Imagine trying to compare two clouds in the sky; regardless of how similar they appear, no two clouds are ever identical. Clouds have abstract shapes that anyone can recognize, yet it is very challenging to describe precisely what shape they have. The odd shapes that clouds could take are what make it hard to teach a computer what a 'normal' cloud should look like. Scientists face difficulties in automating the revision of incoming rainfall images captured by satellites.

Satellites can provide vital rainfall data and help the public measure rain globally. The number of satellite images available today is growing much faster than the number of trained experts available to review them. Without accurate checks, misleading rainfall information might prevent timely warnings for dangerous storms, floods, or hurricanes.

To address this problem, we developed a tool using artificial intelligence (AI). This tool utilizes a type of computer algorithm called Neural Networks to detect mistakes or unusual rain patterns (we call them artifacts). Additionally, the algorithm's performance can improve with a small amount of feedback from scientists in charge when it fails. Tests using two important satellite sensors (named SSMIS and SSMI) show promising results.