Agri-environmental schemes are found in most European countries and now account for approximately 4 per cent part of EU expenditure on UK agriculture. A significant part of that expenditure is the cost of monitoring farmer compliance with input restrictions. This paper analyses the design of monitoring schedules for long duration agri-environmental schemes where the aim is to reinstate preferred ecosystems using a Partially Observed Markov Decision Process (POMDP). The approach has much in common with the Arrow-Fisher-Henry model of irreversible land development where there is uncertainty over environmental value. Uncertainty in the Partially Observed Markov Decision Process (POMDP) model analysed here, relates to the current vegetation state, stochastic transitions between vegetation states and monitoring accuracy. It applies to both irreversible changes and changes subject to varying degrees of reversibility. Results from this model present a scheme for monitoring which depends upon the regulators prior probability of vegetation states. Over time, monitoring resolves the uncertainty the regulator has about the vegetation state. For some prior probabilities monitoring is repeated for a number of periods for others no monitoring or one period of monitoring is optimal.