One characterization is the following: $H \succeq 0$ iff the matrix of coefficients
$$\Gamma_{\alpha \beta} = \operatorname{tr}(P_\alpha^\dagger P_\beta H)$$
satisfies $\Gamma \succeq 0$.
This idea is known as a moment matrix or sum of squares proof of positivity.
It derives from the fact that $H \succeq 0$ iff $\operatorname{tr}(H A) \geq 0$ for all $A\succeq 0$.
Because every $A \succeq 0$ can be factorized as $A = B^\dagger B$, it follows that
$H \succeq 0$
iff
$$\operatorname{tr}(BB^\dagger H) \geq 0$$ for all $B = \sum_\alpha b_\alpha P_\alpha$.
This can be rewritten as
$$b^\dagger \Gamma b \geq 0$$
for all vectors $b$, which is equivalent to the claimed condition $\Gamma \succeq 0$.
Fixing the normalization (e.g. if you want a density matrix $\operatorname{tr}(H) = 1$), is done by the requirement $\Gamma_{00} = \operatorname{tr}(\mathbb{1}^\dagger \mathbb{1} H) = 1$, so that $H$ is normalized to trace one.
Requiring that a submatrix of $\Gamma$ is positive semidefinite is a often used relaxation for the positivity of a density matrix (e.g. for quantum Max Cut or maximal violations of Bell non-locality), when one is only given some subset of Pauli coefficients $\operatorname{tr}(P H)$.
One way to think of this is to expand a state not in terms of a vector basis (leading to a density matrix), but a matrix basis (leading to a moment matrix).