Prediction of Soil Organic Carbon across Different Land-use Patterns: A Neural Network Approach

ecosystem type and soil taxonomy has been compared, and the taxonomic approach appeared to be more meanMathematical modeling has widely been used to predict soil organic ingful in creating a real picture of spatial distribution carbon (SOC). However, there are characteristics of the models such as over simplification, ignorance of complex nonlinear interactions of SOC. Attempts have been made to estimate global etc., which limit their use in accurately assessing the distribution of the SOC using the pedon database and extrapolating them C across the landscapes. Artificial neural network (ANN) modeling to soil units of the world soil map (Bohn, 1976, 1982; approach that provides a tool to solve complex problems related to Batjes, 1996; Buringh, 1984; Kimble et al., 1990). The larger data sets was therefore used here to predict SOC contents pedon database of the USDA Soil Conservation Service across different land use patterns in a study conducted in Sri Lanka. and related organizations has been used to estimate the Selection of variables was made using a priori knowledge of the regional distribution of organic C in the USA (Kern, relationships between the variables. Thus, soils of the sites were sam1994). However, previous studies indicated that there pled and analyzed for organic C by internal heat of dilution (Ci) and are uncertainties associated with such SOC estimates external heat of dilution (Ce), and the results were presented as grams and often related to variations in soil map scales and per kilogram (g kg 1). In addition, some landscape attributes and environmental parameters of the sites were also collected. The preseries. As a whole the uncertainties associated with meadictive performance of ANN was compared with multi-linear regressuring and detecting changes in soil C pools remain sion (MLR) models. The best ANN model predicted the measured high, both at individual sites and extrapolating site-level Ci content with R2 of 0.92. However, comparison of the two types of data to regional, national, or global scales (Vance, 2003). models indicated less bias and high accuracy of the ANN compared Accurate and precise approaches yet to be available for with MLR in predicting Ci, but the reverse for Ce. In order to better assessing the effect of management practices and land predict Ce, it is recommended to use other architectures of neural use change on the soil C for the purpose of incorporation networks and training algorithms for improving predictive accuracy. of this important pool into future C accounting systems. The predictive capability of the ANN developed with easily available The Kyoto Protocol, for instance, limits reporting of C climatic and terrain data are of importance in predicting SOC with sequestration activities to “measurable and verifiable” minimum cost, labor, and time. pools (Vance, 2003). Mathematical modeling has been used to predict soil C evolution (Jenkinson and Rayner, 1977; Parton et al., T soil system strongly influences the structure and 1988; Pastor and Post, 1985; Smith, 1982). These models function of ecosystems and acts as a buffer to global possess the ability to simulate the complex processes climatic change. Therefore, understanding of the proin the formation and degradation of organic C and of cesses in the soil is crucial in the context of the ecosystem describing the relationship between a numbers of soil management. The SOC is a vital component, since it properties that control soil C evolution. In these models plays a key role in soil fertility and in hydrology and empirical, stochastic, and mechanistic equations have acts as a sink or source of terrestrial C, which affects been used to describe the simultaneous interactions of the concentration of atmospheric CO2. Soil information soil properties with SOC. These models rely on the is important in modeling ecological processes, vegetaavailable SOC data for predicting evolution in a given tion dynamics, and forecasting agricultural potentials area, and certain models incorporate a limited number (Adams et al., 1990; Levine et al., 1996; Dixon et al., of SOC data points. In other cases, some of the models 1994). use interpolated or extrapolated SOC values. As a reThe SOC estimates with certain degrees of uncertaintsult, the model predicts SOC evolution poorly for a ies are available for regional to global scales. These given area. Further, for a satisfactory prediction of SOC estimates have been made either based on the existing evolution, it is necessary to increase the number of SOC soil databases or modeling techniques. Post et al. (1992) data points rather than interpolation or extrapolation used global soil data up to 1-m depth across Holdridge of existing few SOC data points. The use of statistical or life zones. The aggregation of soil data according to the empirical models may also hinder the real relationships between the SOC and soil properties because strict staS. Somaratne, Dep. of Botany, The Open Univ. of Sri Lanka, Nawala, tistical sampling designs are not generally used in soil Nugegoda, Sri Lanka; G. Seneviratne, Institute of Fundamental Studies, Hantana Road, Kandy, Sri Lanka; U. Coomaraswamy, Vice Chansampling. cellor’s Office, The Open Univ. of Sri Lanka, Nawala, Nugegoda, Sri Lanka. Received 21 Nov. 2003. *Corresponding author (gaminis@ Abbreviations: AIC, Akaike’s information criterion; ANN, artificial ifs.ac.lk). neural network; Ce, soil organic carbon determined by application of external heat; CEC, cation exchange capacity; Ci, SOC determined Published in Soil Sci. Soc. Am. J. 69:1580–1589 (2005). Pedology by internal heat of dilution; LSD, least significant difference; MLR, multi-linear regression; MR, mean residuals; RMSE, root mean sums doi:10.2136/sssaj2003.0293 © Soil Science Society of America of squared error; RMSR, root mean square of residuals; SOC, soil organic carbon; TA, transformed aspect. 677 S. Segoe Rd., Madison, WI 53711 USA 1580 Published online August 25, 2005