Every quantum channel has many Kraus representations that may differ in the number of Kraus operators. For example, for any positive integer $n$ and numbers $p_i$ with $i=1,\dots,n$ and $\sum_{i=1}^np_i=1$ the matrices $E_i=\sqrt{p_i}I$ form a valid, if impractical, Kraus representation of the identity channel with $n$ Kraus operators. This example also shows that there is no maximum on the number of non-zero Kraus operators one can use to describe a quantum channel.
However, there is a minimum. The minimum is called Kraus rank or Choi rank of the channel. The letter term is justified by the fact that Choi rank is equal to the rank - in the usual linear algebraic sense - of the channel's Choi matrix. This implies that the Choi rank of a channel $\mathcal{E}: L(\mathcal{X})\to L(\mathcal{Y})$ is an integer between $1$ and $\dim\mathcal{X} \cdot \dim\mathcal{Y}$.
Occasionally, a channel's Choi rank is also simply called its "rank". However, this can be misleading as every channel is also a linear map and therefore has the regular, linear algebraic rank which may be different$^1$ than its Choi rank. In a context where one uses the term "rank" to refer to the number of operators in a Kraus representation - as it is done in the paper in question - it is likely that this unfortunate convention is being employed.
$^1$ Consider the amplitude damping channel with $K_0=\begin{bmatrix}1&0 \\ 0&0\end{bmatrix}$ and $K_1=\begin{bmatrix}0&1 \\ 0&0\end{bmatrix}$.