Observing landscape changes from space
Generating time-lapses from satellite imagery in Python
Written by Anze Zupanc, Matej Aleksandrov, Matej Batič, Devis Peressutti, and Blaz Sovdat.
The Sentinel-2 satellite mission involves twin satellites orbiting the Earth in tandem, with phase of 180 degrees. This translates into revisit time of five days at the equator; as one moves towards the poles revisit time decreases. High revisit frequency of Sentinels adds another dimension to satellite imagery — time. We’ve become used to still images of stunning landscapes from all over the world. But being able to observe from space how landscapes change in time — either due to our influence or natural causes — is something new.
Generating time-lapses from scratch
Our Sentinel Hub services have offered this capability for quite a while, however, creating time-lapses from scratch requires a lot of manual work.
First, we have to retrieve the set of dates for which data of the area of interest (henceforth abbreviated AOI) are available. Our OpenSearch API is the way to achieve this. The API allows us to filter out images whose cloud coverage exceeds user-set threshold.
Second, once we’ve retrieved the list of available dates, we download an image for each date. This is done via a WMS or WCS request.
But now another issue pops up: as of this writing, the filter in OpenSearch uses the cloud coverage that was estimated over the whole tile (these estimates are provided by ESA). This means that while the tile might have low cloud coverage, our AOI — in general covering only a small fraction of the whole tile — could be almost completely covered by clouds. And vice versa — while the tile might have high cloud coverage, our AOI might be almost cloudless.
One way to alleviate this problem is to use a custom cloud detector to estimate the cloud coverage of the AOI on the fly. We use our recently released cloud detector to solve this problem.
The time-lapse script
We are happy to announce that the sequence of dull steps we’ve just outlined is completely automated by our time-lapse Python module which uses the open-source sentinelhub-py package to interact with Sentinel Hub services and sentinel2-cloud-detector package to estimate cloud coverage of the retrieved scenes.
Using the module
To use the module you need Python 3.5+ installation and instance ID associated with your Sentinel Hub account. If you don’t have an account, you can create a free trial one.
Time-lapse examples
Grand Ethiopian Renaissance Dam
The Grand Ethiopian Renaissance Dam is a gravity dam on the Blue Nile River. Its construction began in 2011 and as of this writing the dam is still under construction. When completed, it will be the largest hydroelectric power plant in Africa.
Mandrare River
The Mandrare River in southern Madagascar dries up in certain months of the year, as seen in the time-lapse.
Sevilla rice fields
The following time-lapse shows rice fields in Sevilla in false color.
Utah Mine Fields
In Moab desert in Utah, water is pumped underground to dissolve the potash ore into a brine. Highly-concentrated brine is then pumped into the evaporation ponds shown in the following time-lapse. As darker water absorbs more sunlight, the water is dyed bright blue to reduce the amount of time it takes for the potash to crystallize.
