Compatible versus annual stand-growth models for thinned Chinese fir plantations

Introduction Chinese fir (Cunninghamia lanceolata) plantations are a major timber resource in subtropical China, where thinning strongly affects stand growth and yield forecasting. Reliable models are needed to balance short-term post-thinning responsiveness with long-term path consistency. This study compares two stand-level systems—Compatible and Annual (periodic-annual) Growth models—to evaluate their performance across thinning regimes and forecast horizons. Methods Data were derived from a 40-year thinning trial in southern China. Stand survival and basal area were modeled for thinned and unthinned stands using two model classes: (i) a Compatible system ensuring algebraic path consistency and (ii) an Annual system emphasizing short-interval responsiveness. Each was fitted with two data structures: consecutive (non-overlapping) and all possible (overlapping) growth pairs. Parameters were estimated using Seemingly Unrelated Regressions (SUR) and validated through threefold cross-validation for short (2–4 yr), medium (6–8 yr), and long (≥10 yr) horizons. Results In unthinned stands, Annual models with all growth pairs achieved the best short-term accuracy, while Compatible models performed best at medium and long horizons. In thinned stands, Annual models most accurately captured short-term survival, whereas Compatible models yielded superior basal area predictions at short to medium horizons. All-pairs estimation generally improved precision, though consecutive pairs reduced bias in short-term post-thinning basal area. Discussion Model behavior revealed a trade-off between responsiveness and path consistency. Annual systems are advantageous immediately after thinning, while Compatible systems provide stable long-term projections. For management of C. lanceolata plantations, we recommend Annual + All-pairs for short-term forecasting and Compatible + All-pairs for medium-to-long horizons, with recalibration every 6–8 years to maintain predictive reliability.