The data have been split into a test and training set. The training set contains 2497 agricultural fields and the test set contains 1074 fields. The crops growing on each field was verified in person and with drones in 2017.

We have provided the satellite images of the entire region across 11 time slices in one year, covering summer and winter months. You will use this satellite imagery to train a model to estimate the probability that each field falls into each crop type. There are 2 tiles with the satelite data. Files with the name JFP contain 10% of the fields and the other files contain 90% of the fields.

Use only the data provided here to train your model. Do NOT use Field_ID as a feature in your model.

There are 7 crop types present in these fields, plus vacant fields, and fields that have both vineyards and pecans intercropped in one field (this is its own classification). The crop IDs are as follows:

1 Cotton

2 Dates

3 Grass

4 Lucern

5 Maize

6 Pecan

7 Vacant

8 Vineyard

9 Vineyard & Pecan ("Intercrop")

Your task is to provide the probability that each field belongs to each of the above listed classes. For each unique field ID you should provide 9 probabilities with value between 0 and 1.

Your submission file should look like:

Field_ID     Crop_ID_1    Crop_ID_2    Crop_ID_3    Crop_ID_4 .....   Crop_ID_9
<string>      <number>    <number>     <number>      <number>          <number>
   5            .034         .215        .567            .975            .123    

The files you have for download here are:

• Orange River Climate: Useful background information
• Orange River Crop Grown Stages: Useful background information
• Train.zip: shapefile containing all of the fields in the training dataset. This is the dataset that you will use to train your model.
• Test.zip: shapefile containing all of the fields in the test dataset. This is the dataset on which you will apply your model to. This dataset contains the same variables as the test data except there is no target. This is what you are predicting.
• Farmpin_sample_submission.csv: An example of what your submission file should look like. The order of the rows does not matter, but the names of the unique Field IDs must be correct. Your submission file should have all of the Field IDs in this file along with estimates of the corresponding Crop IDs.
• Crop_id_list.csv: List of all the unique crops and their Crop IDs
• Satellite images tile 1 are the 11 files named as YYYY-MM-DD.zip: These are Sentinel-2 satellite data captured on the date indicated in the file name. These files are in SENTINEL-SAFE format, including image data in JPEG2000 format, quality indicators (e.g. defective pixels mask), auxiliary data and metadata. This includes all bands and TCI. Each image contains 90% fields in both the training and test set.
• Satellite images tile 2 are the 11 files named as YYYY-MM-DD-JFP.zip: These are Sentinel-2 satellite data captured on the date indicated in the file name. These files are in SENTINEL-SAFE format, including image data in JPEG2000 format, quality indicators (e.g. defective pixels mask), auxiliary data and metadata. This includes all bands and TCI. Each image contains the remaining 10% of fields in both the training and test set.
• The two tiles make up the entire scene. 90% of the fields are in tile 1 and the remaining 10% of the fields are in tile 2.

Note: If you want to download the satellite data using a script, you can contact us for a permanent URL at zindi@zindi.africa. You will still have to agree to the terms of use for this competition, and you may not share the dataset or URL with anyone who has not also registered on Zindi and agreed to the terms of use for this competition.

If you need cloud computing support to process this data, please complete this form to get $500 USD in Azure credit for this competition. If you have questions about transfering the data to your workspace or setting up your workspace, let us know. We will also try to post some tips and guidance.

Have a look at these blog posts by Johnowhitaker.

• https://datasciencecastnet.home.blog/2019/06/26/tutorial-predicting-crop-types-with-gee/
• https://datasciencecastnet.home.blog/2019/06/27/tutorial-improving-crop-type-predictions/