Parallelize nested loop with running sum in Fortran

Question

I would like to understand if it is possible to parallelize the following nested loop (using OpenMP in Fortran).

To give some background information, I am doing this to initialize a sparse matrix. It turns out that the matrix-vector multiplication is very fast, thanks to the sparse BLAS library. The bottleneck in the code is however the loops that I do to create the COO representation of the matrix (row index, col index and nonzero values). So the question is whether it is possible to parallelize this (perhaps using a reduction clause):

nnz = N*N
        allocate(row_ind(nnz),col_ind(nnz),values(nnz),stat=istat)
        if (istat/=0) then
            error stop "Allocation of row_ind,col_ind,values failed"
        endif

        counter = 1
        do iz = 1,n_z
            do ia = 1,n_a
                ix = (iz-1)*n_a+ia
                opt_ind = pol_ap_ind(ia,iz)
                do izp = 1,n_z
                    if (Pi(iz,izp)/=0.0d0) then
                        ixp = (izp-1)*n_a+opt_ind
                        values(counter) = Pi(iz,izp) 
                        row_ind(counter) = ix
                        col_ind(counter) = ixp
                        counter = counter+1
                    endif 
                enddo
            enddo
        enddo
        nnz = counter -1
        row_ind = row_ind(1:nnz)
        col_ind = col_ind(1:nnz)
        values = values(1:nnz)