用于实时燃料含水量反演的统一多传感器机器学习框架
A Unified Multisensor Machine-Learning Framework for Live Fuel Moisture Content Retrieval
浏览论文内容
中文总结 AI 辅助
研究针对实时燃料含水量大面积估算难的问题，开发统一机器学习框架，结合多源数据估算，处理传感器差异，用随机森林和XGBoost回归器训练模型，验证设计下草、灌木和树模型取得较好R2值，条件具备时可纳入额外光学传感器。
中文摘要 AI 辅助
实时燃料含水量控制着植被可燃性，是火灾管理中的重要变量。但由于实地观测集中在特定区域，大面积估算和绘制其地图仍很困难。我们开发了一个统一的机器学习框架，可根据卫星植被指数、气象变量、地形和季节预测因子估算实时燃料含水量。将GlobeLFMC 2.0测量值与多种卫星表面反射率产品匹配，考虑传感器差异进行处理。初步单产品实验表明，天气、地形和一年中的循环日期提供了大部分预测增益。用随机森林和XGBoost回归器训练了单独的草、灌木和树模型。在主要验证设计下，草、灌木和树的最佳模型分别达到0.715、0.693和0.700的合并R2值。该框架在有兼容反射率波段等条件时可纳入额外光学传感器。
英文摘要
Live fuel moisture content controls vegetation flammability and is a high-importance variable in fire management. Nevertheless, it remains difficult to estimate and map over large areas due to the concentration of field observations in specific regions. We develop a unified machine-learning framework that estimates live fuel moisture content from satellite vegetation indices, meteorological variables, topography and seasonal predictors. GlobeLFMC 2.0 measurements are matched to Terra and Aqua MODIS, VIIRS, Landsat 8/9, Sentinel-2 and Sentinel-3 surface-reflectance products, combining the long MODIS record with finer-resolution recent observations. To account for differences among sensors, optical predictors are restricted to a common red, near-infrared and shortwave-infrared feature space; site--product combinations and field time series are screened for remote-sensing suitability; and spectral response function diagnostics are combined with target-independent empirical reflectance calibration toward a Sentinel-2 reference domain. Preliminary single-product experiments show that weather, topography and cyclic day-of-year provide most of the predictive gain beyond vegetation indices, whereas optional product-specific predictors do not justify their additional dependencies. Separate Grass, Shrub and Tree models are trained with Random Forest and XGBoost regressors. Under the primary validation design, which withholds observation dates from sites represented in training, the best models achieve pooled R2 values of 0.715, 0.693 and 0.700 for Grass, Shrub and Tree, respectively. The framework can incorporate additional optical sensors when compatible reflectance bands, documented spectral responses and sufficient overlap observations are available for calibration and validation.
