Incorporating uncertainty in gully erosion calculations using the random forests modelling approach

Abstract

AbstractGully erosion represents one of the greatest threats to agricultural watersheds around the world and this is exacerbated with continual changes to land practices and changing climatic condition at a temporal scale. Catchment based models which include gully erosion have received increased attention in recent years due to the impact of gully erosion on downstream water quality. These types of models take gully erosion calculated via a deterministic model at a particular spatial location, along with other measures of erosion to provide an end‐of‐catchment load. Although seen as a major contributor to erosion in this type of modelling framework, the error surrounding the gully erosion calculation is ignored in the sediment model. We propose a methodology that investigates the error in the gully erosion model. The methodology uses Random Forests to predict gully density and its associated prediction error across a catchment using environmental variables, which are then incorporated into the gully erosion model. We demonstrate the approach using the Burdekin catchment in Queensland, Australia, where only a very small proportion of gullies (∼0.88%) have been mapped. Our results cast doubt in the predictive ability of models of sediment transportation that use gully erosion where the error is estimated high. Copyright © 2009 John Wiley & Sons, Ltd.