# A Case for Planetary Computing

**Authors:** Amelia Holcomb, Michael Dales, Patrick Ferris, Sadiq Jaffer, Thomas Swinfield, Alison Eyres, Andrew Balmford, David Coomes, Srinivasan Keshav, Anil Madhavapeddy  
**arXiv:** [2303.04501](https://arxiv.org/abs/2303.04501)  
**Date:** 2023-03-08  
**Summary page:** https://www.summarizepaper.com/en/paper/2303.04501v1/

## Key points

- Comprehensive end-to-end system infrastructure is crucial for planetary computing in processing petabytes of global remote-sensing data for environmental analysis.
- The system should be accessible, interoperable, and extensible to handle morphological changes.
- Selective access control and reproducibility of AI-based inferences are essential features.
- Support for dynamic data flow pipelines, AI-based insights, API endpoints integration, and exploratory query interfaces is necessary.
- Existing cloud-based solutions like Google's Earth Engine (GEE) and Microsoft's Planetary Computer (MPC) offer end-to-end capabilities but may lack in non-technical requirements such as traceability and explainability.
- Custom end-to-end systems using open-source components allow for greater customization but may expose users to more system concerns.
- Land use policy considerations emphasize the importance of accessing high-resolution datasets for evaluating the impact on biodiversity and natural habitats.
- Natural resource managers require scalable data processing capabilities for interactive exploration of potential land use policies at different spatial scales.
- Systems research needs to address challenges related to reproducibility across heterogeneous workloads, scalability of algorithms across CPU/GPU operations, traceability of results while protecting private data, and ensuring permanent data availability.

## Summary

In the context of planetary computing for processing petabytes of global remote-sensing data for environmental analysis, it is crucial to have a comprehensive end-to-end system infrastructure. This system should be accessible, interoperable, and extensible to handle morphological changes. It should also provide selective access control and ensure reproducibility of AI-based inferences. Additionally, it should support dynamic data flow pipelines and offer AI-based insights while integrating API endpoints and facilitating exploratory query interfaces. Existing solutions such as Google's Earth Engine (GEE) and Microsoft's Planetary Computer (MPC) offer cloud-based end-to-end solutions with the ability to ingest satellite imagery, interactively explore datasets, and process data using familiar languages like Python and Javascript. While both platforms excel in end-to-end capabilities, they may fall short in non-technical requirements such as traceability and explainability. Building a custom end-to-end system using open-source components may expose users to more system concerns but allows for greater customization. For instance, land use policy considerations emphasize the importance of accessing high-resolution datasets for evaluating the impact of land use changes on biodiversity and natural habitats. Natural resource managers require scalable data processing capabilities for interactive exploration of potential land use policies at different spatial scales. Overall, there is a need for systems research to address challenges related to reproducibility across heterogeneous workloads, scalability of algorithms across CPU/GPU operations, traceability of results while protecting private data, and ensuring permanent data availability. By refining existing systems or building custom solutions tailored to specific scientific scenarios like land use policy evaluation, advancements in planetary computing can contribute towards addressing climate change and biodiversity crises.