H14A-01 INVITED
Rainfall Redistribution in a Tropical Forest: Spatial Patterns and Temporal Persistence.
The quest for hidden spatial patterns has left few near-surface processes untouched, and throughfall is not
one of them. Indeed, the spatial patterns of several hydrological and biogeochemical processes at the forest
floor have been linked to throughfall patterns.
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 patterns of throughfall. We employed
a mixed design- and model-based and event-based sampling strategy with an extent of one hectare and 220
throughfall collectors in a tropical rainforest in an attempt to separate spatial patterns from spatial illusions.
For most of 56 sampled events, throughfall frequency distributions called for robust variogram estimation
techniques. In the presence of outliers, the classical, non-robust variogram estimator not only overestimates
the sill variance but also induces artificial autocorrelation structures. Throughfall patterns typically displayed
no or only weak autocorrelations, except in areas with little understory where measurements correlated over
10 – 15 m. The temporal persistence of throughfall was high, that is, measurements at individual sampling
points were significantly correlated over consecutive wet seasons. Interestingly, seasonality, and hence
deciduousness, had a negligible influence, whereas the magnitude of an event strongly determined the
temporal stability of throughfall. Our results demonstrate that vegetation characteristics strongly influence
spatial patterns of throughfall, whereas the temporal persistence of throughfall largely depends on event
size.
H14A-02
Sampling Strategies for Throughfall Monitoring: a Simulation Study to Capture Space- Time Heterogeneity
For decades, scientists have searched for appropriate sampling strategies to quantify throughfall magnitudes
in forest ecosystems; accordingly, a number of studies focused on some aspects of this issue, for example
the relative merits of different sampling designs or an increased sample support. Due to the empirical
character of those investigations, no overall assessment has yet been possible. The availability of extensive
spatio-temporal datasets of throughfall in a tropical rainforest enables us to perform a simulation study, which
aims at identifying the most appropriate sampling strategy to quantify mean throughfall and subsequently,
interception loss. The assessment includes the test of different sampling strategies, for instance simple
random sampling, stratified simple random sampling and cluster sampling; the sample size requirements to
achieve a certain accuracy of the estimated sample mean; and the relative merits of an increased sample
support, i.e. trough-type versus funnel-type collectors. Since throughfall data exhibit extreme values, we
removed them by a statistical criterion prior to the geostatistical analysis, but reinstalled them before
sampling from the simulated fields.
Preliminary results for datasets without extreme values show that for funnel-type collectors a sample size
between 50 and 150 is needed to go below a relative error of the mean of 10% regardless of the sampling
design. Increasing the sample support reduces the sample size, which has to be contrasted with the
additional effort of employing trough-type collectors. We will proceed with the calculations for the
contaminated datasets, and assess the final results with respect to the estimation of interception loss.
H14A-03 INVITED
Consequences of Spatiotemporal Redistribution of Precipitation by Vegetation for Hillslope and Runoff Processes
The effects of vegetation on streamflow generation are currently best understood in terms of how evaporation from canopy interception and transpiration affect watershed-scale responses at timescales of individual storms or longer. Examining processes at finer temporal and spatial scales suggests that spatiotemporal redistribution of precipitation by forest canopies within storm events may have important consequences for transient pore pressures and runoff generation. Temporal variation in rainfall intensity is reduced in throughfall because canopies act as reservoirs that temporarily increase storage during short bursts of high-intensity rainfall; effective hydraulic mean residence times may be 20 minutes or longer in dense canopies. Modeling indicates this intensity smoothing reduces the development of high soil pore pressures responsible for initiation of shallow landslides and has a small effect on peak storm discharge. Associated with temporary storage in the canopy is lateral redistribution that results in spatial patterns in throughfall that persist over time and cause spatial variation in soil moisture. Modeling indicates that throughfall or stemflow infiltrating disproportionately into zones of high soil moisture causes a positive feedback that speeds runoff response and reduces total soil water storage. Few watershed models incorporate these spatiotemporal nonlinearities directly, though more work is needed to understand how these processes influence runoff at the watershed scale and across a range of ecosystem structures.
H14A-04 INVITED
Spatial Variability of Water Repellency After Forest Fires in the Rocky Mountain Region
Increases in runoff and erosion after forest wildfires are often attributed to the development of water repellent soils. The potential for increased overland flow depends on the spatial contiguity of the water repellency as well as its overall strength, but there is limited information on the spatial variability of post-fire soil water repellency. We conducted spatially intensive water repellency measurements in 225 m2 and 1m2 plots in forested areas of Montana and Colorado burned at moderate to high severity, and in unburned control plots. Both the burned and unburned 225 m2 plots contained 10 to 23 water repellent patches in which the repellency was strongest at the surface and declined rapidly with depth. The water repellent patches were closer together and up to 3 times larger in the burned plots, so that 19 to 76 percent of the burned plots were water repellent compared to just 11 percent of the unburned plots. In five of the six burned plots the patches were not laterally connected, suggesting that in most cases Hortonian overland flow generated from water repellent patches will infiltrate near its point of origin. The 1m2 plots were smaller than most of the water repellent patches, so they did not capture the spatial characteristics of soil water repellency. Due to the patchiness of soil water repellency at the 100 to 101 meter scale, overland flow measurements in small (~1m2) plots may overestimate the magnitude and variability of runoff from burned catchments. More broadly, the patchiness of the water repellency seen in our plots suggests that the importance of water repellency as a cause of increased runoff after fires may have been overestimated, and that post-fire increases in runoff and erosion may be due to other factors such as surface sealing and loss of ground cover.
H14A-05
The Seasonal Evolution of Hillslope-Channel Connectivity in a Flashy Forest Catchment
Fast, near-surface flowpaths are a trait of soilscapes with high rainfalls amounts and intensities and with an abrupt decrease of permeability with depth. The Lutz Creek catchment in Panama is a prime example of such an environment, in which overland flow generated at the soil surface proper and return flow (RF) connect hillslopes with its stream channels. Return flow- and saturation overland flow (SOF)-mediated connectivity is, however, not arranged randomly, but follows a topography-controlled pattern. This pattern, in turn, determines a temporal pattern of connectivity: RF-prone parts of the catchment connect hillslopes and channels earlier in the wet season than do SOF-prone parts, and regardless of season, the former establish connectivity in response to low-intensity rainfall, whereas the latter require high-intensity rain, all else being equal. Increasing catchment wetness towards the end of the wet season, however, blurs the distinction between RF- and SOF-controlled parts of the catchment, and this is when maximal connectivity is achieved.
H14A-06
A Stochastic Model of Surface Runoff and Pollutant Connectivity Between Hillslopes and Streams
The overland flow pathway between hillslopes and streams remains a critical, yet poorly represented transport pathway linking hillslope constituent generation and catchment exports via the stream network. The poor representation of this runoff and pollutant pathway largely results from the mismatch in scales between the pervasive point-scale representations of infiltration (eg Green and Ampt) within hydrologic and erosion models, and the reality of often extreme within-catchment spatial variability in the critical hydrologic properties of the system. Attempted solutions to date fall into three categories; 1) simple empirical sediment delivery ratios, 2) numerical simulations of spatially variable fields of soil and rainfall parameters, and 3) gross simplifications of the variability and arrangement of these parameters so as to admit analytical solutions to the runoff delivery problem. A key aspect of this problem is that in real catchments, transient runoff may be generated, infiltrating further downslope at a point where soil hydraulic conductivity is greater. This paper proposes an alternative way (entitled Stochastic Runoff Connectivity, SRC) to model and quantify the delivery of overland flow to the stream boundary under these spatially variable hillslope conditions. The approach utilizes queuing system theory and a stochastic storage process model whereby the 'connected area', and the volume of water reaching the stream via this pathway, can be described using probability density functions determined (analytically) from the asymptotic states of the stochastic process. The connected area in this instance is defined as an infiltration-excess equivalent to the well know variable source area concept, and provides an important input to raindrop-energy based hillslope erosion models. The validity of the SRC model is tested against field data and numerical simulations.
H14A-07
On the Representativeness of Plots for Scaling Ecohydrologic Processes in Forests
Scaling hydrological and ecological processes to stands or larger regions still relies on observational data collected in plots. We tested the validity of this approach with respect to reference canopy stomatal conductance, Gsref, leaf area index, L, canopy gap fraction, Fgap, and canopy transpiration, Ec, at plot to stand scales. We also tested whether or not plot representativeness was dependent on environmental drivers. The study was conducted in a 1.7 ha aspen-dominated stand in northern Wisconsin. We measured the size and relative location of every aspen tree (total of 756 individuals), and in a subset of 106 of these trees we collected sap flow data. We also measured gas exchange and developed site-specific allometry for crown cross-sectional area and leaf area. Different scales of observation were made using pseudo-plots spanning 2m to 12m radii with randomly sampled plot centers. We tested the reliability of parameters, Gsref and L, at each scale using an ecosystem model. Ec and L both declined non-linearly with increasing plot size, resulting in only small variation in transpiration per unit leaf area across the plot sizes. Log-log variance-area plots of Ec binned by vapor pressure deficit, D, were linear with no significant change in slope among bins. Gsref generally declined and total reference canopy conductance (= Gsref * L) declined monotonically with plot area, respectively. Simulations using Gsref and L showed increasing overestimation of predicted Ec with increasing plot size, with up to 50 percent error at low D and 30 percent error at moderate to high D. These errors were explained by accounting for Fgap, derived from allometry. The results implicate three spatial scale-dependent, time-independent canopy parameters (Gsref, L, Fgap) as critical for scaling from plot to stand level. General implications for scaling processes in forest ecosystems will be discussed.