From Pixels to Places: Harnessing Geospatial Data with Machine Learning.

Author(s): Stephen Chege-Tierra Insights

Originally published on Towards AI.

Machine learning algorithms are the “cool kids” of the tech industry; everyone is talking about them as if they were the newest, greatest meme. Amidst the hoopla, do people actually understand what machine learning is, or are they just using the word as a text thread equivalent of emoticons? It would be like giving away expensive ingredients for a fine meal and then forgetting the recipe.

Shall we unravel the true meaning of machine learning algorithms and their practicability?

Machine learning (ML) has proven that it is here with us for the long haul, everyone who had their doubts by calling it a phase should by now realize how wrong they are, ML has being used in various sector’s of society such as medicine, geospatial data, finance, statistics and robotics.

According to IBM, machine learning is a subfield of computer science and artificial intelligence (AI) that focuses on using data and algorithms to simulate human learning processes while progressively increasing their accuracy.

Machine learning, originating in the 1950s, has progressed through various stages of development, including the resurgence of neural networks in the 1980s, the emergence of statistical learning methods in the 1990s, and the revolutionary impact of deep learning in the 2010s, catalyzed by advancements in computing power, big data, and algorithmic innovation, ultimately transforming industries, redefining research paradigms, and raising profound ethical and societal considerations as Forbes states.

A sector that is currently being influenced by machine learning is the geospatial sector, through well-crafted algorithms that improve data analysis through mapping techniques such as image classification, object detection, spatial clustering, and predictive modeling, revolutionizing how we understand and interact with geographic information.

Lets look at some of this algorithm and their code snippet with the main platform being google earth engine focusing on supervised learning.

Random Forest

IBM states Leo Breiman and Adele Cutler are the trademark holders of the widely used machine learning technique known as “random forest,” which aggregates the output of several decision trees to produce a single conclusion. Its versatility and ease of use, combined with its ability to handle both regression and classification problems, have driven its popularity.

In non-technical terms, Random forest is like asking a group of friends for advice. Every friend (tree) decides what to do depending on several considerations. The most well-liked option among them is then used to determine the ultimate conclusion. It creates precise classifications or predictions by combining numerous of these trees.

Random Forest is frequently used in geospatial analysis for tasks including classifying land cover, mapping vegetation, planning urban areas, and keeping an eye on the environment. By using the ensemble of decision trees to effectively capture spatial patterns, classify land cover types, predict changes over time, and identify significant features relevant to geographic phenomena, it excels in handling complex and high-dimensional geospatial data, such as satellite imagery or LiDAR data.

This allows for informed decision-making and resource management in a variety of domains, including agriculture, forestry, conservation, and urban development.

Radom Forest Code Snippet

// Load an image from the Landsat 8 collection
var image = ee.Image('LANDSAT/LC09/C01/T1_TOA/LC08_044034_20140318');

// Select bands for classification
var bands = ['B2', 'B3', 'B4', 'B5', 'B6', 'B7'];// Create a region of interest (ROI)
var roi = ee.Geometry.Rectangle(-122.45, 37.74, -122.4, 37.79);// Sample the training data using the ROI
var training = image.sample({
 region: roi,
 scale: 30,
 numPixels: 5000
});// Set the class property based on a land cover map
var classProperty = 'landcover';// Train a Random Forest classifier
var classifier = ee.Classifier.randomForest(10)
 .train({
 features: training,
 classProperty: classProperty,
 inputProperties: bands
 });// Classify the image
var classified = image.classify(classifier);// Display the classified image
Map.centerObject(roi, 12);
Map.addLayer(classified,
 {min: 0, max: 2, palette: ['red', 'green', 'blue']},
 'classification');

K-Nearest Neighbors

k-Nearest Neighbors (k-NN) is like asking your neighbors for advice: you look at what your closest neighbors are doing to decide what to do next. To categorize a place on a map, for instance, by figuring out if it’s a city or a forest, you look at the spots that are closest to you and identify what they are. If the majority of them are woodlands, you could assume that the new site is likewise a forest.

Geospatial applications of k-NN in include:

Land Cover Classification: Based on the properties of neighboring pixels, satellite imagery’s pixels are assigned to one of several land cover categories (forest, water, or urban).

Spatial Interpolation: Using the values of nearby sites as a guide, spatial interpolation estimates values for places where data is either lacking or unavailable.

Geospatial Clustering: This is the process of classifying geographical features into clusters or groupings according to how close they are to one another.
Environmental monitoring: Forecasting local environmental factors, such as temperature, precipitation, or air quality, using data from adjacent sensor networks or monitoring stations

K-nearest Neighbor Code Snippet

// Load an image from the Landsat 8 collection
var image = ee.Image('LANDSAT/LC08/C01/T1_TOA/LC08_044034_20140318');

// Select bands for classification
var bands = ['B2', 'B3', 'B4', 'B5', 'B6', 'B7'];// Create a region of interest (ROI)
var roi = ee.Geometry.Rectangle(-122.45, 37.74, -122.4, 37.79);// Sample the training data using the ROI
var training = image.sample({
 region: roi,
 scale: 30,
 numPixels: 5000
});// Set the class property based on a land cover map
var classProperty = 'landcover';// Train a k-Nearest Neighbors classifier
var classifier = ee.Classifier.kNearestNeighbors(10)
 .train({
 features: training,
 classProperty: classProperty,
 inputProperties: bands
 });// Classify the image
var classified = image.classify(classifier);// Display the classified image
Map.centerObject(roi, 12);
Map.addLayer(classified,
 {min: 0, max: 2, palette: ['red', 'green', 'blue']},
 'classification');

Naïve Bayes

And last but not least, we have The Naïve Bayes classifier-this is a supervised machine learning algorithm, which is used for classification tasks, like text classification. It is also part of a family of generative learning algorithms, meaning that it seeks to model the distribution of inputs of a given class or category.

When it comes to geospatial analysis, the category of a location (like forest, water, or urban) based on the probability of certain features being associated with that category, assuming that these features are independent of each other, making it efficient for tasks like land cover mapping and environmental monitoring.

Naïve Bayes Code Snippet.

// Load an image from the Landsat 8 collection
var image = ee.Image('LANDSAT/LC08/C01/T1_TOA/LC08_044034_20140318');

// Select bands for classification
var bands = ['B2', 'B3', 'B4', 'B5', 'B6', 'B7'];// Create a region of interest (ROI)
var roi = ee.Geometry.Rectangle(-122.45, 37.74, -122.4, 37.79);// Sample the training data using the ROI
var training = image.sample({
 region: roi,
 scale: 30,
 numPixels: 5000
});// Set the class property based on a land cover map
var classProperty = 'landcover';// Train a k-Nearest Neighbors classifier
var classifier = ee.Classifier.kNearestNeighbors(10)
 .train({
 features: training,
 classProperty: classProperty,
 inputProperties: bands
 });// Classify the image
var classified = image.classify(classifier);// Display the classified image
Map.centerObject(roi, 12);
Map.addLayer(classified,
 {min: 0, max: 2, palette: ['red', 'green', 'blue']},
 'classification');

These are the 3 most important machine learning techniques and their code snippet; I would recommend running them on the Google Earth engine for maximum efficiency. I will soon cover on how to test the accuracy of your algorithm in a later blog.

As for now, mastering these fundamental machine learning techniques and implementing them on Google Earth Engine lays a solid foundation for harnessing the power of geospatial data analysis, paving the way for more advanced applications and insights in remote sensing, environmental science, and spatial modeling

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