Validation Rigor Determines Apparent Predictive Skill of UAV-LiDAR Carbon Models: A Cautionary Case Study in a Heterogeneous Tropical Savanna
バリデーションの厳密性がUAV-LiDAR炭素モデルの予測精度を左右する：不均一な熱帯サバンナにおける注意喚起事例 (AI 翻訳)
Louzada RO, Silva RHd, Correia SAC, Couto RMP, Vasconcelos DC, Bozza AN, Chaib JG, Sales MG, Tavares RMB
🤖 gxceed AI 要約
日本語
UAV-LiDARと機械学習を用いた炭素貯留量推定では、空間的自己相関を無視したバリデーションが精度を過大評価することを実証。ブラジルのサバンナで98通りのシナリオを検証した結果、単一分割では見かけ上の精度が高いが、複数の空間分割では予測力がほぼゼロに低下。バリデーション戦略が精度評価の主要因であることを示唆。
English
This study demonstrates that ignoring spatial autocorrelation in validation inflates apparent accuracy of UAV-LiDAR and machine-learning carbon stock models. Testing 98 scenario-response combinations in a Brazilian savanna, single-partition validation showed moderate accuracy (best R²=0.27 for AGB), but repeating with 20 spatial partitions collapsed predictive skill to near-zero or negative R². Validation strategy, not sensor choice, drove apparent performance, cautioning against overinterpreting results from single-partition validation in heterogeneous landscapes.
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
本論文の知見は、日本の森林や農地でのUAV-LiDARや衛星データを用いた炭素貯留量推定にも適用可能。特にJCMなどの炭素クレジット制度において、精度評価の厳密性が求められる。
In the global GX context
This paper provides a critical caution for global carbon accounting efforts using remote sensing and ML, especially for nature-based carbon credits and REDD+. It highlights the risk of overestimating model accuracy when spatial autocorrelation is not addressed, which is essential for credible carbon reporting under frameworks like TCFD and ISSB.
👥 読者別の含意
🔬研究者:Emphasizes the necessity of multiple spatial partitions in validation to avoid inflated accuracy estimates in carbon stock models.
🏢実務担当者:Carbon project developers should verify that model validation accounts for spatial autocorrelation before relying on accuracy claims.
🏛政策担当者:Carbon credit methodologies should mandate rigorous validation that addresses spatial autocorrelation to ensure environmental integrity.
📄 Abstract（原文）
Machine-learning models trained on remote sensing data are widely used to map ecosystem carbon stocks, but their validation often ignores spatial autocorrelation, inflating apparent predictive skill. We tested whether this concern, established for large-scale satellite-based biomass mapping, also holds for UAV-LiDAR carbon estimation at the plot scale, comparing continuous structural metrics, segmentation-derived crown metrics, and Sentinel-2 spectral variables for predicting aboveground biomass (AGB) and soil organic carbon (SOC) across 29 inventory plots in a Cerradão-dominated Pantanal–Cerrado ecotone, Brazil. Random Forest models were evaluated under fourteen predictor scenarios and two allometric formulations. A single spatial partition yielded seemingly reasonable accuracy for AGB (best R² = 0.27), whereas SOC showed weak apparent skill even under a single partition (best rRMSE = 33.7%, R² = −0.14); repeating validation across 20 independent spatial partitions, combined with leakage-free nested variable selection, showed predictive skill collapsing to near-zero or negative R² for all 98 tested scenario–response combinations (mean R²: −0.44 to −0.003). Variable selection remained comparatively stable across partitions, with a Sentinel-2 shortwave-infrared band and canopy vertical-complexity metrics consistently retained, and cross-pool coupling tests showed no meaningful gain from using AGB to predict SOC or vice versa. Validation strategy, rather than sensor combination or vegetation representation, was the primary determinant of apparent predictive skill, cautioning against overinterpreting single-partition accuracy in small, heterogeneous tropical landscapes.
🔗 Provenance — このレコードを発見したソース
- Research Square https://doi.org/10.20944/preprints202607.1819.v1first seen 2026-07-30 04:54:39
🔔 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。