Phase 1: Local Frog Discovery Tool - Closed

1193
$2,000
0 Days Left
challenge-header-img
Novice level. The task was to build a computational model that can identify the occurrence of frogs for a single location using a single data source.
  • Overview
  • Eligibility Requirements
  • FAQs


Overview

  • This challenge is to predict the occurrence of a single species of frog for a single location using a single data source at a coarse spatial resolution.
  • The output will be a species distribution model of one species of frog. Species distribution models are one of the most widely used ecological tools, a cornerstone in many countries worldwide of environmental regulation and conservation.
  • Why frogs? Frogs are an indicator species. This means they are a go-to for scientists wanting to find out more about the environmental health of a particular ecosystem.
  • Because they have permeable skin, frogs are very sensitive to pollutants, and because they can live on both land and in the water, they are a good indicator of the health of these two different environments.
  • Frogs are poorly served by existing species distribution models. They have very localized distributions, more restricted than suggested by a potentially suitable habitat, and therefore existing models struggle to represent their range accurately.
  • As indicators of ecological health and proxies for biological diversity, the disappearance of frogs is of great concern. Where frogs occur, we see healthy, thriving, resilient ecosystems. Where frogs have disappeared, we see ecosystems in poor health. All the 2030 Sustainable Development Goals (SDGs) are underpinned by healthy ecosystems. This means we won’t reach our goals if we don’t prevent and reverse the loss of healthy ecosystems.
  • By joining this challenge, you are part of an important community who have decided to engage in activism using space tech and AI to monitor biodiversity at scale. Monitoring is key to prioritizing actions intended to protect and restore biodiversity.
  • The estimated time effort this level requires is approximately 16 hours.
  • Watch this video for some tips on how to get started: https://challenge.ey.com/challenge-level-1-2/help
  • Learn how to make a submission. Follow our video guide with instructions on how to make a submission and claim your spot on the Better Working World Data Challenge ranking board.
  • The team, or individual, placing 1st will receive: US$2,000.  
  • The team, or individual, placing 2nd will receive: US$1,000.  
  • Additionally, all individuals or team members who complete a submission that meets the specified criteria, as outlined in the terms and conditions, will receive a personalized certificate of completion.
  • Global Semi-Finalists may be eligible for a mentoring experience with an EY Firm and will receive a personalized certificate stating that he/she was a Global Semi-Finalist.


Skills

  • Participants in this challenge will benefit from a basic understanding of math and statistics, as well as some experience in coding, but there are no pre-requisites for taking part.
  • Participating in this challenge will improve your skills in Python for data science, machine learning and managing large volumes of data.
  • An example Jupyter notebook has been built, which has a preliminary F1 score of 0.61. The notebook connects to public facing data sources on Microsoft’s Planetary Computer. The output of the notebook is a .csv file that can be uploaded to the challenge platform to give a relatively low score on the ranking board that can be improved over the course of the challenge.
  • Opportunities for improvement include but are not limited to: modifying the way weather data is sampled over both space and time (i.e. spatiotemporal sampling), modifying the biomes that are sampled (e.g. exclude non-forest biomes), modifying the spatiotemporal sampling of frog occurrence data (e.g. limit to 2021-2022), standardizing the variables, selecting a different set of weather variables (there are 18 to choose from), engineering new features, hyperparameter tuning.
  • This challenge is considered low complexity because participants are expected to build models that are optimized to work across a single country (Australia) using a single source of data (weather). The course spatial resolution (4 kms) of the test set will also make computation requirements lower and reduce complexities associated with model building at very fine spatial scales.
  • This challenge has been co-created with remote sensing expertise from NASA (Dr. Brian Killough) and Swinburne University (Dr. Jack White), and biodiversity expertise from the Australian Museum (Dr. Jodi Rowley) and the CSIRO (Dr. Cam Slatyer).
  • Head to the 2022 GitHub repository for instructions on how to setup your own data and analytics environment in Microsoft Azure using a ‘magic link’.


Challenge Objective

  • Build a Species Distribution Model for the frog species - "Litoria Fallax" across Australia using TerraClimate variables.


Dataset Used

  • For Target variable - Frog occurrence datasets (Australia)
  • For Predictor variables - TerraClimate dataset (Climatic variables) available on Microsoft Planetary Computer portal (You will find additional information for the data to be used in the Data Description tab).

 

Evaluation

During the in progress phase:

  • A score out of 1.0 will be generated based on the F1 performance metric.
  • Participants will be evaluated on the extent to which they improve the accuracy of an existing model.


During the final assessment phase: please check the shortlisting process section in the terms and conditions of the challenge

  • The organizer will publicly announce the global semi-finalists and will ask those teams to submit a “content package” to support their submissions that will be used to judge whether they are chosen as finalists.
  • Additional details on how these deliverables are evaluated are detailed in the terms and conditions of the challenge.