Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Cropland Mapping using Geospatial Embeddings

Ivan Zvonkov, Gabriel Tseng, Inbal Becker-Reshef, Hannah Kerner

TL;DR

The paper tackles the need for up-to-date cropland maps to study land-use change and its climate implications. It investigates geospatial embeddings, or GeoFMs, for cropland mapping by generating embeddings from Presto and AlphaEarth over all of Togo at 10 m resolution with 128 features per pixel and using a Random Forest classifier trained on CropHarvest labels. The results show that embedding-based cropland maps are simple to produce, fast (under 5 seconds per map), and achieve high accuracy, with Presto achieving the best metrics and both embedding approaches aligning well with, yet sometimes exceeding, WorldCover in cropland detection. The work demonstrates a scalable workflow for rapid, future-ready land-use monitoring that can support climate impact assessments and policy decisions, with plans to broaden geographic scope.

Abstract

Accurate and up-to-date land cover maps are essential for understanding land use change, a key driver of climate change. Geospatial embeddings offer a more efficient and accessible way to map landscape features, yet their use in real-world mapping applications remains underexplored. In this work, we evaluated the utility of geospatial embeddings for cropland mapping in Togo. We produced cropland maps using embeddings from Presto and AlphaEarth. Our findings show that geospatial embeddings can simplify workflows, achieve high-accuracy cropland classification and ultimately support better assessments of land use change and its climate impacts.

Cropland Mapping using Geospatial Embeddings

TL;DR

The paper tackles the need for up-to-date cropland maps to study land-use change and its climate implications. It investigates geospatial embeddings, or GeoFMs, for cropland mapping by generating embeddings from Presto and AlphaEarth over all of Togo at 10 m resolution with 128 features per pixel and using a Random Forest classifier trained on CropHarvest labels. The results show that embedding-based cropland maps are simple to produce, fast (under 5 seconds per map), and achieve high accuracy, with Presto achieving the best metrics and both embedding approaches aligning well with, yet sometimes exceeding, WorldCover in cropland detection. The work demonstrates a scalable workflow for rapid, future-ready land-use monitoring that can support climate impact assessments and policy decisions, with plans to broaden geographic scope.

Abstract

Accurate and up-to-date land cover maps are essential for understanding land use change, a key driver of climate change. Geospatial embeddings offer a more efficient and accessible way to map landscape features, yet their use in real-world mapping applications remains underexplored. In this work, we evaluated the utility of geospatial embeddings for cropland mapping in Togo. We produced cropland maps using embeddings from Presto and AlphaEarth. Our findings show that geospatial embeddings can simplify workflows, achieve high-accuracy cropland classification and ultimately support better assessments of land use change and its climate impacts.

Paper Structure

This paper contains 8 sections, 13 figures, 2 tables.

Table of Contents

  1. Introduction & Related Work
  2. Methodology
  3. Embedding generation
  4. Mapping using embeddings
  5. Results & Conclusion
  6. Conclusion
  7. Appendix
  8. Additional Qualitative Analysis

Figures (13)

  • Figure 1: Comparison of WorldCover land cover map and clustered embeddings in Togo. We used the default WorldCover color scheme. The cluster colors were randomly generated.
  • Figure 2: Steps involved in generating embeddings and mapping cropland.
  • Figure 4: High resolution satellite image zoomed in on coordinate: (0.84445, 6.46924).
  • Figure 5: GLAD cropland map GLAD zoomed in on coordinate: (0.84445, 6.46924).
  • Figure 6: WorldCover cropland map WorldCoverMap zoomed in on coordinate: (0.84445, 6.46924).
  • ...and 8 more figures