Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Contract-Governed Training for Earth Observation: Observed Service Agreement Graphs and Coverage-Accuracy Trade-offs

Wenzhang Du

TL;DR

EO training often optimizes global accuracy without guarantees on coverage for mission-critical strata. OSAG introduces a contract-governed training layer with contract-normalized sampling and two control knobs, alpha and lambda_C, to enable explicit governance over which contracts are served. The paper provides a toy theory linking coverage, service-risk bounds, and contract design, and demonstrates substantial reductions in priority coverage error with maintained global accuracy and improved high-priority performance on HSI and MSI datasets; a coarse-vs-fine contract ablation on EuroSAT confirms governance-cost modulation via contract design. Collectively, OSAG offers a practical, explainable approach to policy-driven EO model training that can complement existing backbones and operational constraints.

Abstract

Earth observation (EO) models are frequently trained under implicit sampling policies that optimize global accuracy but provide no explicit guarantees on who (which regions, classes, or mission-critical strata) is being served throughout training. This paper introduces a contract-governed training paradigm for EO in which training samples are grouped into service contracts -- semantically meaningful units such as (dataset, region, rare-crop indicator) -- and each contract is assigned a target service share. We instantiate this paradigm as an Observed Service Agreement Graph (OSAG), a lightweight governance layer that (i) monitors contract-level exposure (coverage) during optimization, (ii) drives empirical coverage toward target shares via contract-normalized sampling weights, and (iii) exposes explicit accuracy-governance trade-offs through two knobs: a sampling mixture coefficient alpha and a contract-regularization weight lambda_C. We provide a compact theory in a toy setting: OSAG sampling concentrates empirical coverage to targets; coverage deviations upper-bound service-risk deviations; and contract design (coarse vs. fine) modulates governance cost. Experiments on AVIRIS hyperspectral scenes (Indian Pines plus Salinas) and multispectral Sentinel-2 EuroSAT demonstrate that OSAG can substantially reduce priority coverage error while maintaining global accuracy and improving high-priority accuracy. A EuroSAT coarse-vs-fine contract ablation further evidences how semantically refined contracts can reduce the accuracy cost per unit of governance improvement.

Contract-Governed Training for Earth Observation: Observed Service Agreement Graphs and Coverage-Accuracy Trade-offs

TL;DR

EO training often optimizes global accuracy without guarantees on coverage for mission-critical strata. OSAG introduces a contract-governed training layer with contract-normalized sampling and two control knobs, alpha and lambda_C, to enable explicit governance over which contracts are served. The paper provides a toy theory linking coverage, service-risk bounds, and contract design, and demonstrates substantial reductions in priority coverage error with maintained global accuracy and improved high-priority performance on HSI and MSI datasets; a coarse-vs-fine contract ablation on EuroSAT confirms governance-cost modulation via contract design. Collectively, OSAG offers a practical, explainable approach to policy-driven EO model training that can complement existing backbones and operational constraints.

Abstract

Earth observation (EO) models are frequently trained under implicit sampling policies that optimize global accuracy but provide no explicit guarantees on who (which regions, classes, or mission-critical strata) is being served throughout training. This paper introduces a contract-governed training paradigm for EO in which training samples are grouped into service contracts -- semantically meaningful units such as (dataset, region, rare-crop indicator) -- and each contract is assigned a target service share. We instantiate this paradigm as an Observed Service Agreement Graph (OSAG), a lightweight governance layer that (i) monitors contract-level exposure (coverage) during optimization, (ii) drives empirical coverage toward target shares via contract-normalized sampling weights, and (iii) exposes explicit accuracy-governance trade-offs through two knobs: a sampling mixture coefficient alpha and a contract-regularization weight lambda_C. We provide a compact theory in a toy setting: OSAG sampling concentrates empirical coverage to targets; coverage deviations upper-bound service-risk deviations; and contract design (coarse vs. fine) modulates governance cost. Experiments on AVIRIS hyperspectral scenes (Indian Pines plus Salinas) and multispectral Sentinel-2 EuroSAT demonstrate that OSAG can substantially reduce priority coverage error while maintaining global accuracy and improving high-priority accuracy. A EuroSAT coarse-vs-fine contract ablation further evidences how semantically refined contracts can reduce the accuracy cost per unit of governance improvement.

Paper Structure

This paper contains 20 sections, 9 equations, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Problem Setup: Contracts, Coverage, and Service Risk
  3. Contracts and target service shares
  4. Empirical coverage and priority coverage error
  5. Contract-level service risk
  6. OSAG: Governance via Contract-Normalized Sampling
  7. Trade-off knob 1: sampling mixture $\alpha$
  8. Trade-off knob 2: contract regularization $\lambda_C$
  9. Toy Theory (Compact): Why Coverage Control Matters
  10. Experimental Protocol
  11. Datasets
  12. Representation and model (architecture-orthogonal evaluation)
  13. Contract definitions and priority targets
  14. Baselines and OSAG policies
  15. Metrics
  16. ...and 5 more sections

Figures (2)

  • Figure 1: Governance trade-off on Indian+Salinas and EuroSAT: Acc$_{\text{high}}$ vs. PrioCovErr (log, lower is better). Error bars show std over 3 seeds.
  • Figure 2: EuroSAT coarse vs. fine contracts: baseline $\rightarrow$ OSAG arrows in (PrioCovErr, Acc$_{\text{all}}$).