Envision| Quiet Planet

Quiet Planet

The COVID-19 outbreak and the resulting social distancing recommendations and related restrictions have led to numerous short-term changes in economic and social activity around the world, all of which may have impacts on our environment. Your challenge is to use space-based data to document the local to global environmental changes caused by COVID-19 and the associated societal responses.

Analysis of NASA Image Datasets to Predict, Correlate and Propose Solutions to the COVID-19 Pandemic

Summary

In our project, we predicted relations between various climate change factors like greenhouse emissions and human activity like transportation as an effect of COVID-19. A knowledge graph network was created in the form of connections on the basis of a schema of several rules. Using these predictions, we mapped out various regions to target in the crisis, via K Means Clustering; we also made predictions for the values of the various factors accounted for, via multivariate regression.

How We Addressed This Challenge

The aim of this challenge was to document and analyse environmental changes and societal responses using space-based data as a result of COVID-19. 

  • We used space-based data - NASA Datasets on Vegetation, Fires, CO2, NO2, Cloud Fraction, Water Vapour, Aerosol Particles, Snow Cover, Ground Temperatures, etc. [Your solution may focus on any type or combination of environmental phenomena...]
  • We did a time-series analysis of these image datasets. We looked at peaks and dips over the past few months and used regression to predicts dataset images  for the next months. 
  • We used a KGCN model to predict correlations between COVID-19 and the environmental changes. [How would you determine if observed environmental phenomena are impacts of the COVID- 19 pandemic?]
  • We also correlated changes in human activity such as transportation with COVID-19. [Now, the global standstill has unveiled a new realm or change in ecosystem dynamics.]
  • We predicted and correlated changes in atmospheric gases in accordance to COVID-19. Looking at extremes of certain regions, we also manually looked at the government policies leading to that extreme. [Therefore, the changes in air pollution associated with the pandemic will serve as a natural experiment...] [Therefore, you may be able to assess the impact of the pandemic on atmospheric gas emissions.]
  • We used K Means Clustering to give a visual mapping of regions worst affected by the virus and in need of economic help. [Aid organisations could also use the data to identify those areas hit hardest by the virus.]
How We Developed This Project

Seeing the situation and impact of COVID-19 in India, our team was inclined on choosing this challenge. On one hand, COVID-19 has hit the economy hard, however, it has also had a positive impact on the climate. This challenge gave us an opportunity to analyse these interesting changes and effects, and provide a solution to combat this crisis. 

Our team has experience with computer vision and image analysis. We used this strength to approach the challenge, by looking for visual datasets that we can then analyse to make predictions and find relations. 

We used the image datasets from the space agency data of various elements affecting the climate. These formed the basis for our chronological analysis model for making future predictions about COVID-19 and its global impact on the environment. These were also used as nodal inputs in our graph model, and helped us predict relations among various factors. We also used other NASA Datasets of CO2, NO2, Water Vapour, Ozone, etc.

Because of the spread of the virus, we used software like Google Collab  to work together on the model. The entire model was written in Python. We used Grakn's graql library for our knowledge graph network. We used graql to develop this project along with Grakn to form correlations. The KGCN was made using tensorflow and grakn provided the query and reasoning engine for the formation of connections. Various libraries were used throughout the project: OpenCV, NumPy, matplotlib, os, PIL,  skimage, etc. 

A multivariate regression model was used to make predictions on a time series analysis of these datasets. K Means Clustering along with Canny-Edge Detection was used to target regions worst effected by the virus, and these were plotted on a bonne world map.

Throughout the project, we faced many problems.

  • The datasets had certain inconsistencies wherein the data was incomplete. 
  • We had to account for certain out-layers while writing the regression model.
  • The graph model had to be integrated with the complex schema based off of the needs of the challenge. This caused numerous errors and failed attempts over the course of the 2 days.
  • Working on the same model, online, was also a tough task for the team. There was a lot of confusion and misunderstandings. We had to be extremely careful while combining and handling the various algorithms for the model.

We also made some achievements.

  • We quickly found the datasets that we would require for our model. These were compiled chronologically, without any mistake. 
  • We were clear about our idea from the very beginning and knew what we were doing throughout the project.
  • Our graph model did well in predicting correlations. We successfully managed to correlate COVID-19 with changes in atmospheric gases as well as human activity like transportation. 
  • Our time series analysis algorithm made predictions for June 2020. In order to check the accuracy, we ran the model for April 2020, and matched the output with our data. We got an 87% accuracy. 
  • We successfully managed to plot the regions that should be targeted during the pandemic for economic help, globally, by various organisations, using the factors that we predicted and correlated. 


Project Demo

LINK TO 5 SLIDE CANVA PRESENTATION

These are the visual outputs of our model to showcase and support our project. 

https://www.canva.com/design/DAD92Eh-LQs/WE1BfKpPKNk-tr41lj1suA/view?utm_content=DAD92Eh-LQs&utm_campaign=designshare&utm_medium=link&utm_source=sharebutton

Project Code
Data & Resources
  • https://earthobservatory.nasa.gov
  • https://www.nasa.gov/nex
  • https://earthdata.nasa.gov/learn/pathfinders/covid-19
  • https://archive.eumetsat.int/umarf/


Tags
#graph-network, #predictions, #correlations, #solutions-mapping, #image-analysis
Global Judging
This project was submitted for consideration during the Space Apps Global Judging process.