Process Overview

AutoMap is designed to be simple to handle while very complex to run. Under the hood, we take care of efficient workflows that are scientifically correct and smart. In general, we can boil the process down to these steps you don't need to worry about:

What AutoMap is doing for you
  1. Cloud Setup
  2. The virtual machine processing the data needs a storage medium capable of handling a very high throughput (Mbits). Data needs to be secured so only you can access your files. All ressources need to be structured and tagged, accordingly and the monitoring system needs to be initialised.

  3. Raw Data Download
    (last step in Download - Mode)
  4. We use freely available satellite data that just needs to be downloaded, but depending on the run parameters, the number of files could go in the thousands and the storage needs in the terabytes.

  5. Cloud Masking
  6. Every single satellite image has a default cloud mask attached to it. That mask is created by NASA or ESA in ways to satisfy the most basic needs of all their users. Such a mask is not either-or, it contains triggered bits for each pixel and depending on the bit position, the information changes. That way it is possible to programmatically differentiate cirrus clouds from cloud shadows or snow.

  7. Pre-Processing
  8. A picture taken with a smartphone actually consists of three separate images for the colors red, green and blue (RGB image). The combination of those three image bands provides the end product.

    At this point in the process a single satellite scene consists of separate image files per (color) band plus the cloud mask and metadata as text. Since in this scenario the pictures were taken in space, the metadata files help us to get rid of all different kinds of distortions. Once this step is finished, every scene is packed in one single image.

  9. Merging
    (last step in Composite - Mode)
  10. Imagine a four-dimensional cube with a temporal dimension (a scene at different points in time), a spectral dimension (the color bands) and two spatial dimensions (X and Y coordinates on the globe). In this step we get rid of the temporal space, e.g. we want to merge the stack in that dimension in a smart way, without clouds and any other outliers.

  11. Classification
    (last step in Classification - Mode)
  12. Based on the (temporally merged) results, training points can be collected and a classifier can be trained. Every classifier itself has parameters that can be tuned (hyperparameters), so the classification results become as accurate as possible with a fixed set of training. A classifier is also just as good as the training data and features that describe the situation. Since a single pixel is never isolated but always connected to its neighbors (we can assume that chances are high that a forest pixel is surrounded by other forest pixels), we can take that into account to further increase accuracy.

  13. Results
  14. For every selected satellite the merge and classification results are provided, as well as a metadata file containing detailed information about the considered individual scenes and other parameters. Also, a clever classification merge product is provided, holding the best results of all individual classifications.

    Please refer to the FAQs of the documentation for detailed information on the output!

  15. Notification and Storage
  16. Once the operation finished, the user will be notified per mail, inviting them to download the compressed results from Amazon AWS S3. The monitoring system will be updated for further references.


AutoMap is designed to work with the most common freely available missions from NASA and ESA. They all are multispectral satellites with exception of NASAs MODIS Land Cover Type Product (MCD12Q1), which is the default classification input. For downloading MODIS or MCD12Q1 data a NASA LAADS Key is required.

Satellite Name Product Resolution (Spatial) Bands (Spectral)
Landsat 5 Top-of-Atmosphere (TOA) 30x30 m 1, 2, 3, 4, 5, 7
Landsat 7 Top-of-Atmosphere (TOA) 30x30 m 1, 2, 3, 4, 5, 7
Landsat 8 Top-of-Atmosphere (TOA) 30x30 m 2, 3, 4, 5, 6, 7
MODIS Surface Reflectance (SR) 500x500 m 3, 4, 1, 2, 6, 7
Sentinel 2 Top-of-Atmosphere (TOA) 10x10 m 2, 3, 4, 8, 11, 12
MCD12Q1 Land Cover Type1 500x500 m Class, Probability

Launch a new run

The easiest way to generate a new call for action is the click-and-go user-interface that is made available in the Menu after login. Pick a mode for a set of satellites, specify a time interval, select one of your uploaded vector files as boundary for your area of interest and start a run. The description field is here to make it easier to identify your runs later on. After a successful launch, the monitoring system will keep you updated on the current state of the process. When finished, an email will be sent to the specified email adress(es), containing final informations and a download link that remains valid for 48 hours.

Useful Tip:

Please refer to the Manual of the documentation for detailed information on all parameters!

Code Launcher

The Code Launcher has a dedicated documentation page. The Code Launcher provides programmatic access to AutoMap with a simple Python interface. Write your own script to iterate over different time ranges, areas and satellites, tag the runs with individual descriptions and easily process in bulk. Note: The code launcher is reserved for premium users.


The AutoMap API has a dedicated documentation page. The API enables your team to automate the launch of AutoMap runs. Note: The API access is reserved for premium users.


Upload and delete your boundary and sample files here. All files relevant for a processing job need to be uploaded to the server first.


Keep track of your orders. We maintain a database of submitted jobs, their current status, description, costs and other important information.