Abstract

Considerable field research into the effect of forestry and agroforestry operations on soil erosion has taken place in South-East Asia over the last 40 years (e.g., Bruijnzeel, 1990). Recently developed, physics-based models of soil erosion and resultant sediment flow - 'SSflux' (e.g., Favis-Mortlock et al., 1996) 'promise' an ability to simulate the site-specific relationships (between rainfall characteristics, physical properties and forestry operations) contained within these field studies. Clearly, a capability to separate the effects of specific forestry operations from other site factors would aid in the identification (and demonstration) of 'best' or 'most sustainable' forestry practices (see e.g., Bruijnzeel, 1982; Abdul Rahim Nik et al., 1997). Model calibration of a single-output-variable combined with limited catchment data on soil-geological properties does, however, mean that such physics-based models can give highly uncertain predictions; though this is rarely quantified in rigorous uncertainty analyses (Beven, 1996). As a result, such approaches can sometimes give a false sense of understanding of the processes or likely scenarios.This presentation demonstrates how the mobilisation and transport of suspended-sediments within multi-scale catchments can be modelled without the use of complex (high order) and hence uncertain, model formulations. The modelling uses rainfall, waterflow and sediment data collected within the Baru Experimental Catchment of the Ulu Segama Forest Reserve - which is part of Yayasan Sabah timber concession in Sabah, Malaysian Borneo. Some of the localised erosional-landforms within the catchment relate to selective forestry operations that took place some five years prior to the collection of the model data-set. The results of a Data-Based-Mechanistic (DBM) modelling approach, which combines physically-based understanding with model-structure identification based on transfer-functions and objective statistical inference (Young and Beven, 1994), are presented here, though conceptual models (e.g., TOPMODEL) have also been applied to the same data-series (e.g., Chappell et al., 1998). Model realisations of the 'rainfall to riverflow' and 'rainfall to SSflux' behaviour of five of the nested, contributory areas of the Baru Experimental Catchment indicate (i) that these systems exhibit relatively low-order behaviour that can be modelled with a few (and hence relatively certain) model parameters (Chappell et al., in press), (ii) that parametric differences between the different contributory areas can be (tentatively) explained in process-based terms, and (iii) that the exploratory modelling approach adopted is capable of identifying new processes and properties that should be measured to give better explanation. With this initial success - more complex model formulations that retain a similar degree of statistical credibility are being developed. The importance of road-landslides to the understanding and successful modelling of even low-order erosional systems was, however, very clear.