Chris Mattioli

Active Research Endeavors

MIT Lincoln Laboratory, Air Traffic Control Group

The Offshore Precipitation Capability

Air traffic controllers rely heavily on weather radar information when making decisions regarding aircraft positioning. Unfortunately, the weather radar network that covers the national air space suffers from limited coverage offshore. The Offshore Precipitation Capability (OPC) was designed to address this gap of data. At its core, the OPC is a deep neural network trained to predict weather radar data based upon three dissparate sets of weather data that are available offshore. These data include geostationary satellite imagery, lightning strike location data, and numerical weather prediction model fields. Click the image to the right (or above if you're on mobile) for more information.

NWP Model Super-resolution

Numerical weather prediction (NWP) models are the physics resolving models of the atmosphere that are capable of forecasting many different quantities at long time horizons. Such models include the Global Forecasting System (GFS) and the ECMWF (European Centre for Medium-Range Weather Forecasts). These models are initialized with up-to-date data such as temperature, pressure, and radar reflectivity. However, the spatial resolution of these models is typically very large (0.25-0.5 degrees lat/lon). There is a need to enchance this resolution so that these forecasts can be used in a more tactical situation. Super-resolution (also known as statisitcal downscaling) is a method by which resolution is "enhanced" through machine learning. I've been currently attempting to perform super-resolution on these models with a high level of fidelity and accuracy.

Convective Weather Avoidance Modeling

There is a wealth of various weather data available to the air traffic controllers of the national airspace. These data not only include "time 0" analysis data, but also forecast data. Decision making at this level involves condensing this raw information in such a way that addresses safety and efficiency. A convective-weather-avoidance-model seeks to map this wealth of weather data into actionable information. It attempts to couple weather hazards with pilot weather-avoidance behavior, so that air traffic controllers can be confident in their action plans.

Active Research Interests

Bayesian Statistics & Inference

I'm a Bayesian thinker through and through. Not only in my work, but also as a member of society. There's just something so elegant about the expression of Bayes Rule, that you can instill beliefs into decision making. How in many statistical models, the notion of prior knowledge precipitates out as regularization. It's so cool! Be sure to check out my lecture on Variational Bayesian Inference as well as my lecture on the integration of Bayesian statistics in machine learning via the Variational Auto-Encoder.

Descision Making with Uncertainty

Borrowing from my interests in probability, statistics, and computer science, I find the field of automated decision support particularly interesting. It is union of mathematical models, efficient algorithms, and human factors (and more than likely messy data too). In my work with MIT Lincoln Laboratory, I've had the opportunity to work on these systems for FAA applications. Not only in their design assumptions, but also their implementation and deployment. How one is able to construct question/answer and cause/effect into a mathematical, programmable, and learnable structure is extremely fascinating to me.