Long time series of environmental variables are reflecting the dynamics and functioning of ecosystems. Here, we investigate data from a long-term monitoring site in Germany, the Bramke valley in the Harz mountains, where time series of ion concentrations in stream water are obtained since the 1970ies at four measurement locations from three small adjacent forested catchments. Since for (only) one of the catchments daily runoff rates are also available, we invent a method to generate time series of nutrient output from the catchments. Both concentrations and outputs show a number of remarkable long-term changes, including ones not obviously related to changes in atmospheric deposition, management or properties of the forest stands. For the analysis of the Bramke data, we investigate Horizontal Visibility Graphs (HVGs), a recently developed method to construct networks based on time series. Values (the nodes of the network) of the time series are linked to each other if there is no value higher between them. The network properties, such as the degree and distance distributions, reflect the nonlinear dynamics of the time series. For certain classes of stochastic processes and for periodic time series, analytic results can be obtained for some network properties. HVGs have the potential to discern between deterministic-chaotic and correlated-stochastic time series. We classify the Bramke series according to their stochastic nature, with a focus on inter-catchment comparison on one hand, on different nutrients for one catchment on the other, and conclude on possible reasons for the observed changes and their ecological interpretation.