I am using the notebook worldcereal_v1_demo_custom_croptype.ipynb, and I would like to better understand how it works, especially regarding the final output.
I have a relatively good understanding of the different steps, but I would like to know if the final generated map recognizes crop types in the selected area in real-time (or at least over a recent period where satellite images are of good quality). If so, how can I determine the exact date of the imagery used?
Additionally, if I want to apply the custom model I generated to satellite data that I have downloaded for a specific period, what format should my file be in, and what variables should it include?
Step 8 in the notebook allows you to generate a map based on your trained and deployed model. In that step, you’ll be prompted to select a processing period in this line: processing_period = slider.get_processing_period(). You can generate a croptype map for a chosen period by moving the slider. Note that currently, this is only possible for processing dates till end of 2023. Next week, we release an update to the system which will allow you to generate maps with data till end of 2024. The processing period is always a year, which you should center around your season of interest.
Currently it is not possible to run the models on locally downloaded satellite data. We’d be curious to hear what you have in mind as a use case for that.
Finally, note that in next week’s update, it will also be possible to include your own (private) reference data, provided it is uploaded to our Reference Data Module (you can still keep your datasets private).
I’m currently working on a project as a student, aiming to develop an application that displays the crop type history over a specific area in France. I already have this historical data using the French Land Parcel Identification System (RPG - Registre Parcellaire Graphique) from IGN, and I’m displaying crop types per parcel from 2015 to 2023. I’m also building a dashboard to provide statistical insights.
It’s still under development, and I’m planning to add vegetation indicators such as NDVI, and climate-related impacts like water stress and soil moisture.
In a next step, I would like to use your classification model to identify crop types and cultivated parcels for the most recent years not yet covered by the RPG. The goal is to have the most up-to-date view possible of cultivated surfaces and crop types — or at least to be able to work from a reliable base to make predictions with acceptable accuracy. Ultimately, I’d like to get close to a near real-time overview, as much as seasonality allows. Your work and the progress you’re making are of great interest to me.
I don’t have private data to work with at this stage, but I was wondering to what extent it would be possible — by training and customizing the model as currently supported — to prepare a dataset using Sentinel-1 and Sentinel-2 imagery (along with other relevant features from the training dataset) to classify current crop types.
For now, I’m working locally, but I plan to get more hands-on with processing hubs like OpenEO and CDSE.
Unfortunately, I wasn’t able to attend the latest webinar, and I was wondering if it would be possible to access a recording of the session.
Thanks again, Jeroen, for sharing the updated version — I’ll definitely spend some time exploring it!
the recording will be made available soon and indeed I think your described use case is pretty much covered by the new demo that Jeroen has shared. You can access public data that was already extracted by us (giving you access to the satellite data) and train your models locally. But they then need to be deployed to the cloud for openEO to use them to make your crop type map. It’s all explained in the webinar and also extensively in the notebook(s) itself.