Yield prediction errors of a stand density management program for black spruce and consequences for model improvement

Abstract

The objectives of this study were to (i) quantify the prediction error associated with estimating density (N (stems/ha)), quadratic mean diameter (Dq (cm)), basal area (G (m2/ha)), total volume (Vt (m3/ha)), and merchantable volume (Vm (m3/ha)) using a stand density management decision-support program (SDMDSP) developed for black spruce (Picea mariana (Mill.) BSP) plantations and (ii) given objective i, assess model adequacy by examining the relationship between prediction error and model input variables (prediction period, site index, initial density, and number of thinning treatments) by yield variate. Specifically, the SDMDSP was evaluated by comparing its yield predictions with corresponding measured values (n = 44) within 19 black spruce plantations. The resultant tolerance intervals indicated that 95% of the relative errors associated with future predictions would be within the following limits 95% of the time (minimum–maximum): (i) –27.3 to 29.7% for N, (ii) –26.1 to 14.3% for Dq, (iii) –48.3 to 26.1% for G, (iv) –64.3 to 37.7% for Vt, and (v) –87.0 to 73.0% for Vm. Graphical analysis indicated that errors for Vt and Vm were associated with the data from thinned plantations. This result is discussed within the context of residual stand structure variation and response delay from which recommendations for model improvement are derived.