Whenever you fly over a crop field to obtain photographs for photo stitching you should always try to fly at a time when there are no clouds in the sky. If clouds are present, they can effect your imagery in several negative ways as illustrated below. As an industry leader in agricultural drone photogrammetry, Agribotix has coined four different terms for the four different types of cloud effects: cloud shadow, sun spot, overcast shadow, and overcast burn. These are illustrated below and discussed in detail. Although clear skies are ideal, they are not a necessity; flying under an overcast cloud layer that is uniform in thickness & density will yield good results. Unfortunately, it can be difficult to discern exact characteristics of overcast clouds and so flying under them can posit some risk of producing inaccurate results. At Agribotix, our machine vision algorithms are trained to spot these inaccurate results by spotting various signatures in the customer submitted photo sets as well as in the final stitched imagery. Although we can sometimes correct these deficiencies ourselves, we advocate that our customers submit good photo sets from the start to minimize errors as those photos make their way through our imagery processing pipeline.
Below is an illustration of a classic cloud shadow. A cloud shadow occurs when part of the ground is covered by a cloud’s shadow and that covered area shows up in a photograph. Cloud shadows are the easiest cloud effects to recognize when looking through a photo set before they are submitted to us for image processing on our web app. If you happen to be flying on a day when clouds are present, take a look through your photographs to see if you have any images with cloud shadows in them.
Below is a photo set taken with a near infrared (NIR) camera and therefore the crop has a pink (rather than green) color to it. Notice that images 2506 (IMG_2506.JPG) through 2520 have darkened areas in them. These dark areas are cloud shadows. If cloud shadows are present they will produce inaccurate results. While other stitching services offer cloud shadow removal capabilities that ‘photoshop’ cloud shadows out of stitched imagery entirely, the Agribotix data analytics team has found these capabilities to work well on the naked eye only; false color NDVI & DVI maps require precision input and cloud shadows corrupt that input process. If your photo set has enough oversampling then photos with cloud shadows can be removed before uploading them to our web app. Otherwise, if there is not enough oversampling to compensate for the missing photos, the entire set must be uploaded with the resulting errors addressed after the fact.
The second most common cloud effect is something called a sun spot. Sun spots and cloud shadows are fundamentally the same thing except sun spots occur when you are flying under an overcast cloud layer and the sunlight occasionally pokes through the cloud deck. The effects are the same except instead of producing the occasional dark spots (as shown in the photoset above) sun spots will manifest as light spots in your photo set.
Below are a series of photographs with cloud effects in them. Photos 1263 (IMG_1263.JPG) through 1295 all contain either cloud shadows or sunspots depending on how you interpret them. If the rest of the photos all had the same pigmentation as image 1296, we would call these other photos cloud shadows. If the rest of the photo set had the same pigmentation as the majority of image 1291, we would call these cloud shadows. The last image (IMG_1300.JPG) shows an overcast burn which is discussed in the last section below.
The schematic of an overcast shadow is shown below. Overcast shadows can be hard to detect with the naked eye and machine vision algorithms will often times have to be employed to catch specific signatures they leave behind. In the false color (NDVI & DVI) maps they show up as regions which appear to have relatively weaker crop health that those areas not affected by the overcast shadow.
The last and least common cloud effect is the overcast burn. As with the overcast shadow, the overcast burn is caused by variation in cloud thickness. In the false color (NDVI & DVI) maps they show up as regions which appear to have relatively stronger crop health that those areas not affected by the overcast burn and so in this way they invert the effects of the cloud shadow.