A lightweight, open-source, early-warning AI framework for near-real-time (<1h) landslide risk forecasting to prevent landslide fatalities around the world. Made in Africa.

Every year, landslides kill over 4,500 people globally, with developing nations bearing a large fraction of casualties despite having the fewest resources for prediction, prevention, and relief.

REQUEST A DEMO

ATLAS can predict landslides with high accuracy using limited data, specifically designed for deployment in resource-constrained regions of the world (namely, sub-Saharan Africa). We modeled landslide risk through mathematically rigorous techniques while making major innovations along the way.

“After a landslide, one more can be expected.”

— Faroese Proverb

Help us save lives.

We hope to work with regional partners to implement this free system for public benefit.

For technical details on the model and methodology, see the "Model Architecture & Training" section of the model.ipynb notebook (https://github.com/shyagehike/atlas-v2/blob/main/model.ipynb). For licensing & legal information, see https://github.com/shyagehike/atlas-v2/blob/main/LICENSE & https://tlo.mit.edu/understand-ip/exploring-mit-open-source-license-comprehensive-guide

GITHUB

CONTACT

  
    $$
\log \mathcal{L}(\lambda)
   = \sum_{i=1}^{n} \log \lambda(x_i)
     \;-\;
     \int_{\mathcal{X}} \lambda(x)\, dx.
$$

  

  
    $$
\log \mathcal{L}(\lambda)
   = \sum_{j:\, y_j = 1} w_j \log \lambda(x_j)
     - \sum_{j:\, y_j = 0} w_j \lambda(x_j)
     - \Omega(w).
$$
  

  
    $$
\mathcal{J}(w,b)
   = -\sum_{j:\, y_j = 1} w_j \mu(x_j)
     + \sum_{j:\, y_j = 0} w_j \exp(\mu(x_j))
     + \Omega(w).
$$

  

  
    $$
\mathrm{NLL\_per\_pos}
   = \frac{
       -\sum_{y_j=1} w_j \mu(x_j)
       + \sum_{y_j=0} w_j e^{\mu(x_j)}
       + \Omega(w)
     }{
       \sum_{y_j=1} w_j
     }.
$$

  

Help us save lives.

Let’s partner up.