在全球气候变化的背景下，干旱、热浪与强降水等极端气候事件发生的频率越来越高，对陆地生态系统生产力及碳循环产生了重要影响。总初级生产力（GPP）作为陆地生态系统的碳输入，直接影响着植被和土壤碳的流动和净的碳收支状况，因而在生态系统碳循环中扮演着重要的作用。水分是影响GPP的重要因素，但水分亏缺的影响往往与其他气候因素（如温度、太阳辐射）等综合在一起发生作用，表现出滞后性和遗产效应。这意味着，当期水分条件和前期水分条件都会对当期的总初级生产力的大小产生影响，并具有空间异质性，量化当年和前期水分条件的影响及其空间差异是学界的一个研究热点。揭示不同区域总初级生产力对干旱胁迫的响应特征，以及不同时期水分亏缺的影响差异，对准确模拟和评估未来气候变化影响下的总初级生产力的变化非常重要。

本研究以中国陆地生态系统为研究区域，开展了前期水分条件对中国总初级生产力的影响研究。在研究中，首先基于中分辨率成像光谱仪（Moderate-resolution Imaging Spectroradiometer，MODIS）的总初级生产力（GPP）遥感数据和气象数据，利用偏相关分析方法对不同气候因子（温度、水分、辐射）敏感区进行了识别，进而分析了不同敏感区内GPP对干旱胁迫的响应差异。在此基础上，以水分敏感区为重点研究区，基于机器学习算法（随机森林、神经网络等），从变量的重要性角度，评估了不同时期水分亏缺条件对GPP的影响大小。最后，利用气候研究中心（Climatic Research Unit, CRU）的气象格点数据，设计了一系列的水分亏缺情景，通过不同情景的对比分析，定量评估了不同时期水分亏缺的影响。

研究结果发现：（1）气候因子（即水分、温度、辐射）对GPP的影响具有较高的空间异质性，水分敏感区相对于非水分敏感区表现出更强的GPP-水分依赖关系，即在水分敏感区内，植被总初级生产力随干旱强度的增加呈明显下降的趋势，而在温度敏感区和辐射敏感区，水分变化对GPP的影响不明显。（2）基于机器学习算法的研究结果表明：当年的水分条件对GPP的影响最大，但前期水分条件的影响也不容忽视。（3）基于水分亏缺情景分析所揭示的结果同样表明，GPP主要受当年的水分条件控制，但也会有条件地反映前期水分条件的影响，表现为前期水分条件的影响具有多样性，其对GPP的影响程度与当年的水分条件有关。具体而言，当当年水分条件处于干旱时，GPP主要受当前年干旱的影响，前期水分条件的影响表现得并不显著；相反，当当年水分条件正常或较湿润时，前第1年水分条件的好坏亦会对GPP产生影响，表现为前第1年的水分条件越好，即植被越健康，则当年的植被生产力也会越高。

本研究揭示了前期水分条件对当期GPP的影响及其时滞效应与遗产效应，这意味着在预测气候变化引起的GPP变化时，不仅要考虑当年的水分条件，还需要考虑前期（尤其是前第1年）水分条件的动态变化。

Under the background of global climate change, the frequency of extreme weather events such as drought, heat waves and heavy rainfall is increasing, which has an important influence on the productivity and carbon cycle of terrestrial ecosystems. As the carbon input of terrestrial ecosystem, gross primary productivity (GPP) plays an important role in the carbon cycle of ecosystem, which directly affects the carbon flow of vegetation and soil and the net carbon budget. Water is an important factor affecting GPP, but the effect of water deficit is often combined with other climatic factors (such as temperature, solar radiation), showing time-lag and legacy effect. This means that both current and previous water conditions have an impact on the gross primary productivity in the current period, and the impact shows spatial heterogeneity. Quantifying the impact of current and previous water conditions and their spatial differences is a hot topic in the academic circle. Revealing the response characteristics of gross primary productivity to drought stress in different regions and the differences of the effects of water deficit in different periods is very important to accurately simulate and evaluate the changes of GPP under the influence of future climate change.

In this study, the impact of water conditions on China's GPP was studied taking the terrestrial ecosystem of China as the research area. Firstly, this study identified the different climate (temperature, water, radiation) sensitive regions with the method of partial correlation analysis based on the gross primary productivity (GPP) of the Moderate resolution Imaging Spectroradiometer (MODIS) of remote sensing data and meteorological data, and then analyzed the response of GPP in the different sensitive regions to drought stress. On this basis, the influence of water deficit conditions in different periods on GPP was evaluated from the perspective of the importance of variables based on the machine learning algorithm (random forest, neural network, etc.) with the water-sensitive areas as the principal research area. Finally, a series of water deficit scenarios were designed using the meteorological data of Climatic Research Unit (CRU) and the effects of water deficit in different periods were quantitatively evaluated through comparing and analyzing different scenarios.

The results showed: (1) The impact of climate factor (i.e., water, temperature and radiation) on GPP has higher spatial heterogeneity. The water-sensitive regions showed a stronger water-GPP relationship than water-insensitive regions. That is, the GPP of vegetation appeared an obvious downward trend with the increase of the drought intensity in water-sensitive regions, while the influence of water on GPP is not obvious in the temperature-sensitive regions and radiation-sensitive regions. (2) The research results based on machine learning algorithm showed that the current water conditions have the greatest influence on GPP, but the influence of previous water conditions cannot be ignored. (3) The results revealed based on the analysis of water deficit scenarios also showed that GPP is mainly controlled by water conditions in the current year, but it also conditionally reflects the influence of water conditions in the previous years. It shows that the influence of previous water conditions is multifarious, and its influence degree on GPP is related to the current year's water conditions. Specifically, when the water conditions in the current year were in drought, GPP was mainly affected by the drought in current year, and the influence of the water conditions in the previous years was not significant. On the contrary, when the water conditions in the first year were normal or wetter, the quality of water conditions in the first previous year would also have an impact on GPP. The better the water conditions in the first previous year were, the healthier the vegetation was, and then the higher the vegetation productivity in the current year would be.

The impact of water conditions in previous years on GPP and its time-lag effect and legacy effect was revealed in this study, which implies that not only current year’s water condition but also its dynamic changes in previous years (especially the first year) should be considered when predicting changes in GPP caused by climate change.