Title: Kriging, Regression and Machine Learning For Predicting Forest Floor Carbon Content In A Temperate Managed Forest
Abstract:
In most temperate forests, the forest floor is an important reservoir of carbon and nutrients that greatly influences forest productivity and soil fertility. Small changes in this store can alter the balance of the ecosystem, which is why it is essential to develop methodologies that improve estimates at the landscape level. However, precise assessments of these reservoirs represent a challenge due to their great variability and spatial dependence. In this study, three popular spatial modeling approaches (Kriging, Regression and Machine Learning) were compared to map carbon contents (CC) in the forest floor of a temperate forest. Methods include Ordinary Kriging (OK), Generalized Linear Model (GLM), Generalized Additive Model (GAM), and Random Forest (RF). The CC estimates were made for two years, 2013 and 2018. The predictor variables used represent the spatial, topographic and canopy structure. All models were evaluated by cross validation (k=10) and the mean absolute error (MAE), root mean square error (RMSE) and the coefficient of determination (R2) were determined. The results showed that the performance of the methods was, in decreasing order, RF, GAM, GLM and OK. The OK method reflected the degree of spatial dependence of the CC but the spatial estimates were unrealistic (R2<0.35). GAM and GLM showed good performance (R2>0.70), but higher levels of CC overestimation. RF obtained the best fit (R2>0.86) to model CC in both years evaluated. This study concludes that the RF model is a promising approach with great potential for improving forest floor carbon estimates at the local scale.
Biography:
Martha Elva Ramírez Guzmán has completed his PhD at the age of 30 years from Reading University, UK. She is a Professor of Statistics of Colegio de Postgraduados, Mexico. She has published more than 30 papers in reputed journals
