Contains the raw terrain data, including elevations, hills, valleys, and mountains.

The challenge of MSTS lies not just in volume, but in heterogeneity. Data sources often differ in sampling rates, signal-to-noise ratios, and relevance to the target variable. Simply concatenating these inputs into a monolithic model often leads to the "curse of dimensionality" and noise amplification.

Let us define a standard multi-source time series setup. We have a set of $N$ source time series denoted as $S = X^(1), X^(2), ..., X^(N)$ and a target series $Y$. Each source $X^(i) \in \mathbbR^T \times D_i$ may have different dimensions $D_i$ and temporal resolutions.