In this work we show the feasibility of empirical testing of socio-environmental models with a relatively simple model of a traditional agro-ecosystem: the Huerta Murcia (Spain). This Mediterranean traditional irrigated land is based on adaptive strategies of irrigation and crop management and has important functions for the conservation of natural and cultural resources. During the last decades, this agro-landscape has been threatened by several factors linked to low profitability of agricultural production under present conditions, the emergence of new irrigated lands outside the river valleys and the land conversion to non-agricultural uses. A dynamic system model has been built to analyse all these processes. Results obtained for the 1932–1995 calibration period show that the model successfully tracks the observed data series for the available variables (area of traditional irrigated lands, number of landowners, population, average farm size and water pollution) and that the Huerta had lost more than 2000ha, explained by competition for land use and the decreasing profitability of farming due to the reduced farm size, the appearance of water deficit and the increasing water pollution. After model testing, including dimensional consistence tests, sensitive analysis and extreme condition tests, a set of policy scenarios (1995–2025) were explored. Land planning policies seem to be more effective than agricultural policies. Then, the observed actual behaviour (1995–2008) of Huerta Murcia was compared to results expected under the analysed scenarios results show an actual rate of loss of Huerta (1995–2008) higher than expected under the base trend (business as usual) scenario. This accelerated loss can be explained by the successive changes in the Municipal Urban Plan, leading to a notable increase in buildable area and to increasing expectations about population growth. When the actual scenario (1995–2008) is implemented by updating exogenous variables or forcing inputs and introducing the potential (virtual) population increase, as derived from the plans, the model is able to predict the whole set of 1995–2008 observed data series with no further calibration against data. This might be considered as a sort of model validation with an independent dataset, which constitutes a remarkable contribution in the area of socio-ecological modelling.