Unsupervised framework to monitor lake dynamics

A system for more efficiently categorizing pixels in images of a surface, the system comprising:

• a memory containing features for each pixel in the images, such that for each sub-area of a geographic location captured by the images there is a time series of features;
• a processor performing steps of:
• determining a distribution for the time series of features for each sub-area;
• forming a graph linking neighboring sub-areas;
• for each pair of linked sub-areas, breaking the link between the two sub-areas based on differences in the distributions for the time series of features for the two sub areas;
• categorizing sub-areas with fewer than a threshold number of links to other sub-areas as a transition category;
• and
• categorizing sub-areas with at least the threshold number of links as one of at least two other categories.

A method for improving identification of land cover from satellite sensor values, the method comprising:

• a processor performing steps of:
• receiving satellite sensor values for a collection of sub-areas;
• forming a graph linking neighboring sub-areas in the collection of sub-areas;
• for each pair of linked sub-areas, breaking the link between the two sub-areas based on differences between the satellite sensor values of the two sub areas;
• categorizing sub-areas with fewer than a threshold number of links to other sub-areas as having a transition land cover;
• and
• categorizing sub-areas with at least the threshold number of links as having one of at least two other land covers.