Classifying Crop Type using Satellite Imagery

Notebook Workflow

1. 01-Pre-processing - read data in from PostGIS and join it with other relational tables in the dataset. This was also a first step to look at the data.
2. 02-create-features - write functions to calculate each vegetation index based on spectral bands. For each veg index calculate the mean and standard deviation across time and save as a multiband raster. I used 6 different vegetation inices.
3. 03-modeling-binary - model and hyperparameter selection using grid search and cross validation using 4 different kinds of models: logistic regression, random forests, support vector classification, and k-nearest neighbors.
4. I included a multi-class notebook to show that I tried!
5. Visualization is not included because it was done in a variety of programs, mainly QGIS and Tableau.

Overview

One of the United Nations sustainable development goals is “zero hunger by 2030.” Hunger and malnutrition remain a significant barrier to development in Africa. Collecting accurate data is critical for monitoring agricultural growth and improving food security. I will be using data from a Zindi challenge, which is an African data science platform that focuses on using data science for social benefit.

This project is a stepping stone to more advanced predictions such as crop yield prediction and predicted food security under future climate change scenarios which I am interested in doing for my final project.

Goals

The objective of this project is to create a classification model to identify vineyards using Sentinel-2 satellite imagery. The end result will be a map of classified fields (Vinyard or not) for the agricultural region along the Orange river.

Data

The fields in the training set are along the Orange River, a major agricultural region in South Africa that have been verified in person and with drones in 2017.

There are 7 crop types present in the fields:

1. Cotton
2. Dates
3. Grass
4. Lucern
5. Maize
6. Pecan
7. Vacant
8. Vineyard
9. Vineyard & Pecan ("Intercrop")

Files included in the dataset:

• 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.
• 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.