Abstract:
Chemical element concentrations and fluxes in the hydrological cycle of a mountain rain forest in southern Ecuador are influenced by annual seasonality, long term trends, inter-annual ENSO cycles and other environmental factors. Some knowledge about this processes was collected in previous studies. But little is known about how much of the variation in the data can be explained by those processes. Goal of this thesis is therefore to build linear models that explain the variation in element concentrations and fluxes in the incident precipitation and in the troughfall of an small forested catchment in southern Ecuador. The linear models contain seasonal terms, trend terms, ENSO temerature anomalie terms and environmental variable terms. For each analysed element four linear model were build to explain variation in concentrations and
fluxes in incident precipitation and troughfall. The models contained all terms at the beginning and were then optimized to a model with only the significant terms for each element in each flux and concentration.
By analysing a time series from 1998 to 2010 with monthly means of element concentrations of weekly measurements of troughfall and incident precipitation, and their resulting fluxes, the following hypothesis are tested. Namely seasonal terms are significantly explaining the variation in the concentrations and fluxes, longterm trends are explaining the variation, ENSO related temperature anomalies are explaining the variation and other environmental factors are explaining the variation. The findings showed that in 45 % of the models seasonality is significantly contributing to the explaining of the variation. A significant trend terms is part of 30% of the models and a significant ENSO term in 18%. The range of percentage of significant environmental variables starts with 16% for wind direction and 18% for flower or fruiting phenology. goes to 25 % and 31 % for fire activity and heavy rain activity respectively and finally goes to 57% for conductivity. To mention is that in this case conductivity is present in 90% of the conductivity
models. The resulting R squares showed that the best models are the troughfall models. The best model
here explains almost 80% of the variation, the median is around 50% of explained variation and the worst model explains 27 % of the variation. In the incident precipitation concentration and in the troughfall and incident precipitation fluxes the best models are between 30 and 40 % of explained variation, the median is about 20% for incident precipitation and between 10 and 15 % for the fluxes and the lowest values are about 5 %. The model quality test shows that the not crucial criteria of normal distribution ot the model residuals is violated in some models. The crucial temporal independence criteria is most likely violated in few models and in one model it is clearly violated. All in all the thesis could show that seasonality, trend, ENSO related temperature anomalies and the environmental variables fire activity, conductivity, wind
direction, heavy rain, and flower and fruiting phenology are in various combinations contributing significantly to the explaining of the variation in concentration and fluxes of incident precipitation
and troughfall. The models are strong in explaining the variation in cases like potassium troughfall concentration, where rainfall seasonality leads to big concentration variation, while in other cases, like magnesium incident precipitation concentrations, where little variation occurs and factors that are not included as model terms lead to clear patterns in the concentrations, the model can explain almost no variation.
