Spatial prediction of soil nutrient in a hilly area using artificial neural network model combined with kriging

Spatially distributed modeling of soil organic carbon across China with improved accuracy

“…The performance of the HASM and RBFNN can also be contributed to improvements of prediction accuracy. The RBFNN model has been proven more effective than the MLR at explaining spatial variations of soil properties in response to changes in site environmental factors due to its weaker smoothing effects and greater explanatory capacity for the spatial variability of soil properties based on the same auxiliary environmental variables [ Li et al ., ]. Previous studies also showed that MLR was not the optimum approach due to the complex relationships between soil and environmental factors [ Li , ; Umali et al ., ; Li et al ., ].…”

“…Furthermore, many of the relationships between soils and environmental factors are not linear [ McBratney et al ., ]. Although linear regression models are successfully used to map SOC, the artificial neural network (ANN) approach was recently found to perform better than MLR owing to its capacity to auto‐analyze nonlinear relationships between multisource inputs by self‐learning [ Li et al ., ]. In addition, such relationships are often spatially nonstationarity in that the relationships between environmental variables and soils vary across space, and it is thus unlikely that a single model can be developed to be applicable to all subareas or units in regional studies [ Mishra et al ., ; Li et al ., ].…”

“…This type of method is based on the premise that the deterministic component of the targeted soil variable caused by correlated environmental factors can be explained by scorpan models while the spatially varying but dependent component can be described by the prediction residuals of the scorpan model and captured by the purely spatial approaches. This idea has formed the methodological framework for the spatially distributed modeling of soil properties in recent studies [ Shi et al ., ; Zhang et al ., ; Li et al ., ; Liu et al ., ].…”

“…Among the three types of methods used for spatially distributed modeling in soil science, methods that introduce correlated environmental factors as auxiliary variables, and especially the third type that considers both components of the targeted soil variable, have been proved to be effective for improving the prediction accuracy and have received increasing attentions, because the spatial distribution of soil is greatly affected by environmental factors and it is difficult to accurately obtain soil property distributions in the absence of obvious environment features [ Shi et al ., ; Mishra et al ., ; Zhang et al ., ; Li et al ., ; Liu et al ., ].…”

Multi-Sensor Prediction of Stand Volume by a Hybrid Model of Support Vector Machine for Regression Kriging

“…Those two-step approaches both consider the spatial heterogeneity conveyed by remote sensing predictors and autocorrelation of neighboring observed data [24,25]. Those approaches, especially machine learning combined ordinary kriging of residuals such as artificial neural network kriging (ANNK) and random forest kriging (RFK), have yielded accurately spatial predictions [26,27]. However, support vector machine for regression kriging (SVRK) modeling for mapping forest volume has rarely been tested and reported.…”