I guess this is standard and there is a citable reference. I think the following is an argument which only uses the formalism (e.g. also works in the étale case). (This is an edited version of my first incorrect answer.) Firstly, if $X \to Y \to Z \stackrel{+1}{\to}$ is a distinguished triangle with $X, Y$ of wt $\ge 0$, then $Z$ is of weight $\ge 0$. (This is easy if one thinks about Frobenius eigenvalues.) Dually, if $X \to Y \to Z \stackrel{+1}{\to}$ is a dt with $Y, Z$ of wt $\le 0$ then $X$ is of wt $\le 0$. Now the constant sheaf $k_Y$ on $Y$ is of wt $\le 0$. Hence the dualizing sheaf $\omega_Y$ on $Y$ is of wt $\ge 0$. Let $f : X \to Y$ be as in your question. Consider the distinguished triangle $K \to f_!f^!\omega_Y \to \omega_Y \stackrel{+1}{\to}$ (where $K$ is defined as the shift of the cone over the adjunction morphism $f_!f^! \to id$). The above remarks show that $K$ is of weights $\ge -1$. <b>Claim:</b> We are done if we can show $K$ is of weights $\ge 0$. <i>Proof:</i> Pushing to a point we get a long exact sequence $\dots \to H^k(Y,K) \to H^k(X,\omega_X) \to H^k(Y,\omega_Y) \to H^{k+1}(Y,K) \to \dots$ now everything in $H^{k+1}(Y,K)$ is of weight $\ge k+1$ (because $*$-pushforward preserves wt $\ge 0$). We conclude that we have a surjection $gr_W^kH^k(X,\omega_X) \to gr_W^kH^k(X,\omega_Y)$ Finally, because $H_k^{BM}(X) = H^{-k}(X,\omega_X)$ (Borel-Moore homology) the claim follows. We now give a sketch of how to prove the claim. Because this argument is getting more complicated than I had first intended, I'll give a sketch. I can try to provide more details if it is useful for you. Consider a weight filtration $W$ on $\omega_Y$ (it is not "the" weight filtration because $\omega_Y$ is not necessarily perverse). I claim that $gr^W_{\le 0}(\omega_Y) = gr^W_0(\omega_Y)$ is isomorphic to $IC(Y)[d_Y](d_Y)$ (where $d_Y = dim Y$). This basic idea is that $\omega_Y$ does not have any sections supported on subvarieties, and any other $IC$ in $gr_{\le 0}(\omega_Y)$ would contribute such a forbidden section. Similarly, if $W$ denotes a weight filtration on $\omega_X$ then $gr^W_{0}(\omega_X)$ is $IC(X)[d_X](d_X)$. Now consider the adjunction map $f_!f^!\omega_Y \to \omega_Y$. The weight zero part is given by the map $f_!IC(X) = f_! gr^W_0(\omega_X) \to gr^W_0(\omega_Y)$. Now by the decomposition theorem (here we use surjectivity) $IC(Y)$ occurs as a summand of $f_!IC(X)$ in smallest degree. Now one deduces (I can only see how to do this using generic smoothness at the moment) that there exists $IC(Y)[?](?) \to f_!\omega_X$ such that the induced map $$IC(Y)[?](?) \to gr^W_0(\omega_Y) = IC(Y)[?](?)$$ is an isomorphism. We conclude that the triangle $K \to f_!\omega_X \to \omega_Y \stackrel{+1}{\to}$ can be replaced by a triangle $K \to L \to gr^W_{\ge 1}(\omega_Y) \stackrel{+1}{\to}$ with $L$ of wts $\ge 0$ and $gr^W_{\ge 1}(\omega_Y)$ of weights $\ge 1$. We conclude that $K$ has weights $\ge 0$ as claimed.