Optimizing Fine-Tuning of Earth Foundation Models via Multidimensional Latin Hypercube Sampling for Small-Scale Burn Scar Identification
小規模火災跡識別のための多次元ラテン超方格サンプリングによる地球基盤モデルの微調整最適化 (AI 翻訳)
Yuchen Du, Daniel Jacome, Jianghao Wang
🤖 gxceed AI 要約
日本語
本研究は、地球基盤モデル（Prithvi）の微調整を多次元ラテン超方格サンプリング（LHS）で最適化し、小規模火災跡の高精度識別を達成。LHSは単純無作為抽出より優れ、100サンプルでピーク精度の94.5%を維持。炭素会計やデータ不足地域での災害モニタリングに貢献。
English
This study optimizes fine-tuning of the Prithvi Earth Foundation Model using Multidimensional Latin Hypercube Sampling (LHS) for small-scale burn scar identification. LHS outperforms Simple Random Sampling, achieving 0.91 mIoU and retaining 94.5% peak accuracy with only 100 samples. The framework enhances carbon accounting and disaster monitoring in data-constrained regions.
Unofficial AI-generated summary based on the public title and abstract. Not an official translation.
📝 gxceed 編集解説 — Why this matters
日本のGX文脈において
日本では森林火災や農地焼却が炭素会計に影響するが、本手法のデータ効率性は日本の地方自治体や民間企業が限られたデータで高精度な火災影響評価を行う際に有用。SSBJや有報でのGHG算定精度向上に寄与する可能性がある。
In the global GX context
Globally, this work addresses a key challenge in carbon accounting: accurate quantification of emissions from small-scale fires. The data-efficient framework aligns with ISSB and TCFD needs for reliable Scope 1 and 3 GHG data, and supports disaster monitoring in developing regions.
👥 読者別の含意
🔬研究者:Provides a novel sampling strategy (LHS) for fine-tuning Earth Foundation Models, improving segmentation accuracy and data efficiency.
🏢実務担当者:Enables accurate small-scale burn scar mapping with minimal training data, enhancing carbon accounting and GHG reporting for companies and governments.
🏛政策担当者:Offers a scalable tool for monitoring small-scale wildfires and their carbon emissions, supporting climate action and disaster management policies.
📄 Abstract（原文）
Identifying small-scale burn scars is critical for global carbon accounting, yet remains computationally challenging due to spectral complexity and ground truth scarcity in heterogeneous landscapes. Conventional deep learning models often fail to generalize in such environments, lacking both domain-specific priors and representative training distributions required for precise segmentation. Here, we show that optimizing the fine-tuning of the Prithvi Earth Foundation Model (EFM) via Multidimensional Latin Hypercube Sampling (LHS) establishes a robust framework for this task. Our comparative analysis reveals that the domain-adapted Prithvi model achieves a Mean Intersection over Union (mIoU) of 0.91, outperforming standard Vision Transformers (ViT) by 31.9% and significantly surpassing reconstruction-based architectures, such as Scale-MAE. We demonstrate that LHS is superior to Simple Random Sampling (SRS) for optimizing foundation models, as it ensures statistical fidelity with a Kolmogorov–Smirnov (KS) statistic below 0.1 and effectively captures the tail distributions of fire weather indices. Furthermore, our framework exhibited exceptional data efficiency, retaining 94.5% of its peak accuracy with only 100 training samples. These findings provide a scalable solution for monitoring small-scale disasters in data-constrained regions and validate the synergy between rigorous sampling strategies and EFMs.
🔗 Provenance — このレコードを発見したソース
- semanticscholar https://doi.org/10.3390/fire9040161first seen 2026-06-29 08:01:31
🔔 こうした論文の新着を逃したくない方は キーワードアラート に登録（無料・3キーワードまで）。
gxceed は公開メタデータに基づく研究支援データセットです。要約・翻訳・解説は AI 支援で生成されています。 最終的な解釈・検証は利用者が原典資料に基づいて行うことを前提とします。