Sims: An Interactive Tool for Geospatial Matching and Clustering

Akram Zaytar, Girmaw Abebe Tadesse, Caleb Robinson, Eduardo G. Bendito, Medha Devare, Meklit Chernet, Gilles Quentin Hacheme, Rahul Dodhia, Juan M. Lavista FerresarXiv 2024 · DOI · arXiv

Abstract

Geospatial analysis often requires significant computational resources and specialized programming skills, creating a "capacity hurdle" for domain experts. We present Sims (Similarity Search), an open-source, no-code web tool designed to democratize geospatial feature exploration. By leveraging Google Earth Engine (GEE) as a high-performance backend, Sims enables users to perform complex clustering and similarity searches on petabytes of satellite and environmental data through an intuitive interface. We demonstrate the tool's utility through a case study in Rwanda, where it successfully identified distinct agricultural zones with statistically significant differences in crop performance.

Introduction

Geospatial data—including satellite imagery, soil maps, and weather grids—is inherently "heavy" and computationally demanding. Processing such data typically requires expertise in Geographic Information Systems (GIS) and specialized programming. Furthermore, spatial data often violates the "independent and identically distributed" (i.i.d.) assumption central to many machine learning models due to spatial autocorrelation: the tendency for nearby locations to be more similar than distant ones.

Currently, many researchers, such as agronomists or disaster relief coordinators, select data layers based on intuition because the iteration loop for rigorous exploration is too slow or technically demanding. Sims aims to bridge this gap by focusing on feature exploration, helping users identify relevant variables before committing to formal model building.

Methods

Sims is built on a modern Python stack utilizing geemap, ipyleaflet, and Solara. To handle massive datasets without requiring high-end local hardware, the tool offloads all heavy computation to Google Earth Engine (GEE). This architecture allows the application to run on a modest virtual machine while accessing petabytes of environmental data.

Overview of the Sims user interface and core functional components.

The tool supports two primary workflows:

  1. Clustering: Utilizing the wekaMeans algorithm (a GEE implementation of K-means), pixels within a region of interest are grouped based on a multi-dimensional "feature profile" (e.g., rainfall, soil pH, and temperature).
  2. Similarity Search: Users define a reference "template" region and a "query" search area. The tool calculates feature-space distances (Euclidean, Manhattan, or Cosine) to produce a similarity heatmap.

A key innovation is the Alias and Expression system, which allows users to rename cryptic GEE data strings (e.g., MODIS/061/MOD11A2) to human-readable terms like "Max Temp" and create custom features like NDVI using simple mathematical expressions.

Case Study: Rwanda Maize Yield

The utility of Sims was evaluated using a case study focused on maize yield in Rwanda. The target variable consisted of simulated yield data from 2005–2015. Feature layers were pulled from GEE, including SoilGrids (clay, sand, nitrogen, etc.), CHIRPS rainfall, MODIS Land Surface Temperature, and FAO WAPOR evapotranspiration.

Step-by-step workflow for geospatial clustering in the Rwanda case study.

Results

In the Rwanda study, Sims was used to segment the country into five distinct clusters ($K=5$). The tool identified regions with highly significant differences in maize performance ($p < 2e-16$ for most pairs). This segmentation helps account for spatial dependency, allowing for more localized and accurate agricultural recommendations, such as fertilizer application.

The tool significantly accelerated the "feature discovery" phase, delivering complex spatial results within minutes. It successfully identified which feature combinations (e.g., integrating soil data with weather data) best captured variations in crop yield, providing immediate visual feedback that allows for rapid iteration.

Discussion and Future Work

While Sims lowers the barrier to entry for geospatial analysis, it still requires users to possess domain knowledge to select appropriate datasets from the GEE catalog. Current limitations include a lack of automated parameter suggestions (e.g., optimal $K$ for clustering), lack of session persistence, and no warnings for computationally expensive large-scale queries.

Future developments will focus on adding template files for workflow sharing, automated meta-selection of layers, and support for time-series analysis to better capture seasonal patterns. By removing the coding requirement, Sims empowers domain experts to interact directly with data, reducing the gap between research and data engineering.

Summary updated

Abstract

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
BibTeX
@misc{zaytar2024sims,
  author = {Akram Zaytar and Girmaw Abebe Tadesse and Caleb Robinson and Eduardo G. Bendito and Medha Devare and Meklit Chernet and Gilles Quentin Hacheme and Rahul Dodhia and Juan M. Lavista Ferres},
  title = {{Sims: An Interactive Tool for Geospatial Matching and Clustering}},
  year = {2024},
  eprint = {2412.10184},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2412.10184},
}
Figure