基于随机森林算法的干旱半干旱区植被物候驱动机制与时滞累积效应研究

Phenological Driving Mechanism and Time-delay Cumulative Effect of Vegetation in Arid and Semi-arid Regions Based on Random Forest

摘要:

植被物候作为表征植被对气候变化响应的关键生态指标，其动态变化对干旱区生态系统稳定性具有重要影响。为深入揭示干旱半干旱区植被生长起始期（SOS）对多气候因子的响应机制及时滞累积效应，基于2001—2020年NDVI数据、多源气候数据和SOS数据，构建了融合随机森林（RF）模型、SHAP解释算法、时滞累积分析的综合评价体系。结果表明：气温日较差（DTR）、日平均温度（Tmean）、云量（CLD）和月总降水量（Pre）的季前长度分别集中在3~4个月(55.6%区域)、1~3个月(60.3%区域)、4~6个月(58.1%区域)和4~6个月(62.9%区域)。Pre是SOS最关键的驱动因子，CLD、DTR和Tmean则在特定情境下提供补充调节，并且不同类型气候因子的响应特征存在明显差异，其中Tmean与DTR等热量因子以滞后1~3个月、无累积效应为主；Pre、地表土壤含水率（SM）与标准化降水蒸散发指数（SPEI）等水分因子则表现出以当前月和短期累积为主的响应机制；CLD则主要体现为滞后2个月、累积1个月的调控效应。本研究揭示了干旱半干旱区植被物候对多气候因子驱动的响应规律及时滞累积效应，为构建区域尺度的气候-植被交互模型、农业生产布局优化及气候变化下植被动态预测提供了理论依据和方法支撑。

Abstract:

Vegetation phenology is a key ecological indicator to characterize the response of vegetation to climate change, and its dynamic changes have important effects on ecosystem stability in arid regions. To further reveal the response mechanism of vegetation onset period (SOS) to multiple climatic factors in arid and semi-arid areas, a comprehensive evaluation system combining random forest (RF) model, SHAP interpretation algorithm and time delay accumulation analysis was constructed based on NDVI data, multi-source climate data and SOS data from 2001 to 2020. The results showed that the pre-season length of diurnal temperature range (DTR), daily mean temperature (Tmean), cloud cover (CLD) and monthly total precipitation (Pre) were concentrated in 3~4 months (55.6% region), 1~3 months (60.3% region), 4~6 months (58.1% region) and 4~6 months (62.9% region), respectively. Pre was the most critical driver of SOS, while CLD, DTR and Tmean provided complementary regulation in specific situations. The response characteristics of different types of climate factors were significantly different, among which the thermal factors such as Tmean and DTR mainly had a lag of 1~3 months and no cumulative effect. The response mechanism of water factors such as Pre, surface soil moisture (SM) and standardized precipitation evapotranspiration index (SPEI) was mainly the current month and short-term accumulation. CLD was mainly reflected in the regulation effect of two months lag and one month accumulation. The results revealed the response of vegetation phenology to multiple climatic factors in arid and semi-arid areas, which provided theoretical basis and method support for the construction of climate-vegetation interaction models at regional scale, the optimization of agricultural production layout, and the prediction of vegetation dynamics under climate change.