The predominant theory of ventral tegmental area (VTA) dopamine neuron function is that they signal reward prediction error (RPE) in their phasic activity. Some recent studies have observed ramping dopamine neuron activity, but its relationship to RPE signalling is currently under debate. We set out to explain the function of this ramp and its relationship to phasic RPEs, using both experimental and theoretical methods. We chose to examine goal-directed navigation, as it requires learning to accurately estimate location and select optimal actions in each location. Given that VTA dopamine neurons are involved in value learning, action selection, and reward location learning, they are ideally placed to provide teaching signals for goal-directed navigation. We characterised VTA dopamine neuron activity by performing calcium imaging using a Miniscope as mice learned to navigate in a closed-loop virtual reality corridor and lick to report a reward location. Across learning, phasic responses resembling RPEs developed, as well as a slow pre-reward ramp in activity. The slope of this ramp was modulated by both learning stage and task engagement. The ramp slope was inversely correlated with locomotor speed, indicating that the ramp did not reflect motor vigour, contradicting previous studies. We considered whether ramping VTA dopamine neuron activity could represent a form of RPE. We devised a Q-learning model that incorporated noisy state inference and an eligibility trace. This model recapitulated our behavioural findings and produced simultaneous phasic and ramping prediction error. The model predicted that a ramp should improve task performance, which we confirmed in our experimental data, indicating that the ramp played a teaching role in the selection of accurate location-specific action during navigation. Our findings provide neural evidence and a theoretical framework to explain ramping dopamine neuron activity as a form of RPE that improves goal-directed navigation.