Hello,
I would ask a question regarding L4A processor. I don’t really understand the process of selecting and extracting training data for the processor.
If I’m not mistaken, the parcels from subsidy aplication are used as training data, meaning those parcels reported by farmers. What if we can’t trust the farmers’ applications and just want to classify the crops to compare with the applications?
It would be useful to have a tutorial that details the process of selecting and extracting data training, as well as the classification process itself.
Best regards,
Dragutin
Hello Dragutin,
Indeed, what we apply is a supervised classification, meaning that a part of the parcels are used to train the classification model, and then this classification model is applied to all the parcels (that are included in the classification). On top of that, the parcels that are classified but not used for the calibration are used for the classification validation. This comes from the assumption that in the CAP context, the big majority of the farmers’ declarations are correct. And the type of classification which is used, a random forest classification, is relatively low-sensitive to outliers; it means that a few not correct declarations will not impact much the classification model (and so the classification accuracy).
Concerning the process of selecting the parcels for the classification, and among them for the calibration, this process is explained in the “ATBD for L4A crop type mapping 1.2” doc, accessible here: http://esa-sen4cap.org/content/technical-documents. Depending on the number of parcels in each crop type, you will see that 3 different approaches are used. The aim of it is to improve the accuracy results in the crop types with relatively less parcels, which would generate otherwise less accurate results. Of course, the parameters defining these 3 approaches can be easily adapted.
Do not hesitate if you have further detailed questions. About a possible tutorial, what kind of tutorial have you in mind (support, etc.)?
Best regards,
Philippe
1 Like
Philippe,
Thanks a lot for the quick and accurate answer.
I mostly understand a lot about supervised classification. I want to test the performance and capabilities of the sen4cap system, more specifically its module for classifying cultures. I have already explored the possibilities of Google Earth Engine in the field of crop type classification, then some other applications (e-cognition, SNAP, OTB, etc.). In principle, whether for supervised or unsupervised classification, all applications use the same or similar algorithms. The only difference is in the process of selecting training data, their stratification. The process of selecting training data within sen4cap seems a little too procedural (many attributes, LUTs, etc.), so I suggested that a video tutorial be created with original LPIS data or subsidy aplication data, to show the data structure.
Also, in my particular case, the data from subsidy aplication is incorrect, so classification and its in situ validation need to be done. Farmers intentionally enter the wrong data into the application to obtain subsidy, and on the other hand, the Paying Agency does not have a system of effective controls in place to break the practice.
For this reason, I’m trying to find a way to use sen4cap, specifically a classification module with training data from subsidy aplication that isn’t very reliable.
Best regards,
Dragutin
Hello Dragutin,
For the crop type classification and the complexity of the L4A crop type LUT, in fact only the first columns of this LUT are used for the crop type classification in itself (until CTL4A); the rest of the columns are used for the crop diversification use case, which you can decide not to process. So you could let these last columns empty. So it is easier than it looks at first glance :). For the tutorial, indeed we are planning to give an online training in April. As soon we settle the date, we will share the information about this training on the website. In the meanwhile, you can refer to the presentation slides (http://esa-sen4cap.org/content/presentations) of the hands-on training that was given in Louvain-la-Neuve in January.
Best regards,
Philippe
thanks a lot Philippe! Despite the current situation in the world, that people struggle to work, even in extremely difficult conditions, where the focus should be on the things that we essentially deal with.
Looking forward to the fact that in the near future there will be video material or something similar, that will provide the information necessary to implement such a system.
I look forward to online training to sign up and follow up.
Thanks again a lot, and I wish everyone a good health and a happy life.
Dragutin