Random effect with compound symmetric covariance structure in glmmTMB

Question

I have a dataset of abundance of a kind of rodents(AA) in 63 sites, and want to know which environmental factors can explain the rodent abundance, and because the sampling sites are basically locate in two parts of my research area (northern part and southern part, NS), I used glmmTMB function to build a negative-binomial mixed model with NS as the random effect.

Random effect are fitted with compound symmetric variance-covariance structure, here is the code:

glmmTMB(AA~ x1 + x2 + x3 + x4 + x5 + cs(1|sites/NS),  
                         family="nbinom2"(link = "log"), data=Rglmm2)

And the error returns

Model convergence problem; non-positive-definite Hessian matrix. See vignette('troubleshooting')

According to the information in vignette('troubleshooting'), it looks like the setting of my random effect and the variance-covariance structure part (cs(1|sites/NS)) went wrong, since sites are nested in the NS, how should I set it up right?

Thanks!