You are provided with several sets of training data (tabular data); one with the values of remote sensing reflectance (year and dry season composites from Harmonized Landsat Sentinel imagery) as well as bioclimate data for precipitation and temperature, relief (elevation, slope, aspect, topographical position), canopy height among others, for each record in the training data set. A second set with the original noisy vegetation labels, and a third set with revised labels that can be used to improve the original data. This second data set (improved labels) has been thoroughly but NOT exhaustively checked, so it can also contain a number of wrong labels; so part of the challenge is to have a robust method to improve the training data.

The data provided correspond to one tile of the Mexican Geospatial Data Cube; about 150 Km by 150 Km, in the Pacific Coast of Mexico, almost centered in Puerto Vallarta. It contains a variety of ecosystems; going from tropical semidry to humid forests, as well as temperate forests, grasslands (cultivated and induced), mangroves and other wetlands, several types of agriculture and built up areas.

You are free to apply the ML techniques that you consider more convenient to achieve the best possible results.