We make use of the Navy's high-resolution tropical cyclone prediction system COAMPS-TC to provide ensemble forecasts, numerical experiments with and without the assimilation of specific observation types (e.g., satellite, dropsondes, high-frequency radiosonde), as well as mesoscale nested adjoint sensitivity and observation impact calculations; all of which provide insight into the initial state sensitivity and predictability issues. We assess the impact of observations in sensitive regions in the TC environment (including outflow regions away from the TC inner core) on predictions of TC intensity and structure. Overall the results underscore the importance of multiple scales that influence the predictability of TC intensification. During HS3, the assimilation of Global Hawk dropsondes has been shown to reduce the maximum wind error from 15 knots to less than 10 knots at 48 h for Hurricane Nadine. In this particular case, the adjoint model shows strong sensitivity in the TC outflow near the entrance region of an upper-level jet. The impact of dropsondes from data denial experiments and adjoint-based observation impact calculations will be discussed for other cases from the HS3 program, including Nadine (2012), Leslie (2012), Gabrielle (2013), Edouard (2014), as well as Fred (2015). Additionally, we will present some preliminary results from the ONR Tropical Cyclone Intensity (TCI-14) experiment that featured dropsondes deployed from the NASA WB-57 and observed Hurricanes Marty, Joaquin and Patricia in 2015.