Geospatial AI and catastrophe modeling under one roof

  1. 2017Cambridge, Massachusetts
    Sketch of a red-brick nineteenth-century museum behind bare trees.

    Harvard

    Two of the cofounders begin their PhD program in Harvard’s Earth and Planetary Sciences Department. Both are applying Bayesian statistics and machine learning towards climate problems.

  2. The years after

    Aon, MLB, and NBA teams

    Ted spends years building hurricane catastrophe models at Aon. The cofounding team’s applied machine-learning experience includes work for MLB and NBA teams.

  3. 2026Ann Arbor, Michigan
    Sketch of the Downstream flag before a house, the ground behind it resolving into a fine digital grid of houses.

    Downstream

    “Because there’s plenty of scientific research left to be done to prove out this hunch, I’m cofounding Downstream with two brilliant applied ML experts. Together, we’re going to listen closely to Nature.”

    Team
    Ted Amdur, cofounder and CEO, and two cofounders in applied machine learning. Training and operating experience from Harvard, Berkeley, Aon, MLB, and NBA teams.
    Built so far
    A hurricane wind model run across a 12-storm library, roof damage labels for millions of hurricane-affected properties, and a component-based simulator for structural wind vulnerability.
    In training
    The deep learning model that predicts the susceptibility modifier.
  4. 1992 to today
    Photograph of a Dade County street after Hurricane Andrew: a house reduced to a pile of debris, with neighboring houses standing behind it, roofs largely intact.

    Why this is possible now

    Survey crews after Hurricane Andrew in 1992 saw exposed houses fail beside sheltered ones. That effect could not yet be quantified house by house.

    Imagery
    High-resolution public aerial imagery now covers the coast before and after major landfalls.
    Computational tractability
    GPUs make massively parallel image processing possible.
    Deep learning
    Models built off of state of the art approaches, trained on millions of observed outcomes, can learn how the houses and trees around a building affect losses.
Pen sketch of a shore neighborhood at sunrise, hip-roofed houses and palms above a beach.

Get in touch

We love feedback. If you’re working through interesting problems in wind risk, let’s chat.

ted@downstream.earth
marketing@downstream.earth

Updates by email

Sign up for monthly updates, new research releases, and live event analysis