How can I know how many processing units I need and how to calculate the cost of my requests?

Each request is worth a certain amount of processing units. If we only focus on rate limiting, the subscription package you need depends on how many processing units (PU) and requests per minute or month you intend to use. This in turn depends on how large your images are in pixels, how many of them you order, how many bands you order, the format used and whether some more complex processing options are enabled.

See documentation on how processing units are defined and calculated here. The following are detailed calculation examples for Sentinel-1, exploring three different scenarios, including a simple 1 image request, many small requests, long time ranges and large area processing.

Example 1

1 request that outputs an image 2000 x 2000 px large, with 2 bands ordered, 16 bit tiff format and orthorectification applied, would result in 20,34 PUs used. See the multiplication factors and the calculation below:

The calculation: (4 000 000 / 262 144) x (2 / 3) x 1 x 1 x 2 = 20,34 PU per request;

If you wanted to get 500 images (e.g. for time series), you would need to create 500 separate processing requests. If parameters for each image are equal as above, we can simply multiply our calculation with 500 and see, that 500 similar requests would use about 10 172 PU in sum.

The calculation: 20,34 PU x 500 = ~10 172 PU per 500 requests

If you look at pricing, you will see that the Exploration package is enough for 10 172 PUs, as it gives you 30 000 PU per month.

If your output image would be smaller, e.g. 500 x 500 px large, the cost of a single request would be 1.27 PU. Multiplied by 500, that’s just 635,78 PU in total, so much less. On the other hand, if you wanted an image with quite a large output of 5000 x 5000 px, a request would cost you 127.15 PU, and 500 of these would cost ~63 578 PU, which you might think would be covered by the Basic account, as it offers up to 70 000 PUs per month. However, processing API is limited to output images with width and height to up to 2500 px, so setting it to 5000 would result in an error. To output images this large, you would need an Enterprise account to use batch processing API. You will notice that batch processing divides your PU cost by 3, as it’s only a third of the price of processing API. So your 500 requests would cost 63 578 / 3 = 21 192 PU.

Example 2

If you wanted to create a single request that uses 500 scenes in a multi-temporal script, where you e.g. calculate max NDVI over several years, using up 500 acquisitions, your request would also use 10 172 PU, just like our request for 2000 px x 2000 px output in Example 1 (given that all the other parameters are the same as in Example 1).

The calculation: (4 000 000 / 262 144) x (2 / 3) x 1 x 500 x 2 = 10 172 PU

The issue here is, that you would use more than 2 000 PU in a single minute, as it’s just one request, and use this heavy isn’t supported by any package. Each package specifies how many PUs and requests you can use per month, as well as per minute. Your request would time out or fail. If you’re interested in very large time ranges, it’s best to use batch processing API.

Example 3

Let’s suppose you want to order an orthorectified 16-bit Sentinel-1 image with 2 bands for the whole Australia in full resolution, which let’s say is covered with a 4000 x 4000 km bbox, which equals to 4 million meters x 4 million meters. As Sentinel-1 resolution is 10 meters, we know that our output image will have to be 400 000 pixels x 400 000 pixels large.

The calculation: (160 000 000 000 ‬/ 262 144) x (2 / 3) x 1 x 1 x 2 = 813 802 PU.

Processing API is limited to 2500 pixels for output width and height, so you would have to run this request with batch processing API. To do so, you would need to have an Enterprise account. As batch processing is cheaper, your batch processing request would use 813 802 / 3 = 271 267 PU. You can see that Enterprise S would be sufficient for this, as it offers 400 000 PU per month. And because batch processing is asynchronous and takes a while to ingest the tiles, you won’t be stopped by per minute rate limiting for a single request.

In general, for smaller requests, even if you need many, Exploration and Basic should suffice. Note that Exploration account doesn’t support commercial use. If you need large requests, such as the whole countries or continents, or if you want longer time periods, you will need an Enterprise account to use batch processing. Custom options are also available, if you need heavier use than the packages allow.