Remote imaging spectroscopy, also known as hyperspectral imaging, uses Radiative Transfer Models (RTMs) to predict the measured radiance spectrum for a specific surface and atmospheric state. Discrepancies between RTM assumptions and physical reality can cause systematic errors in surface property estimates. We present a statistical method to quantify these model errors without invoking ground reference data. Our approach exploits scene invariants — properties of the environment which are stable over space or time — to estimate RTM discrepancies. We describe techniques for discovering these features opportunistically in flight data. We then demonstrate data-driven methods that estimate the aggregate errors due to model discrepancies without having to explicitly identify the underlying physical mechanisms. The resulting distributions can improve posterior uncertainty predictions in operational retrievals.

远程成像光谱学（也称为高光谱成像）使用辐射传递模型（RTM）来预测特定表面和大气状态下测得的辐射光谱。RTM假设与物理现实之间的差异会导致表面性质估计中的系统误差。我们提出一种统计方法来量化这些模型误差而无需调用地面参考数据。我们的方法利用场景不变性-在空间或时间上稳定的环境属性-估算RTM差异。我们描述了在飞行数据中机会性地发现这些特征的技术。然后，我们演示了数据驱动的方法，该方法可以估计由于模型差异而引起的汇总错误，而不必明确标识底层的物理机制。所得的分布可以改善操作检索中的后验不确定性预测。