In this competition, you are given a list of maize fields associated with images from Planet and Sentinel-2 satellites. Original positions are considered images' centers as (0,0) and you are given a displacement vector for each field in the training set and you are required to predict these vectors for each vector in the test set. To illustrate more the displacement vector, the following is an image with original position as blue point, corrected position as red point and the displacement vector between the two positions in yellow:

Data sources

You are provided with several data sources in this competition and you may use any (or all) of them in your solution including:

Planet Lab images:

These are images captured using Planet Lab Satellite in ~4.7 m resolution in different timestamps. Each image is 84X84 pixels and has RGB channels only. It should help you identify the farm boundary and what is laying inside beside Maize (like houses and trees).

Sentinel-2 images

These are images captured using Sentinel-2 Satellite in ~10 m resolution in different timestamps. Each image has 16 bands with 41X41 pixels for each of 12 months, giving a total of 192 image bands. The band names are provided in the ImageBands.docx file. These images are aligned on the same area and centered around the same center as Planet Lab images. It should help you in yield estimation for Maize fields.

Meta-data

These are variables captured during the yield estimation process in the field which includes:

• yield estimate (kgs/m^2)
• field area
• year in which the yield estimation process was done
• annotation quality (available only in training and auxiliary data)

All these data sources are available for training, test and auxiliary data. The annotation process applied on the auxiliary data is slightly different from what is applied on both training and test data but we believe it should be useful when added in the training phase.

External data is allowed in this competition, but before a dataset can be used it must be requested in a discussion post, at which point we will review it and add it to the list of allowed datasets. 6 weeks into the competition we will stop accepting requests for additional datasets.

If, when you click to download the starter notebook it takes you to another page, ctrl-S and it will save to your downloads folder.

Files available for download:

• train-unique.csv - Field_ID, Year, Quality (of location label), Yield and x,y of Displacement `Vector for the training set
• extra_train.csv - Field_ID, Year, Quality (of location label), Yield and x,y of Displacement `Vector for the extra training set
• test.csv - Field_ID, Year, and Yield for the test set
• sample_submission.csv - The Field_IDs for which you must submit predicted x&y of the Displacement Vector
• auxilary_data-unique.csv - Field_ID, Year, Quality (of location label), Yield and x,y of Displacement Vector for the auxiliary set
• sentinal.zip - sentinal images for all the data points saved as tif images (shape (192, 41, 41) or (192, 40, 41))
• sentinel_for_points_collected_in_2015.zip - corrected sentinel images for data points captured at 2015 saved as tif images (shape (192, 41, 41) or (192, 40, 41))
• planet-jun17.zip, planet-dec17.zip, planet-jun18.zip, planet-dec18.zip - planet Labs images for all data points saved as png (shape (84, 84, 3) or (83, 84, 3)) captured at June 2017, December 2017, June 2018 and December 2018 respectively (Imagery © 2017 and 2018 Planet Labs Inc.)
• extra_train-planet-jun17.zip, extra_train-planet-dec17.zip, extra_train-planet-jun18.zip, extra_train-planet-dec18.zip - planet Labs images for extra training data points saved as png (shape (84, 84, 3) or (83, 84, 3)) captured at June 2017, December 2017, June 2018 and December 2018 respectively (Imagery © 2017 and 2018 Planet Labs Inc.)
• ImageBands.docx - The image band names for Sentinel-2
• StarterNotebook.ipynb - shows how to load and visualize the data and create a baseline submission.