what she meant. Since the start of the NBA playoffs, many pundits, analytical predictions and bookmakers have considered the Boston Celtics as favorites not only to win the Eastern Conference, but also the outright NBA Championship. My spouse’s astute observation was made after many of those predictions held firm even when the Celtics fell 0-2 to the Miami Heat. In all of this there are lessons about weather forecasting and occasional “raids”.

First, let me outline some generic steps in how weather forecasts are made, and then build my analogy with the Celtics and the Heat. The key steps in weather forecasting are:

1. Collection of observations of the current state of the atmosphere and boundary conditions (weather balloons, satellites, aircraft observations, automated meteorological stations, buoys, etc.).
2. Quality control of those observations.
3. Plotting data (weather maps, soundings, meteograms).
4. Initializing computer models with observations using a basic “physical” and “mathematical” understanding of the atmosphere and how it is expected to evolve.
5. Assimilation of new information (through computer algorithms or human interpretation).
6. Analysis and interpretation based on historical information, analogs and model output.
7. Presentation and verification of forecasts.

As I have written many times, the public’s perception of the accuracy of weather forecasts often does not match reality. As meteorologists, we often hear about how “weather forecasts are always wrong” or that “it must be nice to be in a profession where you’re wrong 50% of the time”. Honestly, these statements often say more about the person than the actual facts. Studies show that many people struggle with the concepts of probability and uncertainty, but we use them all the time in weather forecasting. I commonly see people misinterpreting things like “20% chance of rain” or “hurricane cone of uncertainty”.

According to NOAA Web page, “A seven-day forecast can accurately predict the weather about 80 percent of the time, and a five-day forecast can accurately predict the weather approximately 90 percent of the time. However, a 10-day—or longer—forecast is accurate only half the time.” It’s human nature to think back to the occasional “freak” that affected your daughter’s outdoor wedding or the big game. Now back to the Celtics and Heat.

Minus 0-2 in the series, a little analysis tools they still gave the Celtics a 65% chance to win the series. Hell, some of them tools still give the Celtics a better than 30% chance to come back from 0-3. Like weather forecasting techniques, I’m sure these analytics are based on data collection, analysis, and historical trends. However, they do not check. Here’s where it gets tricky. Even with the best weather models, there will always be some uncertainty and potential for “misses”.

Weather models cannot always represent all the details of the atmosphere due to computational limitations. For example, a model with a grid spacing of 10 miles will miss a minor thunderstorm or small processes that may cause it. To alleviate these blind spots, a “parameterization” is used to broadly represent the processes that the model cannot “see”. In the sports analytics models that project the Heat – Celtics series, it is clear that some processes are not solved or revealed by numbers. I’m not sure they properly “parameterized” Jimmy Butler’s tenacity and will or the contributions of a roster consisting of several raw players.

Another aspect of weather forecasting is data assimilation. Here’s what the European Center for Medium-Range Weather Forecasts (ECMWF) Web page says about assimilation – “The method used at ECMWF for the atmosphere is called Four-Dimensional Variational Data Assimilation (4D-Var)….iteratively adjusts short-term forecast initial conditions to approximate meteorological observations in space and time. Adjustments are made in a way that respects the laws of physics…” Advances in assimilation techniques have dramatically improved weather forecasting. For example, there are scientific studies that show that the effectiveness in predicting Hurricane Sandy (2012) was related to parameterization as well as the quality of the satellite data used initialization or assimilation. If you’re watching the 2023 NBA playoffs, you’ll wonder how much updated information about the Heat and Celtics is being “assimilated” into the analytics to adjust for the initial conditions.

While most of my analogies so far have been based on technical aspects, there is something else I see in time as well. This is called “giving wishes”. Often the weather forecast is very clear about what will happen, but people “wish” because they want a certain outcome. Maybe they love the snow, hope their cooking won’t fail, or want to get in that round of golf. I often wonder if there is a bit of “wanting” in the game with the current Celtics – Heat matchup as well.

As I write this piece, Game 4 of the series will be played tonight. The Los Angeles Lakers were just swept by the Denver Nuggets. Will we see another purge tonight? As the NBA Finals are not scheduled to start until June, it is clear to me that, based on projections and “NBA climatology”, the NBA and television planners expected longer series. That’s the bane of forecasting.