Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TEMPO: Global Temporal Building Density and Height Estimation from Satellite Imagery

Tammy Glazer, Gilles Q. Hacheme, Akram Zaytar, Luana Marotti, Amy Michaels, Girmaw Abebe Tadesse, Kevin White, Rahul Dodhia, Andrew Zolli, Inbal Becker-Reshef, Juan M. Lavista Ferres, Caleb Robinson

TL;DR

TEMPO delivers global, quarterly maps of building density and height derived from PlanetScope imagery using a single multi-task U-Net trained with weak supervision from complementary building datasets. The approach integrates a density and height regression, rolling time-window smoothing, and geographically informed masking to achieve temporally stable estimates with strong cross-dataset alignment. Key findings include robust performance against external references, improved temporal consistency through post-processing, and clear signals of change in regions such as Dallas and Chad, all achieved with scalable distributed inference. The work enables low-cost, large-scale monitoring of urban development and humanitarian settlement dynamics for resilience and policy planning.

Abstract

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.

TEMPO: Global Temporal Building Density and Height Estimation from Satellite Imagery

TL;DR

TEMPO delivers global, quarterly maps of building density and height derived from PlanetScope imagery using a single multi-task U-Net trained with weak supervision from complementary building datasets. The approach integrates a density and height regression, rolling time-window smoothing, and geographically informed masking to achieve temporally stable estimates with strong cross-dataset alignment. Key findings include robust performance against external references, improved temporal consistency through post-processing, and clear signals of change in regions such as Dallas and Chad, all achieved with scalable distributed inference. The work enables low-cost, large-scale monitoring of urban development and humanitarian settlement dynamics for resilience and policy planning.

Abstract

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.

Paper Structure

This paper contains 33 sections, 2 equations, 6 figures, 5 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Data
  3. Planet Imagery
  4. Building Data
  5. Methods
  6. Generating Building Labels
  7. Multi-task Modeling of Buildings
  8. Model Architecture
  9. Sampling Strategy
  10. Loss Function
  11. Training
  12. Post-Processing
  13. Rolling Time-Window Aggregation
  14. Masking Uninhabitable Locations
  15. Ensuring Agreement
  16. ...and 18 more sections

Figures (6)

  • Figure 1: Estimated global building density for Q2 2025.
  • Figure 2: Our methods are split into a training workflow (top) and inference workflow (bottom). The training workflow consists of fitting a multi-task building density & height segmentation model from weak labels sampled around the globe. The inference workflow consists of running the model on nearly 1PB of quarterly PlanetScope basemap imagery from Q1 2018 through Q2 2025 with post-processing to improve temporal consistency.
  • Figure 3: SpaceNet7 building density labels by geographic location.
  • Figure 4: Examples of Planet quarterly imagery for 2023, digital elevation model data, water masks, UDMs, and UDM confidence values over part of Quevedo, Ecuador.
  • Figure 5: Locations in the greater Dallas, Texas, area that experienced in the top 5% of urban growth from Q2 2020 to Q2 2025. We highlight a concentration of new warehouse and logistics infrastructure north of Forth Worth, Texas.
  • ...and 1 more figures