Azavea to Present on Various Open Source Topics at FOSS4G North America

March 26th, 2018

FOSS4G North America 2018 Conference

The FOSS4G NA 2018 Conference Committee recently released the full program schedule for the May conference. Azavea was fortunate to have multiple presentations accepted and will have several staff in attendance in addition to an exhibitor booth. We’re also a proud sponsor of FOSS4G NA 2018.

Read more about our talks below.


Cloud Optimized GeoTiffs: enabling efficient cloud workflows

Eugene Cheipesh

The Cloud Optimized GeoTIFF (COG) is a key component to enabling cloud-native geospatial workflows. An important benefit is the ability for the client to stream just the portion of data that is needed through the use of HTTP GET byte-range requests. COGs enable faster reading, writing, and processing of data on the cloud without the need for local copies.

GeoTrellis is an open source, geographic data processing library designed to work with large geospatial raster data sets. It is written in Scala and has an open-source Apache 2.0 license. GeoTrellis could always read COGs, but until GeoTrellis 2.0, the internal structure for saving and serving GeoTrellis layers was based on Apache Avro encoding.

This talk will start with a high level introduction to COGs and a demonstration of what they enable. We will then go deeper into the technical challenges that were faced when implementing COG layer support for distributed, GeoTrellis workflows. This will include discussions of indexing strategies that rely on space-filling curves, overview generation, and query optimization.

The goal of this talk is to appeal to a broad audience. For people that have never heard of COGs, it will give a high level overview of a few use cases that COGs enable. For people that are already familiar with COGs, it will go into some of the deeper technical implementation challenges.

Go to listing –>


Leveraging Vector Tiles to Query Statewide Habitat Data in a Browser

Dan Ford

NJ Landscape Project data combines species location information with land-use/land-cover classes to form a dataset of hundreds of thousands of polygons with continuous, topological coverage for the entire state and several associated tables with rich attribute information about habitat type and wildlife. Basically, this is an impressive dataset with a ton of useful information that drives regulatory, development, and planning decisions across the state. But, this data is difficult to work with because it is provided as several Geodatabases, each with regional vector data sets and related tables. It’s hard to run queries for this species-specific habitat data even if you understand the data and have experience with geospatial technologies. Previously, GIS technicians had to wait minutes or hours to process this huge data set for custom operations in order to answer questions about habitat details on specific sites. Now, a user can draw a custom area in a browser window and return results in seconds, enabling efficient compliance with land-use regulations.

In this talk, we’ll dive into the process we used to make this data more accessible. This project investigates a workflow to process the dataset into vector tiles that can be rendered responsively in a web browser. Learn how we used Python and GDAL for data processing, Tippecanoe for vector tile creation, and Mapbox GL for client side map rendering.

Go to listing –>


Raster Vision: Deep Learning for Aerial and Satellite Imagery

Lewis Fishgold

Increasingly, deep learning is being used to make sense of the vast quantity of satellite and aerial imagery being generated each day. Although there are many open source libraries for deep learning, they are unable to handle geospatial datasets without the addition of a non-trivial amount of extra data processing glue code. Over the past year, we have prototyped workflows for the computer vision tasks of classification, object detection, and semantic segmentation. After reflecting on these early prototypes, we have assembled a set of objectives and architectural principles for a generic library for doing deep learning on geospatial imagery.

We are currently developing an open source library based on these principles called Raster Vision. The overall goal of the library is to make it easy to prepare training data, train models, make predictions, and evaluate models for the three computer vision tasks listed above. In addition, the library is designed to be easy to extend for new computer vision tasks and implementations, and new data sources. In this talk, we will discuss the general problem of doing deep learning on geospatial imagery, the objectives, roadmap, and architecture of the library, and a demo of how to use Raster Vision.

Go to listing –>


Working with OpenStreetMap using Apache Spark and GeoTrellis

Rob Emanuele

OpenStreetMap (OSM) is a rich trove of geospatial data. However, the OSM data model of nodes, ways and relations presents challenges to those of us who are used to working with geospatial data encoded into the Points, LineStrings and Polygons we all know and love. On top of this, it’s quite a large and unwieldy dataset to work with.

In this talk, I’ll describe how Azavea and others have been using Apache Spark and GeoTrellis to do analytics on historical OSM data to answer questions like “how many kilometers of roads have been edited, ever, as part of the #missingmaps mapping campaign?”. I’ll go over the way we’ve utilized the OSM public dataset that Amazon Web Services released last summer, approaches for deriving feature data from OSM data using Apache Spark, and how we used GeoTrellis to bake vector tiles that show heat maps of edit history in MapboxGL. Attendees will get a tour of this new ecosystem of geospatial open source tooling that is just starting to be applied to OSM, and shown how they can get started in using these tools to answer their own big questions from OSM.

Go to listing –>


We’ll also be participating in a panel discussion on OpenStreetMap:

Contributing Every Day: How I Made a Career working with OSM

OpenStreetMap (OSM) is a collaborative project to create a free editable map of the world. It is being recognized as a useful dataset for training machine learning algorithms, deriving analytics on mapping behaviors, providing vital datasets to first responders to natural disasters, and a casual place for geonerds to continue to contribute to a living map of the world.

At Radiant Solutions, we are using OpenStreetMap on a number of different efforts, and are proposing a panel session where we, along with some of our partners in the Open Source Community (including but not limited to Azavea, Development Seed, and/or Terranodo) to discuss and answer questions about how we use OSM each and every day to support the open source community.

Go to listing –>


Let us know if you plan to attend FOSS4G NA 2018 so we can meet up!