Submitted to: Catena
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 9/20/2002
Publication Date: N/A
Citation: N/A Interpretive Summary: Conservationists and land managers use soil erosion models to assess the erosion that is occurring on the land and to help to make the best possible decisions about land management to control erosion. Models of soil erosion have historically been of two types, statistical and process-based. Statistical models are based on the use of statistical analysis of erosion data. Our most commonly used model of soil erosion in the United States is the Universal Soil Loss Equation (USLE), which is a statistically based model of erosion. Process-based models are based on equations of physics, chemistry, and biology that describe the myriad of individual processes that affect all aspects of erosion. These models tend to be very sophisticated and difficult for all but the experts to use. This study undertakes to introduce a third type of model based on Neural Networks. Neural Networks are logical computer models that mimic to an extent the workings of neural networks in the biological brain. In this study we demonstrate the possibilities of using these networks to predict erosion. The initial results indicate that the predictive capabilities of these new erosion model types may rival or exceed that of both process-based and statistical models of erosion. The result of this work is that we may be able to develop better and easier to use tools for helping land managers make decisions in managing their land to reduce erosion. This implies ultimately a more productive agricultural capability for future generations of Americans.
Technical Abstract: Neural networks may provide a convenient and potentially powerful method for developing prediction tools for soil erosion. The purpose of this study was to investigate the applicability of using neural networks to predict soil loss from natural runoff plots. Data from 2879 erosion events from eight locations in the United States were used. Neural networks were developed for data from each individual site using only eight input parameters, and for the complete data set using ten input parameters. Results indicated that the neural networks performed generally better than the WEPP model in predicting both event runoff volumes and soil loss amounts. Linear correlation coefficients (r) for the resulting predictions from the networks versus measured values were generally in the range of 0.7 to 0.9. Networks that predicted both runoff and soil loss together did not perform better than those that predicted each variable individually. The type of transfer function and the number of neurons used within the neural network structure did not make a difference to the quality of the results. Soil loss was somewhat better predicted when values were processed using a natural logarithm transformation prior to network development. The results of this study suggest a significant potential for using neural networks to predict soil erosion by water.