Skip to content

Home

Jump to bottom
EmmaNguyen edited this page Sep 2, 2019 · 3 revisions

Who reads this?

This document is written for data and tech team of forrestsee to get scope, goal, overview, and summary of this project, especially in the pilot mode.

Scope

  • Learn the process of building a simple predictive model by Python;
  • Learn the structure of design a machine learning-based solution;
  • Experience with a framework of an advanced algorithm (open-source code in Python).

Goal

  • Build a simple predictive model to predict the green cover given data collected from * * Stateline photo and visualize the future based on a simple model.

Overview

Through this project, we hope to raise awareness of the youth community on sustainable development topics, especially environments, renewable energy, and climate change.

This guideline introduces a step-by-step process to help you get a high-level overview of a process, e.g. an input, a process, an expectation for the outcome and some possibilities. References here include useful content directing you to the source to get a better idea of something we used in the pilot project.

This pilot project is a short run for us to validate ideas and assumptions as well as call more resource in testing time. After this, we hope a better plan will be proposed and more activities will be followed-up to build a community of who care about sustainable development.

Summary of Forrest See - the pilot - the first iteration

The project is built on top of a simple 5-step process, e.g. data preparation, data processing, feature extraction, build a predictive model and communication. Before that, we take lessons from previous projects in climate change for Amazon Forrest.

In this early stage, we use visualization and storytelling to show a picture of climate change affect a particular area of Vietnam - Son Tra Pennuisila in Danang city.

  1. Prepare data
  • Data is photos collected from satellites with 30MB data
  1. Process data
  • Normalize data set by time
  1. Extract feature
  • Extract green pixels of each image by counting those followed by color codes
  • Calculate the number of green pixels.
  • Calculate the percentage of green pixels
  1. Build a simple predictive model
  • Predict green cover over time by linear regression with time series for the next K units of time. (Unit: years, months, days, etc.)
  1. Build a simple visualization
  • Using geoplotlib and matplotlib to plot a variance of data with respect to time.

Advanced techniques for step 4 after the

  • Using mindsdb with a tutorial.

Future improvement

  • Data: more time and more quality;
  • Process: combine data with other economical and social factor to show how climate change would affect the local;
  • Visualization: try with different kinds of a plot to detect a better storyline.
  • Predictive solution: find more meaningful features to increase the performance of a naive model;
  • Advanced algorithms for predictive solution + visualization: applying for neural style transfer with parametric extracted to depict a visualization of a potential (destructive) scenario.

Reference

Clone this wiki locally