We propose to take a new perspective on the construction and interpretation of multivariate curve resolution (MCR) models for the decomposition of spectral mixture data. We start by introducing archetypes, i.e. points that approximate the convex hull of a data cloud and correspond to the most linearly dissimilar observations. Identifying archetypes is a way to select essential samples (ESs) and essential variables (EVs) of a data matrix before MCR decomposition. Working with ESs and EVs, we then identify three main implications. The first is data reduction, which brings simplicity and computational speed. The second is prioritization, with the ESs and EVs profiles being the most dominant features to solve the MCR problem. The third is interpretability: the reduced data sets provide more direct insights and better understanding of final MCR models. Overall, the selection of ESs and EVs offers new opportunities that are worth being explored.

我们建议对光谱混合数据分解的多元曲线分辨率（MCR）模型的构建和解释采取新的观点。我们从介绍原型开始，即接近数据云凸包并对应于线性最不相似的观察点的点。识别原型是在MCR分解之前选择数据矩阵的基本样本（ES）和基本变量（EV）的一种方法。然后，通过使用ES和EV，我们确定了三个主要含义。首先是数据缩减，它带来了简单性和计算速度。第二个是优先级划分，ES和EV配置文件是解决MCR问题的最主要特征。第三是可解释性：减少的数据集提供了更直接的见解和对最终MCR模型的更好理解。