Machine learning algorithms for identifying crop types in smallholder farms

Authors: Preeti Rao*, SEAS, University of Michigan, Meha Jain, SEAS, University of Michigan
Topics: Remote Sensing, Sustainability Science, Asia
Keywords: machine learning, remote sensing, classification, smallholder, agriculture
Session Type: Paper
Day: 4/5/2019
Start / End Time: 3:05 PM / 4:45 PM
Room: Roosevelt 3, Marriott, Exhibition Level
Presentation File: No File Uploaded

Smallholder farming systems in the Indo-Gangetic Plains (IGP) are a major part of the rice-wheat production belt of India. Identifying the crop types across the entire IGP provides a critical dataset to help understand cropping patterns, crop yield intensities, and farmer adaptations to climate change. Our study area is a 20 x 20 km area in Eastern IGP where we collected crop type information for four major crops (maize, mustard, tobacco and wheat) during the winter growing season of 2016-17. The mean farm size in our sampled dataset of 324 fields is 745 m2 with 64% of the fields smaller than the mean size. We compare the performance of three machine learning algorithms, Random Forests (RF), Support Vector Machines (SVM) and Artificial Neural Networks (ANN) to develop an ensemble classifier. We apply this ensemble to multi-sensor high-resolution optical (Sentinel-2 and Planet) and radar (Sentinel-1) satellite data to identify the four major crop types in our study area. We identify the critical number and timing of images essential for high classification accuracies. These learnings will be applied towards multi-temporal crop type classification in the entire IGP region.

Abstract Information

This abstract is already part of a session. View the session here.

To access contact information login