Caleb Robinson

Adam J. Stewart, Nils Lehmann, Isaac A. Corley et al.

The Landsat program is the longest-running Earth observation program in history, with 50+ years of data acquisition by 8 satellites. The multispectral imagery captured by sensors onboard these satellites is critical for a wide range of scientific fields. Despite the increasing popularity of deep learning and remote sensing, the majority of researchers still use decision trees and random forests for Landsat image analysis due to the prevalence of small labeled datasets and lack of foundation models. In this paper, we introduce SSL4EO-L, the first ever dataset designed for Self-Supervised Learning for Earth Observation for the Landsat family of satellites (including 3 sensors and 2 product levels) and the largest Landsat dataset in history (5M image patches). Additionally, we modernize and re-release the L7 Irish and L8 Biome cloud detection datasets, and introduce the first ML benchmark datasets for Landsats 4-5 TM and Landsat 7 ETM+ SR. Finally, we pre-train the first foundation models for Landsat imagery using SSL4EO-L and evaluate their performance on multiple semantic segmentation tasks. All datasets and model weights are available via the TorchGeo (https://github.com/microsoft/torchgeo) library, making reproducibility and experimentation easy, and enabling scientific advancements in the burgeoning field of remote sensing for a multitude of downstream applications.

Anthony Cintron Roman, Kevin Xu, Arfon Smith et al.

GitHub is the world's largest platform for collaborative software development, with over 100 million users. GitHub is also used extensively for open data collaboration, hosting more than 800 million open data files, totaling 142 terabytes of data. This study highlights the potential of open data on GitHub and demonstrates how it can accelerate AI research. We analyze the existing landscape of open data on GitHub and the patterns of how users share datasets. Our findings show that GitHub is one of the largest hosts of open data in the world and has experienced an accelerated growth of open data assets over the past four years. By examining the open data landscape on GitHub, we aim to empower users and organizations to leverage existing open datasets and improve their discoverability -- ultimately contributing to the ongoing AI revolution to help address complex societal issues. We release the three datasets that we have collected to support this analysis as open datasets at https://github.com/github/open-data-on-github.

Girmaw Abebe Tadesse, Titien Bartette, Andrew Hassanali et al.

Looting at archaeological sites poses a severe risk to cultural heritage, yet monitoring thousands of remote locations remains operationally difficult. We present a scalable and satellite-based pipeline to detect looted archaeological sites, using PlanetScope monthly mosaics (4.7m/pixel) and a curated dataset of 1,943 archaeological sites in Afghanistan (898 looted, 1,045 preserved) with multi-year imagery (2016--2023) and site-footprint masks. We compare (i) end-to-end CNN classifiers trained on raw RGB patches and (ii) traditional machine learning (ML) trained on handcrafted spectral/texture features and embeddings from recent remote-sensing foundation models. Results indicate that ImageNet-pretrained CNNs combined with spatial masking reach an F1 score of 0.926, clearly surpassing the strongest traditional ML setup, which attains an F1 score of 0.710 using SatCLIP-V+RF+Mean, i.e., location and vision embeddings fed into a Random Forest with mean-based temporal aggregation. Ablation studies demonstrate that ImageNet pretraining (even in the presence of domain shift) and spatial masking enhance performance. In contrast, geospatial foundation model embeddings perform competitively with handcrafted features, suggesting that looting signatures are extremely localized. The repository is available at https://github.com/microsoft/looted_site_detection.

Van Anh Le, Varshini Reddy, Zixi Chen et al.

In this paper we propose a mask-conditional synthetic image generation model for creating synthetic satellite imagery datasets. Given a dataset of real high-resolution images and accompanying land cover masks, we show that it is possible to train an upstream conditional synthetic imagery generator, use that generator to create synthetic imagery with the land cover masks, then train a downstream model on the synthetic imagery and land cover masks that achieves similar test performance to a model that was trained with the real imagery. Further, we find that incorporating a mixture of real and synthetic imagery acts as a data augmentation method, producing better models than using only real imagery (0.5834 vs. 0.5235 mIoU). Finally, we find that encouraging diversity of outputs in the upstream model is a necessary component for improved downstream task performance. We have released code for reproducing our work on GitHub, see https://github.com/ms-synthetic-satellite-image/synthetic-satellite-imagery .

Konstantin Klemmer, Esther Rolf, Caleb Robinson et al.

Geographic information is essential for modeling tasks in fields ranging from ecology to epidemiology. However, extracting relevant location characteristics for a given task can be challenging, often requiring expensive data fusion or distillation from massive global imagery datasets. To address this challenge, we introduce Satellite Contrastive Location-Image Pretraining (SatCLIP). This global, general-purpose geographic location encoder learns an implicit representation of locations by matching CNN and ViT inferred visual patterns of openly available satellite imagery with their geographic coordinates. The resulting SatCLIP location encoder efficiently summarizes the characteristics of any given location for convenient use in downstream tasks. In our experiments, we use SatCLIP embeddings to improve prediction performance on nine diverse location-dependent tasks including temperature prediction, animal recognition, and population density estimation. Across tasks, SatCLIP consistently outperforms alternative location encoders and improves geographic generalization by encoding visual similarities of spatially distant environments. These results demonstrate the potential of vision-location models to learn meaningful representations of our planet from the vast, varied, and largely untapped modalities of geospatial data.

Caleb Robinson, Anthony Ortiz, Hogeun Park et al.

Innovations in computer vision algorithms for satellite image analysis can enable us to explore global challenges such as urbanization and land use change at the planetary level. However, domain shift problems are a common occurrence when trying to replicate models that drive these analyses to new areas, particularly in the developing world. If a model is trained with imagery and labels from one location, then it usually will not generalize well to new locations where the content of the imagery and data distributions are different. In this work, we consider the setting in which we have a single large satellite imagery scene over which we want to solve an applied problem -- building footprint segmentation. Here, we do not necessarily need to worry about creating a model that generalizes past the borders of our scene but can instead train a local model. We show that surprisingly few labels are needed to solve the building segmentation problem with very high-resolution (0.5m/px) satellite imagery with this setting in mind. Our best model trained with just 527 sparse polygon annotations (an equivalent of 1500 x 1500 densely labeled pixels) has a recall of 0.87 over held out footprints and a R2 of 0.93 on the task of counting the number of buildings in 200 x 200-meter windows. We apply our models over high-resolution imagery in Amman, Jordan in a case study on urban change detection.

Hannah Kerner, Snehal Chaudhari, Aninda Ghosh et al.

Crop field boundaries are foundational datasets for agricultural monitoring and assessments but are expensive to collect manually. Machine learning (ML) methods for automatically extracting field boundaries from remotely sensed images could help realize the demand for these datasets at a global scale. However, current ML methods for field instance segmentation lack sufficient geographic coverage, accuracy, and generalization capabilities. Further, research on improving ML methods is restricted by the lack of labeled datasets representing the diversity of global agricultural fields. We present Fields of The World (FTW) -- a novel ML benchmark dataset for agricultural field instance segmentation spanning 24 countries on four continents (Europe, Africa, Asia, and South America). FTW is an order of magnitude larger than previous datasets with 70,462 samples, each containing instance and semantic segmentation masks paired with multi-date, multi-spectral Sentinel-2 satellite images. We provide results from baseline models for the new FTW benchmark, show that models trained on FTW have better zero-shot and fine-tuning performance in held-out countries than models that aren't pre-trained with diverse datasets, and show positive qualitative zero-shot results of FTW models in a real-world scenario -- running on Sentinel-2 scenes over Ethiopia.

Caleb Robinson, Simone Fobi Nsutezo, Anthony Ortiz et al.

Rapid and accurate building damage assessments from high-resolution satellite imagery following a natural disaster is essential to inform and optimize first responder efforts. However, performing such building damage assessments in an automated manner is non-trivial due to the challenges posed by variations in disaster-specific damage, diversity in satellite imagery, and the dearth of extensive, labeled datasets. To circumvent these issues, this paper introduces a human-in-the-loop workflow for rapidly training building damage assessment models after a natural disaster. This article details a case study using this workflow, executed in partnership with the American Red Cross during a tornado event in Rolling Fork, Mississippi in March, 2023. The output from our human-in-the-loop modeling process achieved a precision of 0.86 and recall of 0.80 for damaged buildings when compared to ground truth data collected post-disaster. This workflow was implemented end-to-end in under 2 hours per satellite imagery scene, highlighting its potential for real-time deployment.

Akram Zaytar, Rohan Sawahn, Caleb Robinson et al.

We present ongoing research on agency primitives for GeoAI assistants -- core capabilities that connect Foundation models to the artifact-centric, human-in-the-loop workflows where GIS practitioners actually work. Despite advances in satellite image captioning, visual question answering, and promptable segmentation, these capabilities have not translated into productivity gains for practitioners who spend most of their time producing vector layers, raster maps, and cartographic products. The gap is not model capability alone but the absence of an agency layer that supports iterative collaboration. We propose a vocabulary of $9$ primitives for such a layer -- including navigation, perception, geo-referenced memory, and dual modeling -- along with a benchmark that measures human productivity. Our goal is a vocabulary that makes agentic assistance in GIS implementable, testable, and comparable.

Simone Fobi, Manuel Cardona, Elliott Collins et al.

This work presents an approach for combining household demographic and living standards survey questions with features derived from satellite imagery to predict the poverty rate of a region. Our approach utilizes visual features obtained from a single-step featurization method applied to freely available 10m/px Sentinel-2 surface reflectance satellite imagery. These visual features are combined with ten survey questions in a proxy means test (PMT) to estimate whether a household is below the poverty line. We show that the inclusion of visual features reduces the mean error in poverty rate estimates from 4.09% to 3.88% over a nationally representative out-of-sample test set. In addition to including satellite imagery features in proxy means tests, we propose an approach for selecting a subset of survey questions that are complementary to the visual features extracted from satellite imagery. Specifically, we design a survey variable selection approach guided by the full survey and image features and use the approach to determine the most relevant set of small survey questions to include in a PMT. We validate the choice of small survey questions in a downstream task of predicting the poverty rate using the small set of questions. This approach results in the best performance -- errors in poverty rate decrease from 4.09% to 3.71%. We show that extracted visual features encode geographic and urbanization differences between regions.

Isaac Corley, Caleb Robinson, Inbal Becker-Reshef et al.

As geospatial foundation models shift from patch-level to pixel-level embeddings, practitioners must aggregate thousands of pixel vectors into patch representations that preserve class-discriminative signal while matching downstream label resolution. The default choice, mean pooling, discards within-patch variability and can drop accuracy by more than 10% under spatial shift. To evaluate this effect, we introduce EuroSAT-Embed: 81,000 embedding GeoTIFFs derived from three foundation models: AlphaEarth, OlmoEarth, and Tessera. We benchmark 11 training-free and 2 parametric pooling methods under both random and geographically disjoint test splits. Our results show that richer pooling schemes reduce the geographic generalization gap by up to 40% relative to mean pooling and increases accuracy by up to 5% on spatial splits. We recommend Generalized Mean Pooling (GeM) as a drop-in replacement for mean pooling: it improves accuracy without increasing embedding dimensionality. For maximum accuracy, Stats pooling (concatenation of min/max/mean/std pooling) performs best at 4x the embedding size. We further find that pooling effectiveness varies across embedding sources and that higher-dimensional embeddings benefit most from distributional statistics.

Tammy Glazer, Gilles Q. Hacheme, Akram Zaytar et al.

We present TEMPO, a global, temporally resolved dataset of building density and height derived from high-resolution satellite imagery using deep learning models. We pair building footprint and height data from existing datasets with quarterly PlanetScope basemap satellite images to train a multi-task deep learning model that predicts building density and building height at a 37.6-meter per pixel resolution. We apply this model to global PlanetScope basemaps from Q1 2018 through Q2 2025 to create global, temporal maps of building density and height. We validate these maps by comparing against existing building footprint datasets. Our estimates achieve an F1 score between 85% and 88% on different hand-labeled subsets, and are temporally stable, with a 0.96 five-year trend-consistency score. TEMPO captures quarterly changes in built settlements at a fraction of the computational cost of comparable approaches, unlocking large-scale monitoring of development patterns and climate impacts essential for global resilience and adaptation efforts.

Girmaw Abebe Tadesse, Titien Bartette, Andrew Hassanali et al.

Monitoring archaeological sites at scale is vital for protecting cultural heritage, yet pinpointing when disturbances occur remains difficult because visual cues are subtle and ground-truth data are sparse. We introduce WATCH, a framework for month-level change-event localization over PlanetScope satellite mosaics (2017-2024, 4.7 m/px) that supports three complementary scoring approaches: (i) Temporal Embedding Distance (TED), a training-free method that scores month-to-month deviations from a local temporal reference; (ii) Self-Supervised Change Detection (SSCD), an ensemble of reconstruction, forecasting, and latent-novelty signals; and (iii) a Weakly Supervised (WS) temporal localization model trained with sparse event-month labels. We benchmark WATCH on 1,943 archaeological sites in Afghanistan using embeddings from six foundation models (CLIP, GeoRSCLIP, SatMAE, Prithvi-EO-2.0, DINOv3, and Satlas-Pretrain) alongside a handcrafted spectral and texture baseline, and assess cross-regional generalization on sites in Syria, Turkey, Pakistan, and Egypt. The unsupervised approaches (TED, SSCD) consistently outperform the weakly supervised alternative. TED with SatMAE achieves the highest exact-month recall (55% at m=0), while TED with GeoRSCLIP, CLIP, or Satlas-Pretrain reaches 92.5% within a three-month tolerance (m=3). Handcrafted features remain competitive for exact-month detection under weak supervision. Our directional margin analysis reveals systematic temporal biases: SSCD paired with GeoRSCLIP or Prithvi-EO-2.0 exhibits the strongest early-warning profile, detecting anomalies before the recorded event, while TED favors confirmation-oriented detection after a change has materialized. These results show that satellite imagery combined with foundation-model embeddings enables scalable, decision-relevant heritage monitoring. Code: https://github.com/microsoft/WATCH

Gedeon Muhawenayo, Caleb Robinson, Subash Khanal et al.

Large-scale maps of field boundaries are essential for agricultural monitoring tasks. Existing deep learning approaches for satellite-based field mapping are sensitive to illumination, spatial scale, and changes in geographic location. We conduct the first systematic evaluation of segmentation and geospatial foundation models (GFMs) for global field boundary delineation using the Fields of The World (FTW) benchmark. We evaluate 18 models under unified experimental settings, showing that a U-Net semantic segmentation model outperforms instance-based and GFM alternatives on a suite of performance and deployment metrics. We propose a new segmentation approach that combines a U-Net backbone, composite loss functions, and targeted data augmentations to enhance performance and robustness under real-world conditions. Our model achieves a 76\% IoU and 47\% object-F1 on FTW, an increase of 6\% and 9\% over the previous baseline. Our approach provides a practical framework for reliable, scalable, and reproducible field boundary delineation across model design, training, and inference. We release all models and model-derived field boundary datasets for five countries.

Burak Ekim, Girmaw Abebe Tadesse, Caleb Robinson et al.

Training robust deep learning models is crucial in Earth Observation, where globally deployed models often face distribution shifts that degrade performance, especially in low-data regions. Out-of-distribution (OOD) detection addresses this by identifying inputs that deviate from in-distribution (ID) data. However, existing methods either assume access to OOD data or compromise primary task performance, limiting real-world use. We introduce TARDIS, a post-hoc OOD detection method designed for scalable geospatial deployment. Our core innovation lies in generating surrogate distribution labels by leveraging ID data within the feature space. TARDIS takes a pre-trained model, ID data, and data from an unknown distribution (WILD), separates WILD into surrogate ID and OOD labels based on internal activations, and trains a binary classifier to detect distribution shifts. We validate on EuroSAT and xBD across 17 setups covering covariate and semantic shifts, showing near-upper-bound surrogate labeling performance in 13 cases and matching the performance of top post-hoc activation- and scoring-based methods. Finally, deploying TARDIS on Fields of the World reveals actionable insights into pre-trained model behavior at scale. The code is available at \href{https://github.com/microsoft/geospatial-ood-detection}{https://github.com/microsoft/geospatial-ood-detection}

Caleb Robinson, Nils Lehmann, Adam J. Stewart et al.

Earth observation machine learning pipelines differ fundamentally from standard computer vision workflows. Imagery is typically delivered as large, georeferenced scenes, labels may be raster masks or vector geometries in distinct coordinate reference systems, and both training and evaluation often require spatially aware sampling and splitting strategies. TorchGeo is a PyTorch-based domain library that provides datasets, samplers, transforms and pre-trained models with the goal of making it easy to use geospatial data in machine learning pipelines. In this paper, we introduce a tutorial that demonstrates 1.) the core TorchGeo abstractions through code examples, and 2.) an end-to-end case study on multispectral water segmentation from Sentinel-2 imagery using the Earth Surface Water dataset. This demonstrates how to train a semantic segmentation model using TorchGeo datasets, apply the model to a Sentinel-2 scene over Rio de Janeiro, Brazil, and save the resulting predictions as a GeoTIFF for further geospatial analysis. The tutorial code itself is distributed as two Python notebooks: https://torchgeo.readthedocs.io/en/stable/tutorials/torchgeo.html and https://torchgeo.readthedocs.io/en/stable/tutorials/earth_surface_water.html.

Caleb Robinson, Isaac Corley, Anthony Ortiz et al.

Fully understanding a complex high-resolution satellite or aerial imagery scene often requires spatial reasoning over a broad relevant context. The human object recognition system is able to understand object in a scene over a long-range relevant context. For example, if a human observes an aerial scene that shows sections of road broken up by tree canopy, then they will be unlikely to conclude that the road has actually been broken up into disjoint pieces by trees and instead think that the canopy of nearby trees is occluding the road. However, there is limited research being conducted to understand long-range context understanding of modern machine learning models. In this work we propose a road segmentation benchmark dataset, Chesapeake Roads Spatial Context (RSC), for evaluating the spatial long-range context understanding of geospatial machine learning models and show how commonly used semantic segmentation models can fail at this task. For example, we show that a U-Net trained to segment roads from background in aerial imagery achieves an 84% recall on unoccluded roads, but just 63.5% recall on roads covered by tree canopy despite being trained to model both the same way. We further analyze how the performance of models changes as the relevant context for a decision (unoccluded roads in our case) varies in distance. We release the code to reproduce our experiments and dataset of imagery and masks to encourage future research in this direction -- https://github.com/isaaccorley/ChesapeakeRSC.

Keiller Nogueira, Akram Zaytar, Wanli Ma et al.

The increasing accessibility of remotely sensed data and the potential of such data to inform large-scale decision-making has driven the development of deep learning models for many Earth Observation tasks. Traditionally, such models must be trained on large datasets. However, the common assumption that broadly larger datasets lead to better outcomes tends to overlook the complexities of the data distribution, the potential for introducing biases and noise, and the computational resources required for processing and storing vast datasets. Therefore, effective solutions should consider both the quantity and quality of data. In this paper, we propose six novel core-set selection methods for selecting important subsets of samples from remote sensing image segmentation datasets that rely on imagery only, labels only, and a combination of each. We benchmark these approaches against a random-selection baseline on three commonly used land cover classification datasets: DFC2022, Vaihingen, and Potsdam. In each of the datasets, we demonstrate that training on a subset of samples outperforms the random baseline, and some approaches outperform training on all available data. This result shows the importance and potential of data-centric learning for the remote sensing domain. The code is available at https://github.com/keillernogueira/data-centric-rs-classification/.

Isaac Corley, Nils Lehmann, Caleb Robinson et al.

Geospatial foundation models (GFMs) have been proposed as generalizable backbones for disaster response, land-cover mapping, food-security monitoring, and other high-stakes Earth-observation tasks. Yet the published work about these models does not give reviewers or users enough information to tell which model fits a given task. We argue that nobody knows what the current state of the art is in geospatial foundation models. The methods may be useful, but the GFM literature does not standardize evaluations, training and testing protocols, released weights, or pretraining controls well enough for anyone to compare or rank them. In a 152-paper audit, we find 46 cross-paper disagreements of at least 10 points for the same model, benchmark, and protocol; 94/126 papers with extractable pretraining data use a configuration no other paper uses; and 39% of GFM papers release no model weights. This lack of community standards can be solved. We propose six concrete expectations: named-license weight release, shared core evaluations, copied-versus-rerun baseline annotations, variance reporting, one shared evaluation harness, and data-vs-architecture-vs-algorithm controls. These gaps are a coordination failure, not a fault of any individual lab; the authors of this paper, like many others in the GFM community, have contributed to them. Rather than just critiquing the community, we aim to provide concrete steps toward a shared understanding of how to innovate GFMs.

Caleb Robinson, Gedeon Muhawenayo, Subash Khanal et al.

The agricultural field is the natural unit at which crops are planted, managed, regulated, and reported, yet most global remote-sensing products for agriculture are only available at the pixel level. While some high-quality field-level data products exist, they come from parcel registries covering only parts of Europe or from ML-derived products for individual countries. No openly available, globally consistent map of agricultural field boundaries exists to date. Here we present the first global field boundary dataset at 10\,m resolution for the years 2024 and 2025, comprising 3.17 billion remote-sensing field polygons (1.62 B in 2024 and 1.55 B in 2025) across 241 countries and territories, produced by applying a U-Net segmentation model trained on the Fields of The World dataset to cloud-free Sentinel-2 mosaics. Validated against ground-truth field boundaries in 24 countries, the map achieved a mean pixel-level recall of 0.85 with 14 countries exceeding 0.90. Evaluation against full-country ground-truth datasets in Austria, Latvia, and Finland yielded F1 scores of 0.89, 0.88, and 0.74, respectively. Because reference data for global validation is inherently incomplete, we accompanied the map with a 500 m confidence layer that identifies regions where predictions are reliable. We release the dataset openly as three global maps: the confidence-thresholded default field boundary dataset, the full unfiltered dataset, and the continuous-valued confidence raster. These maps provide the first globally consistent field-level unit of analysis for crop monitoring, food security, and downstream agricultural science.

Caleb Robinson, Kimberly T. Goetz, Christin B. Khan et al.

Effective monitoring of whale populations is critical for conservation, but traditional survey methods are expensive and difficult to scale. While prior work has shown that whales can be identified in very high-resolution (VHR) satellite imagery, large-scale automated detection remains challenging due to a lack of annotated imagery, variability in image quality and environmental conditions, and the cost of building robust machine learning pipelines over massive remote sensing archives. We present a semi-automated approach for surfacing possible whale detections in VHR imagery using a statistical anomaly detection method that flags spatial outliers, i.e. "interesting points". We pair this detector with a web-based labeling interface designed to enable experts to quickly annotate the interesting points. We evaluate our system on three benchmark scenes with known whale annotations and achieve recalls of 90.3% to 96.4%, while reducing the area requiring expert inspection by up to 99.8% -- from over 1,000 sq km to less than 2 sq km in some cases. Our method does not rely on labeled training data and offers a scalable first step toward future machine-assisted marine mammal monitoring from space. We have open sourced this pipeline at https://github.com/microsoft/whales.

Adam J. Stewart, Caleb Robinson, Arindam Banerjee

Recent advances in machine learning have been supported by the emergence of domain-specific software libraries, enabling streamlined workflows and increased reproducibility. For geospatial machine learning (GeoML), the availability of Earth observation data has outpaced the development of domain libraries to handle its unique challenges, such as varying spatial resolutions, spectral properties, temporal cadence, data coverage, coordinate systems, and file formats. This chapter presents a comprehensive overview of GeoML libraries, analyzing their evolution, core functionalities, and the current ecosystem. It also introduces popular GeoML libraries such as TorchGeo, eo-learn, and Raster Vision, detailing their architecture, supported data types, and integration with ML frameworks. Additionally, it discusses common methodologies for data preprocessing, spatial--temporal joins, benchmarking, and the use of pretrained models. Through a case study in crop type mapping, it demonstrates practical applications of these tools. Best practices in software design, licensing, and testing are highlighted, along with open challenges and future directions, particularly the rise of foundation models and the need for governance in open-source geospatial software. Our aim is to guide practitioners, developers, and researchers in navigating and contributing to the rapidly evolving GeoML landscape.

Gilles Quentin Hacheme, Girmaw Abebe Tadesse, Caleb Robinson et al.

Classifying geospatial imagery remains a major bottleneck for applications such as disaster response and land-use monitoring-particularly in regions where annotated data is scarce or unavailable. Existing tools (e.g., RS-CLIP) that claim zero-shot classification capabilities for satellite imagery nonetheless rely on task-specific pretraining and adaptation to reach competitive performance. We introduce GeoVision Labeler (GVL), a strictly zero-shot classification framework: a vision Large Language Model (vLLM) generates rich, human-readable image descriptions, which are then mapped to user-defined classes by a conventional Large Language Model (LLM). This modular, and interpretable pipeline enables flexible image classification for a large range of use cases. We evaluated GVL across three benchmarks-SpaceNet v7, UC Merced, and RESISC45. It achieves up to 93.2% zero-shot accuracy on the binary Buildings vs. No Buildings task on SpaceNet v7. For complex multi-class classification tasks (UC Merced, RESISC45), we implemented a recursive LLM-driven clustering to form meta-classes at successive depths, followed by hierarchical classification-first resolving coarse groups, then finer distinctions-to deliver competitive zero-shot performance. GVL is open-sourced at https://github.com/microsoft/geo-vision-labeler to catalyze adoption in real-world geospatial workflows.

Akram Zaytar, Girmaw Abebe Tadesse, Caleb Robinson et al.

Acquiring, processing, and visualizing geospatial data requires significant computing resources, especially for large spatio-temporal domains. This challenge hinders the rapid discovery of predictive features, which is essential for advancing geospatial modeling. To address this, we developed Similarity Search (Sims), a no-code web tool that allows users to perform clustering and similarity search over defined regions of interest using Google Earth Engine as a backend. Sims is designed to complement existing modeling tools by focusing on feature exploration rather than model creation. We demonstrate the utility of Sims through a case study analyzing simulated maize yield data in Rwanda, where we evaluate how different combinations of soil, weather, and agronomic features affect the clustering of yield response zones. Sims is open source and available at https://github.com/microsoft/Sims

Caleb Robinson, Ben Chugg, Brandon Anderson et al.

Concentrated Animal Feeding Operations (CAFOs) pose serious risks to air, water, and public health, but have proven to be challenging to regulate. The U.S. Government Accountability Office notes that a basic challenge is the lack of comprehensive location information on CAFOs. We use the USDA's National Agricultural Imagery Program (NAIP) 1m/pixel aerial imagery to detect poultry CAFOs across the continental United States. We train convolutional neural network (CNN) models to identify individual poultry barns and apply the best performing model to over 42 TB of imagery to create the first national, open-source dataset of poultry CAFOs. We validate the model predictions against held-out validation set on poultry CAFO facility locations from 10 hand-labeled counties in California and demonstrate that this approach has significant potential to fill gaps in environmental monitoring.

Adam J. Stewart, Caleb Robinson, Isaac A. Corley et al.

Remotely sensed geospatial data are critical for applications including precision agriculture, urban planning, disaster monitoring and response, and climate change research, among others. Deep learning methods are particularly promising for modeling many remote sensing tasks given the success of deep neural networks in similar computer vision tasks and the sheer volume of remotely sensed imagery available. However, the variance in data collection methods and handling of geospatial metadata make the application of deep learning methodology to remotely sensed data nontrivial. For example, satellite imagery often includes additional spectral bands beyond red, green, and blue and must be joined to other geospatial data sources that can have differing coordinate systems, bounds, and resolutions. To help realize the potential of deep learning for remote sensing applications, we introduce TorchGeo, a Python library for integrating geospatial data into the PyTorch deep learning ecosystem. TorchGeo provides data loaders for a variety of benchmark datasets, composable datasets for generic geospatial data sources, samplers for geospatial data, and transforms that work with multispectral imagery. TorchGeo is also the first library to provide pre-trained models for multispectral satellite imagery (e.g., models that use all bands from the Sentinel-2 satellites), allowing for advances in transfer learning on downstream remote sensing tasks with limited labeled data. We use TorchGeo to create reproducible benchmark results on existing datasets and benchmark our proposed method for preprocessing geospatial imagery on the fly. TorchGeo is open source and available on GitHub: https://github.com/microsoft/torchgeo.

Nikolay Malkin, Caleb Robinson, Nebojsa Jojic

We present simple algorithms for land cover change detection in the 2021 IEEE GRSS Data Fusion Contest. The task of the contest is to create high-resolution (1m / pixel) land cover change maps of a study area in Maryland, USA, given multi-resolution imagery and label data. We study several baseline models for this task and discuss directions for further research. See https://dfc2021.blob.core.windows.net/competition-data/dfc2021_index.txt for the data and https://github.com/calebrob6/dfc2021-msd-baseline for an implementation of these baselines.

Fred Hohman, Haekyu Park, Caleb Robinson et al.

Deep learning is increasingly used in decision-making tasks. However, understanding how neural networks produce final predictions remains a fundamental challenge. Existing work on interpreting neural network predictions for images often focuses on explaining predictions for single images or neurons. As predictions are often computed from millions of weights that are optimized over millions of images, such explanations can easily miss a bigger picture. We present Summit, an interactive system that scalably and systematically summarizes and visualizes what features a deep learning model has learned and how those features interact to make predictions. Summit introduces two new scalable summarization techniques: (1) activation aggregation discovers important neurons, and (2) neuron-influence aggregation identifies relationships among such neurons. Summit combines these techniques to create the novel attribution graph that reveals and summarizes crucial neuron associations and substructures that contribute to a model's outcomes. Summit scales to large data, such as the ImageNet dataset with 1.2M images, and leverages neural network feature visualization and dataset examples to help users distill large, complex neural network models into compact, interactive visualizations. We present neural network exploration scenarios where Summit helps us discover multiple surprising insights into a prevalent, large-scale image classifier's learned representations and informs future neural network architecture design. The Summit visualization runs in modern web browsers and is open-sourced.

Esther Rolf, Konstantin Klemmer, Caleb Robinson et al.

Satellite data has the potential to inspire a seismic shift for machine learning -- one in which we rethink existing practices designed for traditional data modalities. As machine learning for satellite data (SatML) gains traction for its real-world impact, our field is at a crossroads. We can either continue applying ill-suited approaches, or we can initiate a new research agenda that centers around the unique characteristics and challenges of satellite data. This position paper argues that satellite data constitutes a distinct modality for machine learning research and that we must recognize it as such to advance the quality and impact of SatML research across theory, methods, and deployment. We outline critical discussion questions and actionable suggestions to transform SatML from merely an intriguing application area to a dedicated research discipline that helps move the needle on big challenges for machine learning and society.

Isaac Corley, Caleb Robinson, Anthony Ortiz

In recent years, there has been an explosion of proposed change detection deep learning architectures in the remote sensing literature. These approaches claim to offer state-of-the-art performance on different standard benchmark datasets. However, has the field truly made significant progress? In this paper we perform experiments which conclude a simple U-Net segmentation baseline without training tricks or complicated architectural changes is still a top performer for the task of change detection.

Anthony Cintron Roman, Jennifer Wortman Vaughan, Valerie See et al. · microsoft-research

This paper introduces a no-code, machine-readable documentation framework for open datasets, with a focus on responsible AI (RAI) considerations. The framework aims to improve comprehensibility, and usability of open datasets, facilitating easier discovery and use, better understanding of content and context, and evaluation of dataset quality and accuracy. The proposed framework is designed to streamline the evaluation of datasets, helping researchers, data scientists, and other open data users quickly identify datasets that meet their needs and organizational policies or regulations. The paper also discusses the implementation of the framework and provides recommendations to maximize its potential. The framework is expected to enhance the quality and reliability of data used in research and decision-making, fostering the development of more responsible and trustworthy AI systems.

Caleb Robinson, Anthony Ortiz, Allen Kim et al.

We present a comprehensive global temporal dataset of commercial solar photovoltaic (PV) farms and onshore wind turbines, derived from high-resolution satellite imagery analyzed quarterly from the fourth quarter of 2017 to the second quarter of 2024. We create this dataset by training deep learning-based segmentation models to identify these renewable energy installations from satellite imagery, then deploy them on over 13 trillion pixels covering the world. For each detected feature, we estimate the construction date and the preceding land use type. This dataset offers crucial insights into progress toward sustainable development goals and serves as a valuable resource for policymakers, researchers, and stakeholders aiming to assess and promote effective strategies for renewable energy deployment. Our final spatial dataset includes 375,197 individual wind turbines and 86,410 solar PV installations. We aggregate our predictions to the country level -- estimating total power capacity based on construction date, solar PV area, and number of windmills -- and find an $r^2$ value of $0.96$ and $0.93$ for solar PV and onshore wind respectively compared to IRENA's most recent 2023 country-level capacity estimates.

Gilles Quentin Hacheme, Akram Zaytar, Girmaw Abebe Tadesse et al.

Cropland mapping can play a vital role in addressing environmental, agricultural, and food security challenges. However, in the context of Africa, practical applications are often hindered by the limited availability of high-resolution cropland maps. Such maps typically require extensive human labeling, thereby creating a scalability bottleneck. To address this, we propose an approach that utilizes unsupervised object clustering to refine existing weak labels, such as those obtained from global cropland maps. The refined labels, in conjunction with sparse human annotations, serve as training data for a semantic segmentation network designed to identify cropland areas. We conduct experiments to demonstrate the benefits of the improved weak labels generated by our method. In a scenario where we train our model with only 33 human-annotated labels, the F_1 score for the cropland category increases from 0.53 to 0.84 when we add the mined negative labels.

Isaac Corley, Simone Fobi Nsutezo, Anthony Ortiz et al.

Remote sensing imagery is dense with objects and contextual visual information. There is a recent trend to combine paired satellite images and text captions for pretraining performant encoders for downstream tasks. However, while contrastive image-text methods like CLIP enable vision-language alignment and zero-shot classification ability, vision-only downstream performance tends to degrade compared to image-only pretraining, such as MAE. In this paper, we propose FLAVARS, a pretraining method that combines the best of both contrastive learning and masked modeling, along with geospatial alignment via contrastive location encoding. We find that FLAVARS significantly outperforms a baseline of SkyCLIP for vision-only tasks such as KNN classification and semantic segmentation, +6\% mIOU on SpaceNet1, while retaining the ability to perform zero-shot classification, unlike MAE pretrained methods.

Girmaw Abebe Tadesse, Caleb Robinson, Charles Mwangi et al.

In 2023, 58.0% of the African population experienced moderate to severe food insecurity, with 21.6% facing severe food insecurity. Land-use and land-cover maps provide crucial insights for addressing food insecurity by improving agricultural efforts, including mapping and monitoring crop types and estimating yield. The development of global land-cover maps has been facilitated by the increasing availability of earth observation data and advancements in geospatial machine learning. However, these global maps exhibit lower accuracy and inconsistencies in Africa, partly due to the lack of representative training data. To address this issue, we propose a data-centric framework with a teacher-student model setup, which uses diverse data sources of satellite images and label examples to produce local land-cover maps. Our method trains a high-resolution teacher model on images with a resolution of 0.331 m/pixel and a low-resolution student model on publicly available images with a resolution of 10 m/pixel. The student model also utilizes the teacher model's output as its weak label examples through knowledge transfer. We evaluated our framework using Murang'a county in Kenya, renowned for its agricultural productivity, as a use case. Our local models achieved higher quality maps, with improvements of 0.14 in the F1 score and 0.21 in Intersection-over-Union, compared to the best global model. Our evaluation also revealed inconsistencies in existing global maps, with a maximum agreement rate of 0.30 among themselves. Our work provides valuable guidance to decision-makers for driving informed decisions to enhance food security.

Ando Shah, Rajveer Singh, Akram Zaytar et al.

Rice cultivation supplies half the world's population with staple food, while also being a major driver of freshwater depletion--consuming roughly a quarter of global freshwater--and accounting for approx. 48% of greenhouse gas emissions from croplands. In regions like Punjab, India, where groundwater levels are plummeting at 41.6 cm/year, adopting water-saving rice farming practices is critical. Direct-Seeded Rice (DSR) and Alternate Wetting and Drying (AWD) can cut irrigation water use by 20-40% without hurting yields, yet lack of spatial data on adoption impedes effective adaptation policy and climate action. We present a machine learning framework to bridge this data gap by monitoring sustainable rice farming at scale. In collaboration with agronomy experts and a large-scale farmer training program, we obtained ground-truth data from 1,400 fields across Punjab. Leveraging this partnership, we developed a novel dimensional classification approach that decouples sowing and irrigation practices, achieving F1 scores of 0.8 and 0.74 respectively, solely employing Sentinel-1 satellite imagery. Explainability analysis reveals that DSR classification is robust while AWD classification depends primarily on planting schedule differences, as Sentinel-1's 12-day revisit frequency cannot capture the higher frequency irrigation cycles characteristic of AWD practices. Applying this model across 3 million fields reveals spatial heterogeneity in adoption at the state level, highlighting gaps and opportunities for policy targeting. Our district-level adoption rates correlate well with government estimates (Spearman's $ρ$=0.69 and Rank Biased Overlap=0.77). This study provides policymakers and sustainability programs a powerful tool to track practice adoption, inform targeted interventions, and drive data-driven policies for water conservation and climate mitigation at regional scale.

Girmaw Abebe Tadesse, Caleb Robinson, Gilles Quentin Hacheme et al.

This study explores object detection in historical aerial photographs of Namibia to identify long-term environmental changes. Specifically, we aim to identify key objects -- Waterholes, Omuti homesteads, and Big trees -- around Oshikango in Namibia using sub-meter gray-scale aerial imagery from 1943 and 1972. In this work, we propose a workflow for analyzing historical aerial imagery using a deep semantic segmentation model on sparse hand-labels. To this end, we employ a number of strategies including class-weighting, pseudo-labeling and empirical p-value-based filtering to balance skewed and sparse representations of objects in the ground truth data. Results demonstrate the benefits of these different training strategies resulting in an average $F_1=0.661$ and $F_1=0.755$ over the three objects of interest for the 1943 and 1972 imagery, respectively. We also identified that the average size of Waterhole and Big trees increased while the average size of Omuti homesteads decreased between 1943 and 1972 reflecting some of the local effects of the massive post-Second World War economic, agricultural, demographic, and environmental changes. This work also highlights the untapped potential of historical aerial photographs in understanding long-term environmental changes beyond Namibia (and Africa). With the lack of adequate satellite technology in the past, archival aerial photography offers a great alternative to uncover decades-long environmental changes.

Akram Zaytar, Caleb Robinson, Gilles Q. Hacheme et al.

Rare object detection is a fundamental task in applied geospatial machine learning, however is often challenging due to large amounts of high-resolution satellite or aerial imagery and few or no labeled positive samples to start with. This paper addresses the problem of bootstrapping such a rare object detection task assuming there is no labeled data and no spatial prior over the area of interest. We propose novel offline and online cluster-based approaches for sampling patches that are significantly more efficient, in terms of exposing positive samples to a human annotator, than random sampling. We apply our methods for identifying bomas, or small enclosures for herd animals, in the Serengeti Mara region of Kenya and Tanzania. We demonstrate a significant enhancement in detection efficiency, achieving a positive sampling rate increase from 2% (random) to 30%. This advancement enables effective machine learning mapping even with minimal labeling budgets, exemplified by an F1 score on the boma detection task of 0.51 with a budget of 300 total patches.

Isaac Corley, Caleb Robinson, Rahul Dodhia et al.

Research in self-supervised learning (SSL) with natural images has progressed rapidly in recent years and is now increasingly being applied to and benchmarked with datasets containing remotely sensed imagery. A common benchmark case is to evaluate SSL pre-trained model embeddings on datasets of remotely sensed imagery with small patch sizes, e.g., 32x32 pixels, whereas standard SSL pre-training takes place with larger patch sizes, e.g., 224x224. Furthermore, pre-training methods tend to use different image normalization preprocessing steps depending on the dataset. In this paper, we show, across seven satellite and aerial imagery datasets of varying resolution, that by simply following the preprocessing steps used in pre-training (precisely, image sizing and normalization methods), one can achieve significant performance improvements when evaluating the extracted features on downstream tasks -- an important detail overlooked in previous work in this space. We show that by following these steps, ImageNet pre-training remains a competitive baseline for satellite imagery based transfer learning tasks -- for example we find that these steps give +32.28 to overall accuracy on the So2Sat random split dataset and +11.16 on the EuroSAT dataset. Finally, we report comprehensive benchmark results with a variety of simple baseline methods for each of the seven datasets, forming an initial benchmark suite for remote sensing imagery.

Esther Rolf, Nikolay Malkin, Alexandros Graikos et al.

We propose a method for jointly inferring labels across a collection of data samples, where each sample consists of an observation and a prior belief about the label. By implicitly assuming the existence of a generative model for which a differentiable predictor is the posterior, we derive a training objective that allows learning under weak beliefs. This formulation unifies various machine learning settings; the weak beliefs can come in the form of noisy or incomplete labels, likelihoods given by a different prediction mechanism on auxiliary input, or common-sense priors reflecting knowledge about the structure of the problem at hand. We demonstrate the proposed algorithms on diverse problems: classification with negative training examples, learning from rankings, weakly and self-supervised aerial imagery segmentation, co-segmentation of video frames, and coarsely supervised text classification.

Anthony Ortiz, Dhaval Negandhi, Sagar R Mysorekar et al.

Rapid development of renewable energy sources, particularly solar photovoltaics (PV), is critical to mitigate climate change. As a result, India has set ambitious goals to install 500 gigawatts of solar energy capacity by 2030. Given the large footprint projected to meet renewables energy targets, the potential for land use conflicts over environmental values is high. To expedite development of solar energy, land use planners will need access to up-to-date and accurate geo-spatial information of PV infrastructure. In this work, we developed a spatially explicit machine learning model to map utility-scale solar projects across India using freely available satellite imagery with a mean accuracy of 92%. Our model predictions were validated by human experts to obtain a dataset of 1363 solar PV farms. Using this dataset, we measure the solar footprint across India and quantified the degree of landcover modification associated with the development of PV infrastructure. Our analysis indicates that over 74% of solar development In India was built on landcover types that have natural ecosystem preservation, or agricultural value.

Caleb Robinson, Anthony Ortiz, Lacey Hughey et al.

Localizing and counting large ungulates -- hoofed mammals like cows and elk -- in very high-resolution satellite imagery is an important task for supporting ecological studies. Prior work has shown that this is feasible with deep learning based methods and sub-meter multi-spectral satellite imagery. We extend this line of work by proposing a baseline method, CowNet, that simultaneously estimates the number of animals in an image (counts), as well as predicts their location at a pixel level (localizes). We also propose an methodology for evaluating such models on counting and localization tasks across large scenes that takes the uncertainty of noisy labels and the information needed by stakeholders in ecological monitoring tasks into account. Finally, we benchmark our baseline method with state of the art vision methods for counting objects in scenes. We specifically test the temporal generalization of the resulting models over a large landscape in Point Reyes Seashore, CA. We find that the LC-FCN model performs the best and achieves an average precision between 0.56 and 0.61 and an average recall between 0.78 and 0.92 over three held out test scenes.

Caleb Robinson, Anthony Ortiz, Juan M. Lavista Ferres et al.

Longitudinal studies are vital to understanding dynamic changes of the planet, but labels (e.g., buildings, facilities, roads) are often available only for a single point in time. We propose a general model, Temporal Cluster Matching (TCM), for detecting building changes in time series of remotely sensed imagery when footprint labels are observed only once. The intuition behind the model is that the relationship between spectral values inside and outside of building's footprint will change when a building is constructed (or demolished). For instance, in rural settings, the pre-construction area may look similar to the surrounding environment until the building is constructed. Similarly, in urban settings, the pre-construction areas will look different from the surrounding environment until construction. We further propose a heuristic method for selecting the parameters of our model which allows it to be applied in novel settings without requiring data labeling efforts (to fit the parameters). We apply our model over a dataset of poultry barns from 2016/2017 high-resolution aerial imagery in the Delmarva Peninsula and a dataset of solar farms from a 2020 mosaic of Sentinel 2 imagery in India. Our results show that our model performs as well when fit using the proposed heuristic as it does when fit with labeled data, and further, that supervised versions of our model perform the best among all the baselines we test against. Finally, we show that our proposed approach can act as an effective data augmentation strategy -- it enables researchers to augment existing structure footprint labels along the time dimension and thus use imagery from multiple points in time to train deep learning models. We show that this improves the spatial generalization of such models when evaluated on the same change detection task.

Jean-Francois Rajotte, Sumit Mukherjee, Caleb Robinson et al.

We introduce FELICIA (FEderated LearnIng with a CentralIzed Adversary) a generative mechanism enabling collaborative learning. In particular, we show how a data owner with limited and biased data could benefit from other data owners while keeping data from all the sources private. This is a common scenario in medical image analysis where privacy legislation prevents data from being shared outside local premises. FELICIA works for a large family of Generative Adversarial Networks (GAN) architectures including vanilla and conditional GANs as demonstrated in this work. We show that by using the FELICIA mechanism, a data owner with limited image samples can generate high-quality synthetic images with high utility while neither data owners has to provide access to its data. The sharing happens solely through a central discriminator that has access limited to synthetic data. Here, utility is defined as classification performance on a real test set. We demonstrate these benefits on several realistic healthcare scenarions using benchmark image datasets (MNIST, CIFAR-10) as well as on medical images for the task of skin lesion classification. With multiple experiments, we show that even in the worst cases, combining FELICIA with real data gracefully achieves performance on par with real data while most results significantly improves the utility.

Lucas Hu, Caleb Robinson, Bistra Dilkina

Recent work has shown that deep learning models can be used to classify land-use data from geospatial satellite imagery. We show that when these deep learning models are trained on data from specific continents/seasons, there is a high degree of variability in model performance on out-of-sample continents/seasons. This suggests that just because a model accurately predicts land-use classes in one continent or season does not mean that the model will accurately predict land-use classes in a different continent or season. We then use clustering techniques on satellite imagery from different continents to visualize the differences in landscapes that make geospatial generalization particularly difficult, and summarize our takeaways for future satellite imagery-related applications.

Nikolay Malkin, Anthony Ortiz, Caleb Robinson et al.

We show that simple patch-based models, such as epitomes, can have superior performance to the current state of the art in semantic segmentation and label super-resolution, which uses deep convolutional neural networks. We derive a new training algorithm for epitomes which allows, for the first time, learning from very large data sets and derive a label super-resolution algorithm as a statistical inference algorithm over epitomic representations. We illustrate our methods on land cover mapping and medical image analysis tasks.

Anthony Ortiz, Caleb Robinson, Dan Morris et al.

Normalization layers have been shown to improve convergence in deep neural networks, and even add useful inductive biases. In many vision applications the local spatial context of the features is important, but most common normalization schemes including Group Normalization (GN), Instance Normalization (IN), and Layer Normalization (LN) normalize over the entire spatial dimension of a feature. This can wash out important signals and degrade performance. For example, in applications that use satellite imagery, input images can be arbitrarily large; consequently, it is nonsensical to normalize over the entire area. Positional Normalization (PN), on the other hand, only normalizes over a single spatial position at a time. A natural compromise is to normalize features by local context, while also taking into account group level information. In this paper, we propose Local Context Normalization (LCN): a normalization layer where every feature is normalized based on a window around it and the filters in its group. We propose an algorithmic solution to make LCN efficient for arbitrary window sizes, even if every point in the image has a unique window. LCN outperforms its Batch Normalization (BN), GN, IN, and LN counterparts for object detection, semantic segmentation, and instance segmentation applications in several benchmark datasets, while keeping performance independent of the batch size and facilitating transfer learning.

Caleb Robinson, Anthony Ortiz, Kolya Malkin et al.

We propose incorporating human labelers in a model fine-tuning system that provides immediate user feedback. In our framework, human labelers can interactively query model predictions on unlabeled data, choose which data to label, and see the resulting effect on the model's predictions. This bi-directional feedback loop allows humans to learn how the model responds to new data. Our hypothesis is that this rich feedback allows human labelers to create mental models that enable them to better choose which biases to introduce to the model. We compare human-selected points to points selected using standard active learning methods. We further investigate how the fine-tuning methodology impacts the human labelers' performance. We implement this framework for fine-tuning high-resolution land cover segmentation models. Specifically, we fine-tune a deep neural network -- trained to segment high-resolution aerial imagery into different land cover classes in Maryland, USA -- to a new spatial area in New York, USA. The tight loop turns the algorithm and the human operator into a hybrid system that can produce land cover maps of a large area much more efficiently than the traditional workflows. Our framework has applications in geospatial machine learning settings where there is a practically limitless supply of unlabeled data, of which only a small fraction can feasibly be labeled through human efforts.

Caleb Robinson, Bistra Dilkina

Human migration is a type of human mobility, where a trip involves a person moving with the intention of changing their home location. Predicting human migration as accurately as possible is important in city planning applications, international trade, spread of infectious diseases, conservation planning, and public policy development. Traditional human mobility models, such as gravity models or the more recent radiation model, predict human mobility flows based on population and distance features only. These models have been validated on commuting flows, a different type of human mobility, and are mainly used in modeling scenarios where large amounts of prior ground truth mobility data are not available. One downside of these models is that they have a fixed form and are therefore not able to capture more complicated migration dynamics. We propose machine learning models that are able to incorporate any number of exogenous features, to predict origin/destination human migration flows. Our machine learning models outperform traditional human mobility models on a variety of evaluation metrics, both in the task of predicting migrations between US counties as well as international migrations. In general, predictive machine learning models of human migration will provide a flexible base with which to model human migration under different what-if conditions, such as potential sea level rise or population growth scenarios.

Caleb Robinson, Fred Hohman, Bistra Dilkina

Knowing where people live is a fundamental component of many decision making processes such as urban development, infectious disease containment, evacuation planning, risk management, conservation planning, and more. While bottom-up, survey driven censuses can provide a comprehensive view into the population landscape of a country, they are expensive to realize, are infrequently performed, and only provide population counts over broad areas. Population disaggregation techniques and population projection methods individually address these shortcomings, but also have shortcomings of their own. To jointly answer the questions of "where do people live" and "how many people live there," we propose a deep learning model for creating high-resolution population estimations from satellite imagery. Specifically, we train convolutional neural networks to predict population in the USA at a $0.01^{\circ} \times 0.01^{\circ}$ resolution grid from 1-year composite Landsat imagery. We validate these models in two ways: quantitatively, by comparing our model's grid cell estimates aggregated at a county-level to several US Census county-level population projections, and qualitatively, by directly interpreting the model's predictions in terms of the satellite image inputs. We find that aggregating our model's estimates gives comparable results to the Census county-level population projections and that the predictions made by our model can be directly interpreted, which give it advantages over traditional population disaggregation methods. In general, our model is an example of how machine learning techniques can be an effective tool for extracting information from inherently unstructured, remotely sensed data to provide effective solutions to social problems.