Dear experts,

I am working on a project to detect fire based on the multi-spectral data.

My current process is manual - reviewing, based on the actual fire infirmation or based on data from MODIS, each image for the specific location and time period and flag them accordingly. Besides being time consuming, this has scope for many errors.

Is there a better option to download the fire and no-fire data (either image or the underlying bands). There are some eval script that does the fire masking (which I am following to revalidate), but I want to minimize the false positives. Without proper labeling, the rest of the code will not function properly. Any pointer would be highly appreciated.

My preference is to use the non-visible spectrum, but I am not able to find the right way to label them.

Also, when downloading .png or .jpg, what would be an ideal resolution, especially if you only have longitude and latitude? Does it make sense to add a bigger buffer for the bounding box or a smaller buffer? I tried the various combinations and many of them look very grainy. I also want to keep the file size reasonably small (say, under 100K per image)