Help Center

Carto Gebruikershandleiding

This manual guides you step by step through Carto’s features and provides all the information you need to make the most of the application. It explains how to set up Carto, use its features, and train high-quality satellite-AI models.

Download

Model Cards

These model cards provide a technical description of the AI models used in Carto: what the architecture is, which datasets were used for training, and more.

Download

Frequently asked questions

Here you'll find answers to frequently asked questions. Still need help? Feel free to reach out to us by email (contact@spheer.ai), phone (+316 23758739), or LinkedIn (Spheer).

How secure is Carto’s AI? Does Carto comply with the European AI Act?: Carto’s artificial intelligence consists of two parts: our foundation model, and the Carto system in which specific models are trained for each use case. To ensure compliance with the European AI Act of August 2025, Spheer follows the guidelines and compliance checker provided by the Future of Life Institute:
Foundation model: our foundation model qualifies as a General Purpose AI Model without Systemic Risk. To comply with the AI Act, we published the Model Card on our website, explaining how the model was developed and how it should be used.
Carto: Carto falls under the category of ‘Limited Risk AI Systems’. In line with the AI Act, we clearly state within Carto that users are interacting with an AI model, and we include in our terms of use that users remain responsible for any decisions made based on insights from Carto.
Do I need technical knowledge to use Carto?: No, you don’t!
Carto is designed for everyone, regardless of technical background. You don’t need to know anything about AI or satellite data—just use your browser to train a model and get valuable insights within minutes. Your expertise in the subject matter on the ground is all you need.
And if you ever need help using Carto, we’re happy to assist!
How accurate is Carto?: Carto is highly accurate, thanks to our foundation model trained on years of Sentinel-2 satellite data.
Field validations with our users show that Carto is often as accurate as manual mapping by experts. The platform delivers detailed, trustworthy maps—giving you confidence in your decisions.
Read more about how users apply Carto.
How quickly will I see results?: Within minutes!
Carto delivers fast, accurate maps for both small and large areas. Upload a few examples or draw them directly in Carto, click “train model,” and receive insights for your area of interest, almost instantly, going back to 2017. Refine as needed until your map is just right.
How does Carto handle personal data? Is Carto GDPR compliant?: Regarding personal data, Carto only stores users’ usernames and email addresses, which are not visible to anyone outside Carto. Spheer records each user’s activity within Carto. Upon request, this information can be provided and/or deleted from our systems.
How large an area can I monitor with Carto?: Carto lets you consistently monitor large areas, up to entire provinces or beyond! Thanks to our smart and scalable architecture, users can train models and generate maps for province-sized areas in just minutes.
Carto also works just as well for smaller areas, whether it’s a single municipality or nature reserve, whatever you need.
Where and how do I get access to Carto?: Carto is only accessible via a user account through client organizations. When you first start using Carto, you'll receive a login link via email. Be sure to check your spam folder. The first time you log in, you’ll set a password.
After setting your password, you can log in directly at carto.spheer.ai using your email address and password.

Carto tips

Quality over quantity

A few accurate observations are more useful than a large number of less precise ones. The more specific and accurate your examples, the easier it is for Carto to detect meaningful patterns.

Be specific in your examples

Split observations when necessary. Instead of one large 50% block, divide it into separate 100% and 0% areas if that better represents reality. This helps Carto distinguish subtle differences and leads to more refined models.

Vary your observations

Be sure to include not only examples of what you're interested in, but also examples of ‘other’ classes. This helps Carto make better distinctions and avoids misclassifying unlabelled areas. This applies to both regression (where you indicate levels of presence) and classification (where you assign different categories including a potential ‘other’ class).

Include observations from different years

This helps Carto to better understand growth patterns and temporal variation. Growth doesn’t look the same every year. Experiments show that including data from at least two different years already improves predictions significantly. In general: the more years you provide data for, the more stable the monitoring becomes.

Improve Carto iteratively

Start with a small number of observations and train the model. Gradually add more examples and see if predictions improve. Start with areas you are completely sure about, and only add edge cases in later iterations. This step-by-step process allows for quick feedback and efficient model tuning.

Make sure your observations are the right size

Observations should be at least 10m x 10m and no more than 1km x 1km. Smaller areas don’t contain enough satellite pixels for training, while larger areas slow down processing unnecessarily.