Modeling Soil Carbon Stocks in Morobe Province, PNG

Abstract

This study was done to develop a best fit model for soil carbon distribution to 1-m depth in Papua New Guinea’s Morobe Province using multiple linear regression (MLR) technique on environmental variables elevation, slope, aspect, normalized difference vegetative index, mean annual temperature (MAT) and mean annual precipitation (MAP) using SPSS and ArcGis 10 software. Descriptive, correlation and MLR analyses were performed, and the data revealed that elevation and MAT were skewed. Slope and elevation were significantly negatively correlated to soil carbon distribution (R = −.725 at p < .05; R = −.862 at p < .01), while MAT was significantly positively correlated to the soil carbon distribution (R = .906 at p < .01). Three models developed from MLR analysis revealed that Model 2 (adjusted R2 = .987 at p < .05) and Model 3 (adjusted R2 = .990 at p < .05) were both significant, hence Model 2 developed Equation (3), whereas Model 3 developed Equation (4). Prediction accuracy of the two equations revealed that Equation (3) (root mean square error [RMSE] = 2.597) performed better than Equation (4) (RMSE = 2.764), hence Equation (3) was the best fit model that developed the predicted map of soil carbon distribution (Model 1 predicted approximately 271 t/ha). This study shows that environmental variables can be used to predict soil carbon distribution. However, the limited number of sites (n = 8) could have greatly affected the model development exercise (e.g., the surprising positive correlation of soil carbon with MAT) and consequently the accuracy of the prediction map.