What is the most appropriate sampling scheme to estimate event-based average throughfall? A satisfactory answer to this seemingly simple question has yet to be found, a failure which we attribute to previous efforts' dependence on empirical studies. Here we try to answer this question by simulating stochastic throughfall fields based on parameters for statistical models of large monitoring data sets. We subsequently sampled these fields with different sampling designs and variable sample supports. We evaluated the performance of a particular sampling scheme with respect to the uncertainty of possible estimated means of throughfall volumes. Even for a relative error limit of 20%, an impractically large number of small, funnel-type collectors would be required to estimate mean throughfall, particularly for small events. While stratification of the target area is not superior to simple random sampling, cluster random sampling involves the risk of being less efficient. A larger sample support, e.g., the use of trough-type collectors, considerably reduces the necessary sample sizes and eliminates the sensitivity of the mean to outliers. Since the gain in time associated with the manual handling of troughs versus funnels depends on the local precipitation regime, the employment of automatically recording clusters of long troughs emerges as the most promising sampling scheme. Even so, a relative error of less than 5% appears out of reach for throughfall under heterogeneous canopies. We therefore suspect a considerable uncertainty of input parameters for interception models derived from measured throughfall, in particular, for those requiring data of small throughfall events.

The investigation of throughfall patterns has received considerable interest over the last decades. And yet, the geographical bias of pertinent previous studies and their methodologies and approaches to data analysis cast a doubt on the general validity of claims regarding spatial and temporal patterns of throughfall. We employed 220 collectors in a 1-ha plot of semideciduous tropical rain forest in Panama and sampled throughfall during a period of 14 months. Our analysis of spatial patterns is based on 60 data sets, whereas the temporal analysis comprises 91 events. Both data sets show skewed frequency distributions. When skewness arises from large outliers, the classical, nonrobust variogram estimator overestimates the sill variance and, in some cases, even induces spurious autocorrelation structures. In these situations, robust variogram estimation techniques offer a solution. Throughfall in our plot typically displayed no or only weak spatial autocorrelations. In contrast, temporal correlations were strong, that is, wet and dry locations persisted over consecutive wet seasons. Interestingly, seasonality and hence deciduousness had no influence on spatial and temporal patterns. We argue that if throughfall patterns are to have any explanatory power with respect to patterns of near-surface processes, data analytical artifacts must be ruled out lest spurious correlation be confounded with causality; furthermore, temporal stability over the domain of interest is essential.

Changes in soil hydraulic properties following ecosystem disturbances can become relevant for regional water cycles depending on the prevailing rainfall regime. In a tropical montane rainforest ecosystem in southern Ecuador, plot- scale investigations revealed that man-made disturbances were accompanied by a decrease in mean saturated hydraulic conductivity (Ks), whereas mean Ks of two different aged landslides was undistinguishable from the reference forest. Ks spatial structure weakened after disturbances in the topsoil. We used this spatial-temporal information combined with local rain intensities to assess the probability of impermeable soil layers under undisturbed, disturbed, and regenerating land-cover types. We furthermore compared the Ecuadorian man-made disturbance cycle with a similar land-use sequence in a tropical lowland rainforest region in Brazil. The studied montane rainforest is characterized by prevailing vertical flowpaths in the topsoil, whereas larger rainstorms in the study area potentially result in impermeable layers below 20 cm depth. In spite of the low frequency of such higher-intensity events, they transport a high portion of the annual runoff and may therefore significant for the regional water cycle. Hydrological flowpaths under two studied landslides are similar to the natural forest except for a somewhat higher probability of impermeable layer formation in the topsoil of the 2-year-old landslide. In contrast, human disturbances likely affect near-surface hydrology. Under a pasture and a young fallow, impermeable layers potentially develop already in the topsoil for larger rain events. A 10-year-old fallow indicates regeneration towards the original vertical flowpaths, though the land-use signal was still detectable. The consequences of land-cover change on near-surface hydrological behaviour are of similar magnitude in the tropical montane and the lowland rainforest region. This similarity can be explained by a more pronounced drop of soil permeability after pasture establishment in the montane rainforest region in spite of the prevailing much lower rain intensities.