Click on “Download PDF” for the PDF version or on the title for the HTML version.


If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.

Riparian Sediment Delivery ratio: Stiff Diagrams and Artificial Neural Networks

Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.org

Citation:  21st Century Watershed Technology: Improving Water Quality and Environment Conference Proceedings, 21-24 February 2010, Universidad EARTH, Costa Rica  701P0210cd.(doi:10.13031/2013.29449)
Authors:   Ernest William Tollner, Herbert Ssegane, Steven C McCutcheon

A stiff diagram with a logistic function was tested to empirically integrate the effects of seven riparian buffer parameters in calculating the sediment delivery ratio as initially applied by the U.S. Forest Service in 1980 silvicultural nonpoint source pollution guidance. The use of artificial neural networks to heuristically capture the stochastic nature of sediment transport and deposition was also tested and compared using sediment yield measured on 30 small experimental buffers located in the U.S., Canada, and Germany. No other comparison existed before this study of the application of the stiff diagram to calculate sediment delivery ratios with ratios calculated from actual measurements. An average of the sediment delivery ratios from the measurements on the 30 buffers is better than estimates obtained from the stiff diagram used to composite the effect of the seven characteristicsrunoff, soil texture, ground cover, slope shape, delivery distance, roughness, and slope gradient. Although only three experimental buffers were forested, these tests establish that the stiff diagram poorly estimates sediment delivery, leaving the original U.S. Forest Service hypothesis unproven. The estimated delivery ratio was over an order of magnitude too small for all three forested buffers compared to sediment delivery ratios calculated from field observations. Specification of the dimensionless stiff diagram area for an artificial neural network established that the logistic function selected by the Forest Service to relate stiff diagram area to delivery ratio does not adequately represent the complex nonlinearity of sediment delivery. Specifying the same seven buffer parameters provided adequate estimates because neural networks inherently account for nonlinearities using more than just the logistic or the other functions calibrated and tested. The overall single best trained network structure with an optimum of 10 hidden neurons with just five buffer characteristics achieved even better performance with a dimensionless mean square error of 0.001, a coefficient of determination of 0.98 and a Nash-Sutcliffe efficiency of 0.98. Although artificial neural networks tend to require large numbers of data, this investigation showed that with appropriately independent variables that explain larger portions of the variability, the minimum recommended number of data can be used to estimate sediment delivery ratios. Thus, artificial neural networks should be trained to estimate sediment delivery ratios instead of using the stiff diagram, which is limited to one function to describe nonlinearities. The paper has been accepted to appear in Transactions of ASABE.

(Download PDF)    (Export to EndNotes)