Consider the stochastic partial differential equation $$ \frac{\partial }{\partial t}u_t(\mathbf{x})= -(-Δ)^{\fracα{2}}u_t(\mathbf{x}) +b\left(u_t(\mathbf{x})\right)+σ\left(u_t(\mathbf{x})\right) \dot F(t, \mathbf{x}), \ \ \ t\ge0, \mathbf{x}\in \mathbb R^d, $$ where $-(-Δ)^{\fracα{2}}$ denotes the fractional Laplacian with the power $α/2\in (1/2,1]$, and the driving noise $\dot F$ is a centered Gaussian field which is white in time and with a spatial homogeneous covariance given by the Riesz kernel. We study the detailed behavior of the approximation spatial gradient $u_t(\mathbf{x})-u_t(\mathbf{x}-\varepsilon \mathbf e)$ at any fixed time $t>0$, as $\varepsilon\downarrow 0$, where $\mathbf e$ is the unit vector in $\mathbb R^d$. As applications, we deduce the law of iterated logarithm and the behavior of the $q$-variations of the solution in space.