Abstract
Litter is essential to promote nutrient cycling and maintain the sustainability of forest resources. However, its variability has not been sufficiently studied at the local scale. The prediction of litter amount using ordinary cokriging with Pearson correlation analysis (COK(P)) and ordinary cokriging with principal component analysis (COK(PCA)) was compared with that using ordinary kriging (OK) based on cross-validation at the local scale of a 1-ha plot over natural spruce-fir mixed forest in Jilin Province, China. Litter samples in semidecomposed (F) and complete decomposed (H) horizons were collected using an equidistant grid point sampling of 10 m × 10 m. Pearson correlation analysis and principal component analysis (PCA) were used to confirm auxiliary variables. The results showed that the amount of litter was 19.65 t/ha in the F horizon and 10.37 t/ha in the H horizon. The spatial structure variance ratio in the H horizon was smaller than that in the F horizon, indicative of its stronger spatial autocorrelation. Spatial distributions of litter amount in both horizons exhibited a patchy and heterogeneous pattern. Of the selected stand characteristics and litter properties, litter moisture content indicated the strongest relationship with litter amount. Cross-validation revealed that COK(PCA) using the comprehensive score as an auxiliary variable produced the most accurate map. The average standard error and root-mean-square error between the predicted and measured values were always smaller, the mean error and mean standardized error were much closer to 0, and the root-mean-square standardized error was closer to 1 than COK(P) using litter moisture and OK. Therefore, a clear advantage of cokriging based on principal component analysis was observed and COK(PCA) was found to be a very useful approach for further interpolation prediction.