Introducing PEARL — AI Accelerated Land Cover Mapping Platform


1 min read

Sajjad Anwar, Lane Goodman, Martha Morrisey, Nick Ingalls, Vincent Sarago, Vitor George, Sanjay Bhangar, Jeevan Farias, Zhuangfang NaNa Yi

Development Seed and Microsoft AI for Earth are launching PEARL, an AI-accelerated platform for fast land cover mapping, for experimental use. PEARL represents a new approach to combining human intelligence and scalable AI for fast, accurate land cover mapping. It leverages capabilities and research under the Microsoft Planetary Computer initiative that significantly reduce the effort to create land cover maps, allowing scientists and researchers to focus on the most pressing environmental and climate research questions.

Two key innovations make PEARL possible. Working with the Microsoft team, we extended their research into cloud infrastructure that makes machine learning model retraining fast enough to be conducted interactively in the browser. We then designed an entirely new user experience that leverages this infrastructure to create intuitive prediction, feedback, and retraining interactions. The result is a tool that allows any user to create and refine a land cover classification model in the browser.

PEARL currently is experimental and is available for public use. The current version includes imagery from the National Agriculture Imagery Program (NAIP), which limits its availability to the United States. Our next release will include global coverage using Sentinel-2 imagery.

Rethinking land cover mapping

Accurate, timely and accessible land cover maps are critical for conservation, climate research, and planning. Scientists and analysts currently rely on costly and time-intensive processes to generate bespoke land cover maps. Global land cover datasets exist and are useful for some purposes. However, publicly accessible maps are often out-of-date, low-resolution, or inaccurate, particularly outside of the US and Western Europe.

We believe that AI offers an immediate opportunity to rapidly speed up land cover mapping, but we also believe that the expertise of geospatial analysts is still critical to produce accurate land cover maps. Consequently, Development Seed has invested heavily in human-in-the-loop AI tools — our Data Team has used this approach to speed up high quality mapping by as much as 30x. Microsoft AI for Earth and Microsoft Research scientists Nebojsa Jojic, Caleb Robinson, Kolya Malkin, Dan Morris et al. have confirmed that overall accuracy of the system goes up in a human collaboration approach in their recent research titled Human-Machine Collaboration for Fast Land Cover Mapping.

Make a land cover map in minutes

PEARL leverages this approach and new research at Microsoft and Development Seed, allowing users to produce a land cover map in minutes. The platform provides imagery and starter models that immediately get you running on your area of interest. You provide feedback to the model on where by correctly labelling areas that it predicted incorrectly. Along the way we provide stats that help you to provide balanced feedback that will improve the model accuracy. You can iterate this process for as long as you like. When you are happy you can export the end result as a GeoTIFF or an interactive map.

PEARL combines machine learning, open data, and open source software with scalable infrastructure on Microsoft Azure. PEARL does not require any data to be brought to the platform — the user gets access to imagery and models. PEARL abstracts away infrastructure and enables scientists to start mapping immediately, instead of engaging in data procurement and complex preprocessing steps.

A core principle behind the design of PEARL is to provide scientists and researchers easy access to infrastructure that is otherwise expensive, hard to setup and manage. A user working on PEARL connects their browser directly to a GPU that runs models and does all the computation. Behind the scenes, a Kubernetes cluster enables scheduling multiple users to perform inference and retraining. Development Seed’s TiTiler provides an imagery service that generates tiles dynamically from mosaics hosted on Azure Blob Storage. The backend infrastructure provides persistent sessions to each user though a REST API and WebSocket connections.

Effortless machine learning

We have been building tools to make machine learning more accessible, so that scientists and researchers can focus on decision making and policy changes. A key element of this work is developing new but familiar user interactions. The application borrows many of the common patterns of map editing applications, giving users an improved retraining sample selection experience. Map creators can select individual points or draw freehand shapes to define image areas for retraining.

Retraining sometimes may not be enough for a high quality map. We introduce map refinements to make final adjustments. Users can treat classes and previous checkpoints as a brush, filling in freehand shapes. In this way, the final map can sample from any of the trained results, and users can remove unwanted pixel noise or minor aberrations.

PEARL Starter Model Metadata

Throughout the map creation process, we focus on surfacing key metrics. The model metadata cards provide the user with context to help select a starter model. This involved the creation of a new metadata schema for models that captures information like class distribution of the training dataset, labels, label source, imagery source, geographic location of the training data set, and per class performance over the hold out test dataset.

AI that keeps getting better

This initial version of PEARL contains two FCN segmentation starter models trained with nine and four land cover classes based on labeled data from the Chesapeake Conservancy’s dataset. Both of these starter models have a global F1 score of just under 90%. When users provide feedback we retrain the model by updating the parameters of the last layer of the model using the point labels provided by the users. Users can improve the performance of the model for a local area and can even define new LULC classes.

As people use PEARL they produce new model checkpoints that work better for their area of interest. These checkpoints can them be applied to larger areas of interest. Future versions of PEARL will include more starter models that cover more regions and classes. We also intend to allow users to save and share their checkpoints and to contribute high performing models back to the starter model library.

We hope PEARL helps speed up your land cover mapping workflow. If you have any questions or comments, let us know!

What we're doing.