Using artificial neural network models to produce soil organic carbon content distribution maps across landscapes

Abstract

network models to produce soil organic carbon content distribution maps across landscapes. Can. J. Soil Sci. 90: 75Á87. Soil organic carbon (SOC) content is an important soil quality indicator that plays an important role in regulating physical, chemical and biological properties of soil. Field assessment of SOC is time consuming and expensive. It is difficult to obtain highresolution SOC distribution maps that are needed for landscape analysis of large areas. An artificial neural network (ANN) model was developed to predict SOC based on parameters derived from digital elevation model (DEM) together with soil properties extracted from widely available coarse resolution soil maps (1:1 000 000 scale). Field estimated SOC content data extracted from high-resolution soil maps (1:10 000 scale) in Black Brook Watershed in northwestern New Brunswick, Canada, were used to calibrate and validate the model. We found that vertical slope position (VSP) was the most important variable that determines distributions of SOC across the landscape. Other variables such as slope steepness, and potential solar radiation (PSR) also had significant influence on SOC distributions. The prediction of the selected two-input-node SOC model (VSP and coarse resolution soil map recorded SOC as inputs) had a correlation coefficient of 0.92 with measured values, and model predicted SOC values had 47.9% of the total points within 90.5% of the measured values and 70.6% within 91% of the measured values. The prediction of the selected four-input-node model (VSP, slope steepness, PSR and coarse resolution SOC values as inputs) had a correlation coefficient of 0.98 with measured values and model predicted SOC values had 75% of the total points within 90.5% of the measured values and 87% within 91% of the measured values. The prediction of the five-input-nodes model with soil drainage as additional input had a correlation coefficient of 0.99 with measured values, and model predicted SOC values had 87% of the total points within 90.5% of the measured values and 98% of the total points within 91% of the measured values. The calibrated SOC prediction model was used to produce a high-resolution SOC map for the Black Brook Watershed and the resulting SOC distribution map is considered to be realistic. Results indicated that DEM-derived hydrological parameters together with widely available coarse resolution soil map data could be used to produce high-resolution SOC maps with the ANN method. For personal use only.

Predicting soil organic carbon density using auxiliary environmental variables in northern Iran

“…Our results also showed a negative correlation (r= -0.15 and r=-0.19) for the soil layers at 0-20 cm and 20-40 cm, respectively, between SOCD and wetness index in the studied area. These results are in contrary to findings of other researchers (Zhao et al 2010;Schwanghart & Jarmer 2011). This may be related to the high vegetation density with low wetness index in the high elevation locations with lower human activities.…”

“…The aspect did not show a high correlation coefficient with SOCD in Deylaman region. Zhao et al (2010) indicated similar results for the aspect, while they reported that slope had a high correlation coefficient (r=0.74) with the SOC content. Our results also showed a negative correlation (r= -0.15 and r=-0.19) for the soil layers at 0-20 cm and 20-40 cm, respectively, between SOCD and wetness index in the studied area.…”

“…The R 2 for the predicted and observed C i and C e was 0.92 and 0.83. Zhao et al (2010) used different ANN models to produce soil organic carbon content distribution maps across different landscapes. They applied mean square error (MSE) and relative overall accuracy (ROA) for accuracy assessment.…”

On the application of Bayesian Networks in Digital Soil Mapping

“…Predictive modelling of the spatial pattern of soil types and properties is based on a quasi-mechanistic understanding of soil formation and the factors that drive soil variation in the landscape, namely the ClORPT factors (Climate, Organic activity, Relief, Parent material and Time; Jenny, 1941). The relationships between soil forming factors and soil properties are complex and several non-linear modelling techniques have been employed to represent them including Random Forests (RFs) (Liaw and Wiener, 2002;Grimm et al, 2008;Wiesmeier et al, 2011) and Artificial Neural Networks (ANNs) (Agyare et al, 2007;Zhao et al, 2010). A principal disadvantage of these methods is that they are 'black-box', meaning that it is often difficult to interpret the relationship between response and predictor variables in physical terms (Suuster et al, 2012).…”

Spatial Prediction of Soil Organic Matter Using a Hybrid Geostatistical Model of an Extreme Learning Machine and Ordinary Kriging

“…In fact, RK still displays a low prediction accuracy because of its weak analytical ability in non-linear relationships and layer structures between target data and multi-source auxiliary variables. Since the development of artificial intelligence and machine learning, neural networks (NN) have been used to solve the complex non-linear problems between soil properties and auxiliary variables, which results in a higher precision than when using classic linear methods [29][30][31]. However, traditional artificial neural networks (ANN) have a low implementation efficiency, which is needed to adjust the complex parameters from the algorithm structure and to avoid the influence of a locally optimal solution; in particular, they require a longer running time when the mapping resolution is increased.…”