Channel initiation in a low-permeability landscape

Where shallow landslides can be ruled out as the cause of channel initiation, the area-threshold concept is commonly invoked to explain the occurrence of channel heads. This concept implies that a certain depth of overland flow must be reached before incision can occur, and that this depth depends on the contributing area upslope of a point of interest.
We investigate the applicability of this concept to Barro Colorado Island (BCI), a tropical rainforest landscape known for a low soil permeability by forest standards and a pronounced wet season with high-intensity rainfall events. We define “low permeability” by the frequency with which selected rainfall intensities exceed soil permeability, expressed as saturated hydraulic conductivity (Ksat), at shallow depths. Thus, while the median of maximum 30-min rainfall intensities of 5 mm/h exceeds on (spatial) average Ksat at a depth of 0.15 m, in some places the same metric does not exceed Ksat until a depth of 0.5 m. BCI is also know for the frequent occurrence of overland flow of the “wide-spread” type, a vague concept occasionally invoked but hitherto undefined. Hence, the area-threshold approach appears applicable.
We conceptualize channel heads as point processes on flow networks. The covariates slope, curvature, flow accumulation and flow convergence were chosen as candidate explanatory variables that control the occurrence of these points. Using stepwise regression as a model-selection technique, we derived a log-linear model involving the covariates flow accumulation and flow convergence and their interaction term. These results suggest that the accumulation of discharge in the direct vicinity of the channel head and on the upslope contributing area dominate channel formation in this environment and that the probability of a pixel being a channel head decreases with increasing flow accumulation or flow convergence. Thus, the discharge amount was a main driver of channel head formation. While this seems to vindicate the area-threshold concept, the difficulty of predicting channel head locations warrant a look at the subsurface, besides considering technical issues such as DEM resolution or the precision of GPS readings under a thick forest canopy. The coincidence of some channel heads with return flow suggests that surface-based metrics alone do not explain their spatial distribution.