关键词: 智慧林业, 森林生态系统, 高质量发展, 影响因素

Abstract:

[Objective] Against the backdrop of smart forestry, digital monitoring, low-altitude operations, information networks, and data-driven governance are reshaping how forest resources are protected, managed, and converted into ecological, economic, and social value. Nevertheless, the high-quality development of forest ecosystems (FEHQD) in China remains spatially uneven, while the dominant drivers, nonlinear responses, and transmission pathways linking technological support, factor allocation, ecological governance, and regional economic conditions have not been adequately clarified. This study evaluates regional disparities and the distributional evolution of FEHQD, identifies its principal influencing factors, and examines the pathways through which these factors are associated with FEHQD. [Methods] Using panel data for 31 provincial-level regions in China from 2005 to 2024, excluding Hong Kong, Macao, and Taiwan, this study integrates the Social-Ecological Systems framework, the Driver-Pressure-State-Impact-Response framework, and the Stochastic Impacts by Regression on Population, Affluence, and Technology framework to construct a multidimensional FEHQD evaluation system. The entropy-weighted Technique for Order Preference by Similarity to an Ideal Solution is used to calculate provincial FEHQD scores. Kernel density estimation is used to depict distributional dynamics, and the Dagum Gini coefficient decomposes disparities into within-region, between-region, and transvariation components. An eXtreme Gradient Boosting model interpreted using Shapley Additive Explanations (XGBoost-SHAP) identifies nonlinear influences, thresholds, interactions, and relative contributions. Partial least squares structural equation modeling (PLS-SEM), with 5000 bootstrap resamples, then assesses the direct and indirect relationships among five leading factors selected from the machine-learning results. [Results] The findings indicate that China’s FEHQD distribution shifted overall from lower levels toward medium and higher levels during the study period, but this improvement did not result in stable spatial convergence. National kernel density curves shifted rightward, gradually changed from a single-peak to a multi-peak pattern, and developed a longer right tail, indicating both an expanding group of higher-performing provinces and persistent stratification. The eastern region experienced the most pronounced rightward shift and developed a relatively stable high-level cluster, although its bimodal pattern reveals internal polarization. The central region advanced more steadily and retained comparatively stronger internal coordination, whereas the western and northeastern regions exhibited increasingly visible multimodal differentiation. Consistent with these patterns, the overall Dagum Gini coefficient rose, with fluctuations, from 0.058 in 2005 to 0.083 in 2024, after reaching its sample-period maximum of 0.117 in 2012. Since 2018, it has fluctuated within a relatively narrow range of 0.079-0.093, remaining above its initial value and providing no evidence of stable convergence. Western China generally recorded the greatest within-region inequality, while the western-northeastern and eastern-western gaps remained comparatively large. More importantly, the contribution of transvariation density increased from 38.730% in 2005 to 57.114% in 2024, whereas the between-region contribution declined from 33.138% to 14.149%; the within-region contribution remained near 28%. Thus, the current disparity is driven less by simple differences between regional averages than by cross-regional overlap, distributional dislocation, and the coexistence of leading and lagging provinces within conventional regional divisions. The XGBoost model achieved an R² of 0.994 for the training set and 0.823 for the test set, demonstrating substantial out-of-sample explanatory capacity while also warranting caution because of the gap between training and test performance. SHAP results show that the coordination degree of forestry elements (CCD) is the most important factor, contributing for 26.44%, followed by economic development level (EDL) at 24.35% and the low-altitude economy (LAE) at 12.09%; together, these three factors account for 62.88% of total importance. Forestry capital productivity improvement (LCPI), land productivity improvement (LPI), forestry labor productivity improvement (LFPI), and information transmission infrastructure upgrading (ITIU) contribute 10.51%, 7.09%, 7.01%, and 6.34%, respectively. Forest pest control (FPC) and fiscal support for forestry (FFS), with shares of 3.25% and 2.92%, respectively, play smaller but still relevant safeguarding roles. These rankings show that FEHQD depends primarily on coordinated factor allocation and economic capacity, while technological infrastructure, productivity, fiscal support, and ecological governance provide complementary conditions. SHAP dependence plots further reveal pronounced nonlinearity and stage dependence. LAE has predominantly negative SHAP values below approximately 0.60 but rises rapidly thereafter, implying that low-altitude monitoring, unmanned aerial vehicle patrols, and disaster identification generate discernible benefits only after applications and supporting conditions reach sufficient scale. CCD shifts from predominantly negative to positive at approximately 0.40 and remains strongly positive above 0.60, confirming a coordination threshold among land, labor, and capital. LCPI shows an increasingly positive contribution, whereas LPI generally declines as its value rises, suggesting that output-oriented land intensification may conflict with ecological carrying capacity. LFPI follows a decline-then-rise pattern; ITIU initially declines and later recovers without displaying a consistently positive contribution; FFS shows diminishing marginal returns; EDL rises continuously; and FPC fluctuates around zero, consistent with a stabilizing rather than growth-driving function. Pairwise SHAP results also identify reinforcing interactions among LAE, ITIU, LCPI, CCD, and EDL. The PLS-SEM results clarify that high predictive importance does not necessarily imply a strong direct path. EDL has the strongest positive direct relationship with FEHQD (β = 0.669, p < 0.001), followed by CCD (β = 0.512, p < 0.001) and LCPI (β = 0.218, p < 0.001). In contrast, LPI is negatively related to FEHQD (β = −0.096, p < 0.001). LAE has no significant direct relationship with FEHQD (β = −0.030, p = 0.242), despite ranking third in SHAP importance; instead, it is strongly associated with EDL (β = 0.747, p < 0.001) and positively associated with LCPI (β = 0.164, p = 0.003), indicating an indirect, condition-dependent role. EDL is also associated with LCPI (β = 0.126, p = 0.045) and LPI (β = 0.384, p < 0.001), showing that economic foundations operate both directly and through productivity channels. These results support a mechanism in which smart-forestry applications require economic and industrial support, while coordinated allocation determines whether technology and investment translate into durable ecological outcomes. [Conclusion] This study contributes by unifying three complementary theoretical frameworks and linking comprehensive evaluation, disparity decomposition, interpretable machine learning, and structural pathway analysis. In practical terms, China should replace uniform interventions with region-specific governance; embed low-altitude technologies and information networks in forest inventory, pest monitoring, fire warning, and precision management; coordinate land, labor, capital, technology, and fiscal investment; and prevent productivity expansion from exceeding ecological constraints. Particular attention should be directed toward lagging and internally polarized provinces rather than relying only on broad east-central-west-northeast classifications. Overall, improving FEHQD requires a policy portfolio that combines economic capacity, efficient capital use, cross-factor coordination, and routine ecological risk governance.

Key words: smart forestry, forest ecosystem, high-quality development, influencing factors