How To Learn Earth Observation from Machine Learning as a GIS Pro-Tips and Tricks.

Author(s): Stephen Chege-Tierra Insights

Originally published on Towards AI.

It seems impossible until it is done- Nelson Mandela.

One thing I have learned in the journey as a GIS data science content creator is never to underestimate the ability of a human brain to grasp something new in a short record of time, especially with the emergence of online line platforms. With dedication, I believe you can learn anything.

I am also sure with much time and practice I could learn how to be a rocket space engineer scientist or a forensic DNA chemical analyst. Of course, I am not interested in these careers but if you and I put in the heart and effort, we could be experts in these fields.

What I am getting at is that with the current state and unpredictability of the technology sector, we need to learn and relearn constantly.

Now, in the spirit of learning, I want to take you on a journey on how a GIS professional can learn how to apply machine learning and earth observation. This will be an overview but I will introduce you to common concepts of earth observation. Let us begin.

Why You Should Learn Machine Learning for Earth Observation

First things first, let me try and advise you on why this is important, why is learning machine learning an earth observation as a GIS pro even practical? I hear you grumble; well, it is. Here is why

1. Progression in Career:

-In-Demand Skills: I’m unsure what will persuade you if this argument doesn’t. Integration of GIS, earth observation, data science, business analysis and machine learning is critical to many industries, including agriculture, urban planning, infrastructure development, digital governance disaster management, and environmental monitoring.

-Competitive Edge: Possessing these skills can make you stand out from the crowd when applying for jobs and open doors to more specialized and senior roles.

2. Stay Ahead of the learning curve

-Technological Evolution: Having a broad skill set in GIS, ML, and EO as technology develops will help you stay robust and adaptive in a job environment that is evolving quickly.

-Continuous Learning: Being involved in these sectors encourages a mindset that is necessary to stay creative and current in your work.

3 Useful Impact

-Disaster Management: By employing ML models to evaluate EO data for early warning systems, damage assessment, and resource allocation, you may enhance disaster response and recovery.

-Agricultural Optimization: Use machine learning algorithms on EO data for yield prediction, crop monitoring, and precision farming to increase agricultural output.

4. Multidisciplinary Cooperation:

-Greater Impact: Working at the intersection of GIS, ML, and EO allows for collaboration with experts from a variety of fields, such as computer science, environmental science, and urban studies.

-Comprehensive Solutions: Using a range of abilities, develop comprehensive solutions that address difficult real-world problems.

Where To Begin?

1. Learn

-Essential Ideas: Acquire an understanding of the principles underlying satellite imagery, remote sensing, and additional EO data kinds (including optical, radar, and LiDAR).

-Data sources: Learn about the primary sources of EO data, which include satellites like Sentinel, MODIS, and Landsat, as well as for-profit businesses like Planet Labs.

-Applications: Look at the several areas where EO is used, including agriculture, forestry, urban planning, and disaster assistance

2. Learn R Python and Google Earth Engine

-R and Python foundations: R: Learn the essentials of R first, such as data types, functions, control edifices, and data operation with dplyr and tidyr.

-Begin with the essentials of Python: including variables, functions, loops, and data structures. Learn how to use data manipulation libraries such as numpy and Pandas.

• GEE: Make use of Google’s official GEE tutorials and materials.
• code snippet

1. Load and Preprocess Landsat 8 Imagery

// Load Landsat 8 Image Collection
var landsat8 = ee.ImageCollection('LANDSAT/LC08/C01/T1_SR')
 .filterDate('2020-01-01', '2020-12-31')
 .filterBounds(ee.Geometry.Point([78.9629, 20.5937])) // Example coordinates
 .median();

// Select bands for classification
var bands = ['B2', 'B3', 'B4', 'B5', 'B6', 'B7'];
var image = landsat8.select(bands);

// Visualization parameters
var visParams = {bands: ['B4', 'B3', 'B2'], min: 0, max: 3000, gamma: 1.4};
Map.centerObject(image, 8);
Map.addLayer(image, visParams, 'Landsat 8 Image');

2. Create Training Data

// Manually create points and assign labels for different land cover types
var points = ee.FeatureCollection([
 ee.Feature(ee.Geometry.Point([78.9, 20.5]), {class: 0}), // Water
 ee.Feature(ee.Geometry.Point([78.95, 20.6]), {class: 1}), // Urban
 ee.Feature(ee.Geometry.Point([79.0, 20.7]), {class: 2}), // Vegetation
 // Add more points as needed
]);

// Sample the image at the locations of the points
var training = image.sampleRegions({
 collection: points,
 properties: ['class'],
 scale: 30
});

3. Train the Random Forest Classifier

// Define the classifier with Random Forest
var classifier = ee.Classifier.smileRandomForest(50).train({
 features: training,
 classProperty: 'class',
 inputProperties: bands
});

4. Classify the Image

// Classify the image
var classified = image.classify(classifier);

// Define visualization parameters for the classification result
var classVisParams = {
 min: 0,
 max: 2,
 palette: ['blue', 'gray', 'green'] // Colors for water, urban, vegetation
};

// Add the classification result to the map
Map.addLayer(classified, classVisParams, 'Classification');

5. Export the Classified Image (Optional)

// Export the classified image to Google Drive
Export.image.toDrive({
 image: classified,
 description: 'LandCoverClassification',
 scale: 30,
 region: image.geometry().bounds(),
 maxPixels: 1e9
});

2. Practical Application

-Projects and contests: Take part in EO and ML-focused projects and contests, such as those on Kaggle or Zindi.

-Online Courses: Take advantage of the practical activities and projects offered by online courses and tutorials. Specialized courses in EO and ML can be found on websites such as Coursera, Udacity, and DataCamp.

-Open-Source Projects: Participate in GitHub repositories and open-source projects related to EO and ML

3. Community and Support

-Forums and Groups: For support and information exchange, which are very crucial, join online forums such as Stack Overflow and GIS Stack Exchange, watch informative YouTube content on data science and earth observation and Reddit’s GIS community

-Meetups and Workshops: Participate in GIS, EO, and data science-related conferences, workshops, and meetups both locally and virtually.

4. Machine Learning Integration

–Learn Various Libraries: Learn how to use caret, randomForest, and xgboost in R and scikit-learn, TensorFlow, and keras in Python as well as other machine learning and EOUse packages.

–Model Training: Learn how to use EO data to train, validate, and test machine learning models.

-R and Python for Integration with Machine Learning: Libraries: Use R’s caret, randomForest, and xgboost, and Python’s scikit-learn, tensorflow, and keras.

-Training Models: Discover how to train, assess, and test machine learning models using EO data. Find out different ways of training and deploying models

5. Stay Informed

-Research Articles: Study articles and studies published in scholarly peer-reviewed journals such as GIS Lounge, Publisher and IEEE Transactions on Geoscience, earth observation, data science, and Remote Sensing. This will help you be ahead of the learning curve

-Community Engagement: Take part in EO and ML-related conferences, webinars, workshops, and forums.

Conclusion

Being a GIS specialist and knowing how to combine Earth observation data with machine learning techniques offers a world of possibilities for meaningful insights and increased analysis. Learning to use R, Python, and Google Earth Engine will enable you to address challenging real-world issues by utilizing sophisticated analytical techniques and vast amounts of geospatial data.

It is imperative to commence with a fundamental understanding of Earth observation and machine learning. On top of this base, specific R and Python libraries and tools, along with the powers of Google Earth Engine, offer the framework required to manage, process, and evaluate enormous volumes of spatial data.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI