Our model scenarios have demonstrated that the basic mechanism of hot spot formation is the same for both micro-topography models. In young, freshly infiltrated subsurface water, electron acceptors are abundant and anaerobic respiration can proceed as soon as the infiltrating water reaches zone 3 where oxygen resupply is assumed to be negligible (τresupply ≪ τdepletion). After depletion of oxygen, denitrification, iron(III) – and sulfate reduction are sequentially initiated for infiltrating water, triggered by the high availability of electron acceptors. Reductive hot spots are generated below infiltration areas located preferentially underneath hummock structures. Initially, below hummocks, infiltrating water passes the unsaturated zone (zone 1) where resupply of oxygen is assumed to occur instantly (τresupply ≫ τdepletion), here aerobic respiration is the only active process. As water infiltrates deeper (zone 2), resupply of atmospheric oxygen is assumed to significantly slow down (reduced diffusivity) and anaerobic processes are being initiated. Denitrification, iron(II) and sulfate reduction become dominant in the part of the saturated zone (zone 3) where the resupply of oxygen is cut off. On the other hand, upwelling zones, where older and already reduced groundwater rises into superficial layers, are characterized by inactivation of reduction processes. Here denitrification, iron(III) and sulfate reduction are inhibited because electron acceptors are not available. Oxidation processes however, are triggered for upwelling areas because here reduced water gets in contact with atmospheric oxygen, which is supplied to zone 2 by diffusion through the saturated pore space. Upwelling of subsurface water preferentially occurs below local depressions. Whether oxidation hot spots can be generated below a depression or not depends on the amount of surface water stored within the superficial depression. If a surface depression is filled with too much water (depth of surface ponding >0.25 m) diffusive penetration of atmospheric oxygen is inhibited and hence the formation of hot spots for oxidation processes is suppressed. In contrast, very pronounced hot spots for oxidation processes can be found below depressions with upwelling groundwater and low ponding depths. Here the saturated pore space is located within zone 2 where diffusion of atmospheric oxygen still exceeds depletion (as opposed to zone 3) and where oxygen can penetrate into shallow layers where it gets in contact with upwelling water carrying high concentrations of reduced species. How fast turnover of reduced species in oxidation hot spots occurs, depends on the availability of oxygen in zone 2, which is controlled by the amount of surface water being stored in the superficial depression. If surface ponding depths are very low (<0.05 m) availability of oxygen is assumed to be very high within zone 2 resulting in fast turnover of reduced species and very pronounced local oxidation hot spots. With increasing surface ponding (0.05 m–0.25 m) oxygen availability drops rapidly resulting in slower turnover rates and less pronounced oxidation hot spots. The availability of oxygen below depressions is therefore mainly controlled by the dynamics of surface water storage, which was found to be highly variable in space and time in wetlands with a hummocky topography, depending on the climatic boundary conditions [Frei et al., 2010]. During intensive rainfall events, surface storage and runoff generation in wetlands with shallow water table can be controlled by a dynamic fill and spill mechanism [Frei et al., 2010]. Depressions are filled with water due to rising groundwater levels during onset of rainfall. With lasting rainfall, isolated ponded depressions start to interconnect with each other building extended surface flow networks [Frei et al., 2010; Antoine et al., 2009]. These surface flow networks can efficiently drain large fractions of the wetland's surface. At times more than 80% of the generated stream discharge may originate from this type of surface flow [Frei et al., 2010]. During high water table conditions, fast diffusion of atmospheric oxygen into the subsurface system is limited to areas of high elevation (hummocks), which remain unsaturated at the surface. During water table recessions and decreasing surface ponding, diffusion of atmospheric oxygen, below depressions with lower surface ponding, becomes more effective in terms of increasing rates for resupply, which triggers oxidation processes for upwelling conditions. Generally field data on oxygen supply in wetlands, its coupling to water table dynamics and peat properties are scarce [Estop-Aragonés and Blodau, 2012], stressing the importance of virtual modeling studies. A special condition can develop during extended dry periods, where depressions become disconnected from the declining water table. Below these disconnected depressions hydraulic gradients may reverse, switching from upwelling to infiltrating conditions. In turn oxidation hot spots will diminish because resupply of reduced species from upwelling groundwater is disrupted. It is reasonable to assume that during droughts hot spot patterns will become less pronounced and may eventually vanish as the system gradually shifts toward a more homogenous distribution of process activities.
 In real wetland systems probably more than one mechanism will be responsible for the formation of biogeochemical hot spots [McClain et al., 2003] and a clear separation of the influence of one specific process is almost impossible under field conditions. The simulations presented here, however, demonstrate that heterogeneous process patterns in hummocky wetlands can be explained by the complex re-distribution of redox sensitive solutes in space as being controlled by micro-topography induced, subsurface transport processes and alternating biogeochemical boundary conditions. Furthermore, the presented concept shows that biogeochemical hot spots can be generated without reference to material heterogeneities which often are hardly observable in horizontally relatively homogenous peat soils [Morris and Waddington, 2011; Holden and Burt, 2003; Reeve et al., 2001, 2006; Clymo, 1984]. Of course the presented concept neglects important aspects of real field conditions. Effects of the wetlands vegetation like e.g., root water uptake and its influence on subsurface flow or the special biogeochemical conditions within the rhizosphere [Crow and Wieder, 2005; Knorr et al., 2008; Wachinger et al., 2000] are not considered as well as the potential effects on dispersion for the availability of electron acceptors and donors. Hydrodynamic dispersion may cause a smearing effect where the boundaries between hot spots and surrounding areas are not as sharp and clearly expressed as in an advectively dominated system, because solutes are also re-distributed along concentration gradients (diffusion) and transversally and longitudinally along the advective flow directions (dispersion). The biogeochemical simulations were performed using 5-day time steps, which was necessary because of computational constraints during the flow modeling (e.g., memory overflow, storage limitations). However, it is known that hydrological events at time scales of hours (e.g., single rainstorm events) can influence the biogeochemical processes within wetlands, as e.g., demonstrated for pulses of N2O emission [Goldberg et al., 2010] or high instantaneous CO2 production [Deppe et al., 2010] after wetting. Dynamics at these time scales, however, were not the main focus of this work and at this point cannot be fully accounted for in the present modeling approach because of computational limitations. Further it is known that organic carbon in wetlands typically consists of a fraction of labile components that can be easily utilized by microorganisms (mostly within shallow layers) and more recalcitrant components (more abundant in deeper layers) [Yavitt and Lang, 1990; Reiche et al., 2010; Moore et al., 2007]. Labile organic carbon is not uniformly available as is assumed in our approach. However, there are two main reasons why we think that our assumption of unlimited carbon supply is nonetheless reasonable. First, labile organic carbon availability is higher in shallow peat layers, in which most of the modeled processes occur, mostly due to inputs from the vegetation and high fermentation activity in the rhizosphere [Knorr et al., 2008; Wachinger et al., 2000; Reiche et al., 2010]. Second, we did not include methanogenesis, for which the supply of electron donors will be the key control, as the ubiquitous CO2 may serve as electron acceptor [Achtnich et al., 1995]. Field observations suggested that if alternative electron acceptors were present, the respective process proceeded, while under methanogenic conditions, respiratory activity slowed down and partly ceased [Beer and Blodau, 2007; Knorr et al., 2009]. Nevertheless, the process rate constant, in this case, depends on the quality of organic matter used and is not universal but substrate specific. The application of the Redfield ratio to simulate release of organic bound nitrogen due to decomposition of organic material in terrestrial ecosystems was probably a weak model assumption. Recent literature reported that C:N:P ratios in terrestrial ecosystems vary depending on vegetation types, but on the global scale average at about 186:13:1 for soil biomass and 60:7:1 for soil microbial biomass [Cleveland and Liptzin, 2007]. In our biogeochemical model we assumed that the majority of organic carbon available to microbes originates from vegetation and fermented plant material processed by microorganisms. The Redfield ratio is, however, narrower than the global average observed for soil biomass (106:16:1 compared to 186:13:1) and nitrogen release would be overestimated by our model. That means that the concentrations of ammonia, rates of nitrification and thus also nitrate pools available for denitrification may also be overestimated. Nevertheless, this should translate into slightly longer phases of nitrification or subsequent denitrification only, thus not fundamentally altering spatial patterns of the model output.