Data_Sheet_1_Antecedent climatic conditions spanning several years influence multiple land-surface phenology events in semi-arid environments.docx

Ecological processes are complex, often exhibiting non-linear, interactive, or hierarchical relationships. Furthermore, models identifying drivers of phenology are constrained by uncertainty regarding predictors, interactions across scales, and legacy impacts of prior climate conditions. Nonetheless, measuring and modeling ecosystem processes such as phenology remains critical for management of ecological systems and the social systems they support. We used random forest models to assess which combination of climate, location, edaphic, vegetation composition, and disturbance variables best predict several phenological responses in three dominant land cover types in the U.S. Northwestern Great Plains (NWP). We derived phenological measures from the 25-year series of AVHRR satellite data and characterized climatic predictors (i.e., multiple moisture and/or temperature based variables) over seasonal and annual timeframes within the current year and up to 4 years prior. We found that antecedent conditions, from seasons to years before the current, were strongly associated with phenological measures, apparently mediating the responses of communities to current-year conditions. For example, at least one measure of antecedent-moisture availability [precipitation or vapor pressure deficit (VPD)] over multiple years was a key predictor of all productivity measures. Variables including longer-term lags or prior year sums, such as multi-year-cumulative moisture conditions of maximum VPD, were top predictors for start of season. Productivity measures were also associated with contextual variables such as soil characteristics and vegetation composition. Phenology is a key process that profoundly affects organism-environment relationships, spatio-temporal patterns in ecosystem structure and function, and other ecosystem dynamics. Phenology, however, is complex, and is mediated by lagged effects, interactions, and a diversity of potential drivers; nonetheless, the incorporation of antecedent conditions and contextual variables can improve models of phenology.

生态过程极为复杂，往往呈现非线性、交互性或层级性的关联关系。此外，针对物候（phenology）驱动因子的建模工作常受限于预测变量、跨尺度交互作用以及前期气候条件遗留效应带来的不确定性。尽管如此，对物候等生态过程进行观测与建模，对于生态系统及其支撑的社会系统的管理仍至关重要。本研究采用随机森林（random forest）模型，评估气候、区位、土壤（edaphic）属性、植被组成及干扰因子的最优组合，以精准预测美国西北大平原（Northwestern Great Plains, NWP）三种优势土地覆盖类型下的多种物候响应。研究基于25年序列的高级甚高分辨率辐射计（Advanced Very-High-Resolution Radiometer, AVHRR）卫星数据提取物候指标，并针对当年及最多4年前的季节与年度时间尺度，刻画了基于水分和/或温度的多类气候预测变量特征。本研究发现，从季节尺度到多年尺度的前期气候条件，与物候指标存在显著关联，且显然介导了群落对当年气候条件的响应。例如，多年序列中至少一项前期水分可利用性指标[降水量或水汽压亏缺（vapor pressure deficit, VPD）]，是所有生产力指标的关键预测因子。包含长期滞后项或前年度累加值的变量（如最大VPD的多年累积水分条件），是物候始期的顶级预测变量。生产力指标同样与土壤属性、植被组成等背景变量存在关联。物候是一类核心生态过程，深刻影响着生物与环境的关联、生态系统结构与功能的时空格局，以及其他生态系统动态。然而，物候过程极为复杂，其响应受滞后效应、交互作用及多样潜在驱动因子的调控；尽管如此，纳入前期条件与背景变量可有效优化物候模型的性能。