Drought Watch

Introduction

This is the Weights & Biases community benchmark for deep learning models for drought detection from satellite. With better models, index insurance companies can monitor drought conditions—and send resources to families in the area—more effectively. The goal is to learn from ~100K expert labels of forage quality in Northern Kenya (concretely, how many cows from 0 to 3+ can the given geolocation feed?) to more accurately predict drought conditions from unlabeled satellite images.

Getting started

Check out this repo.

From there, you can follow the brief instructions to clone the repo, download the data, and start training models in a few lines of code. You can upload your experiment runs to W&B to easily visualize the training, explore the hyperparameters and architecture, and help improve the model.

Reports

You can check out this short report to see how the baseline was trained, explore some of the submissions so far, and get inspiration for how to run or showcase your experiments. We would love to see and feature your reports based on Drought Watch—please include a link to your report with your submission to the benchmark.

Evaluation

This is an open collaborative benchmark—the goal is to build the most accurate model together. We encourage you to include your code and share notes, findings, or observations about your process. You can use this platform to see what others have tried, get ideas, and synthesize different approaches. We will be using the "val_acc" metric to sort all submissions. This is the validation accuracy on data in the 'val' folder provided, computed automatically in the baseline training script. We're also reserving a test set to evaluate long-term generalization.

Resources

A writeup of the project and dataset is now available on Arxiv: Satellite-based Prediction of Forage Conditions for Livestock in Northern Kenya. This work will be presented at ICLR 2020 as part of the Computer Vision for Agriculture workshop--we will link to the recording once it is available. You can also find more information on Drought Watch in this blog post and this latest update.

Dataset credits

The data used in this research was collected through a research collaboration between the International Livestock Research Institute, Cornell University, and UC San Diego. It was supported by the Atkinson Centre for a Sustainable Future’s Academic Venture Fund, Australian Aid through the AusAID Development Research Awards Scheme Agreement No. 66138, the National Science Foundation (0832782, 1059284, 1522054), and ARO grant W911-NF-14-1-0498.