Theisen 气候战争

Climate Wars? 气候战争？

Assessing the Claim That Drought Breeds Conflict
评估干旱引发冲突的说法

Ole Magnus Theisen, Helge Holtermann, and Halvard Buhaug
Ole Magnus Theisen、Helge Holtermann 和 Halvard Buhaug

Climate change is hot. 气候变化是热门话题。

Twice in recent years, the Nobel Peace Prize has been awarded to environmental activists: in 2004 to Wangari Maathai and in 2007 to the United Nations’ Intergovernmental Panel on Climate Change (IPCC) and former U.S. Vice President Al Gore. In April 2007, the UN Security Council held its first ever debate on climate security. The chair of this debate, then British Foreign Secretary Margaret Becket, left no doubt as to the connection between climate and conflict: “What makes wars start? Fights over water. Changing patterns of rainfall. Fights over food production, land use.”

^{1}

In the same year, a report by eleven retired U.S. generals and admirals stated that environmental security is no longer soft politics, concluding that climate change is a “threat multiplier” for instability and conflict that will have repercussions for all.

^{2}

And in a speech to the UN on September 22, 2009, U.S. President and Nobel laureate Barack Obama asserted that “the threat from climate changes is serious, it is urgent, and it is growing,” as more frequent droughts and crop failures “breed hunger and conflict.”

^{3}

Surely, such statements must be based on solid scientific evidence-much in the same manner as the natural sciences inform the debate on likely physical changes? Not so. As a matter of fact, the policy debate on the
近年来，诺贝尔和平奖两次授予环保活动家：2004 年授予旺加里-马塔伊，2007 年授予联合国政府间气候变化专门委员会（IPCC）和美国前副总统戈尔。2007 年 4 月，联合国安理会首次就气候安全问题举行辩论。辩论主席、时任英国外交大臣玛格丽特-贝克特毫无疑问地指出了气候与冲突之间的联系："是什么引发了战争？争夺水资源。降雨模式的变化。争夺粮食生产和土地使用"。

^{1}

同年，11 位退役的美国将军和海军上将在一份报告中指出，环境安全不再是软政治，并得出结论：气候变化是不稳定和冲突的 "威胁倍增器"，将对所有人产生影响。

^{2}

2009年9月22日，美国总统、诺贝尔奖获得者巴拉克-奥巴马（Barack Obama）在联合国发表演讲时断言，"气候变化的威胁是严重的、紧迫的，而且正在不断加剧"，因为更频繁的干旱和作物歉收 "会滋生饥饿和冲突"。

^{3}

当然，这样的说法肯定是基于确凿的科学证据--就像自然科学为关于可能发生的物理变化的辩论提供信息一样？并非如此。事实上，关于 "饥饿与冲突 "的政策辩论是以科学证据为基础的。

The authors would like to thank Andreas Tollefsen at PRIO for his excellent research assistance as well as Brad Lyon at the International Research Institute for Climate and Society and Gregory Yetman at the Center for International Earth Science Information Network, both at Columbia University, for generously providing the precipitation data and advising them on the Standard Precipitation Index, SPI6, indicator. They are also grateful to the anonymous reviewers and to Thomas Homer-Dixon for constructive comments and criticisms, as well as to their colleagues at PRIO and participants at various conferences for helpful input on earlier drafts. Research on this article was carried out with financial support from the Research Council of Norway and NTNU. Replication data and supporting material can be found at http://www.prio.no/CSCW/Datasets/Replication-Data-List/.
作者感谢 PRIO 的 Andreas Tollefsen 提供的出色研究协助，感谢哥伦比亚大学气候与社会国际研究所的 Brad Lyon 和国际地球科学信息网络中心的 Gregory Yetman 慷慨提供降水数据，并就标准降水指数 SPI6 指标提出建议。他们还感谢匿名审稿人和托马斯-霍默-迪克森（Thomas Homer-Dixon）提出的建设性意见和批评，以及 PRIO 的同事和各种会议的与会者对早期草稿提出的有益意见。本文的研究工作得到了挪威研究理事会和挪威国立师范大学的资助。复制数据和辅助材料见http://www.prio.no/CSCW/Datasets/Replication-Data-List/。

security implications of climate change has run far ahead of the scientific evidence base. This study represents one scholarly attempt to catch up with the rhetoric.
气候变化对安全的影响已远远超过了科学证据基础。本研究是学术界追赶言论的一次尝试。

At the heart of the climate security discourse lies the issue of water scarcity. A key characteristic of the world’s poorest and most vulnerable societies is their dependence on rain-fed agriculture for income and food supply. Global warming is likely to affect precipitation patterns and increase the unpredictability of extreme weather events, thereby probably having a negative impact on health and food security in many parts of the world.

^{4}

Some argue that these developments might also have implications for peace and security in a stricter sense. The environmental security literature offers several case-based accounts of armed conflict within the context of competition over scarce resources.

^{5}

Yet, it remains unclear whether these cases are exceptions or whether they epitomize a more systematic pattern of resource scarcity and conflict, in general, and drought and violent conflict, in particular.

^{6}

气候安全讨论的核心是缺水问题。世界上最贫穷和最脆弱社会的一个主要特点是依赖雨水灌溉的农业来获得收入和粮食供应。全球变暖可能会影响降水模式，增加极端天气事件的不可预测性，从而可能对世界许多地区的健康和粮食安全产生负面影响。

^{4}

一些人认为，这些事态发展也可能对更严格意义上的和平与安全产生影响。环境安全文献对争夺稀缺资源背景下的武装冲突进行了若干基于案例的描述。

^{5}

然而，目前仍不清楚这些案例是例外情况，还是资源稀缺与冲突，特别是干旱与暴力冲突的更系统化模式的缩影。

^{6}

This study offers a rigorous assessment of the claim that drought and water shortages increase the risk of civil war.

^{7}

In contrast to earlier attempts to study
本研究对干旱和缺水会增加内战风险的说法进行了严格评估。

^{7}

与先前的研究尝试不同的是

the scarcity-conflict nexus in a comparative manner, we explicitly incorporate the role of ethnopolitical structures. Not all groups in a society are equally vulnerable to environmental shocks. Almost all accounts of land and water conflicts in Africa concern peripheral and neglected groups in weak or oppressive regimes-even though the nature of the political system in these narratives often remains implicit. Environmental hardships, such as prolonged drought, tend to accentuate societal divides, as marginalized groups lack alternative means of livelihood and income and are less likely to be at the receiving end of government-sponsored redistribution programs and relief aid. This leads to a second significant improvement of this study: its geographically disaggregated design. Grievances and human suffering will emerge first, and be most acute, in locations where drought coincides with political and economic marginalization. Local, short-term implications could include lowered opportunity cost of rebel recruitment and a higher motivation for using violence to redress grievances. Therefore, if leading politicians, think tanks, and environmental security scholars are correct-if a regular pattern of increasing water scarcity and increasing risk of violent conflict truly exists-this should be observed where drought strikes marginalized populations in poor, agrarian, nondemocratic societies.
在比较稀缺与冲突的关系时，我们明确纳入了民族政治结构的作用。并非社会中的所有群体都同样容易受到环境冲击的影响。几乎所有关于非洲土地和水冲突的描述都涉及到弱小或专制政权中被忽视的边缘群体--尽管在这些描述中政治体制的性质往往是隐含的。长期干旱等环境困难往往会加剧社会分化，因为边缘化群体缺乏其他谋生手段和收入来源，不太可能接受政府发起的再分配计划和救济援助。这就引出了本研究的第二个重大改进：按地域分类的设计。在干旱与政治和经济边缘化同时发生的地方，人们的不满和痛苦会最先出现，也最严重。对当地的短期影响可能包括降低招募反叛分子的机会成本，以及提高使用暴力平息不满情绪的动机。因此，如果主要政治家、智囊团和环境安全学者的观点是正确的--如果缺水加剧和暴力冲突风险增加的规律真的存在--那么在干旱袭击贫困、农业和非民主社会的边缘化人群时，就应该观察到这种规律。

To evaluate the empirical validity of this general proposition, we employ a high-resolution gridded dataset of Africa from 1960 to 2004 that combines georeferenced and annualized precipitation data with new data on the point location of civil war onset and the location and political status of ethnic groups.

^{8}

We test a large selection of alternative location-specific drought measures and allow for both direct and conditional relationships, where the effect of drought is contingent on various sociopolitical characteristics at the local as well as the national level. In contrast to popular conception, the analysis reveals little evidence of a drought-conflict connection. Although we find strong support for the exclusion perspective-African civil wars break out disproportionately in politically marginalized areas-this statistical regularity is unaf-
为了评估这一一般性命题的实证有效性，我们采用了 1960 年至 2004 年非洲的高分辨率网格数据集，该数据集将地理参照和年化降水量数据与有关内战爆发点位置以及族群位置和政治地位的新数据相结合。

^{8}

我们测试了大量可供选择的特定地点干旱衡量标准，并考虑到直接关系和条件关系，即干旱的影响取决于当地和国家层面的各种社会政治特征。与流行的概念不同，分析几乎没有显示出干旱与冲突之间的联系。尽管我们发现排他性观点得到了强有力的支持--非洲内战不成比例地发生在政治边缘化地区--但这种统计规律性并不可靠。

fected by abrupt local water shortages. This finding calls for moderation when discussing security implications of climate change, particularly within the context of policy advice and practice.
受当地突然缺水的影响。这一发现要求在讨论气候变化对安全的影响时，特别是在政策建议和实践的背景下，要适度。
The article is organized as follows. We begin by briefly describing contemporary precipitation patterns and trends in Africa. Existing arguments linking drought to civil war are then surveyed before we explain why any measurable impact of drought is likely to be contingent on ethnopolitical power relations. Thereafter, we outline how this proposition is tested using a spatially disaggregated design, before turning to the results and our interpretation of them. We conclude by discussing some implications of our findings and point to fruitful avenues for future research.
文章结构如下。首先，我们简要介绍了非洲当代的降水模式和趋势。然后，我们对干旱与内战相关的现有论点进行了调查，然后解释了为什么干旱的任何可衡量影响都可能取决于民族政治权力关系。随后，我们概述了如何利用空间分类设计来检验这一命题，然后再讨论结果和我们对结果的解释。最后，我们将讨论研究结果的一些影响，并指出未来研究的有效途径。

Backdrop: Trends in Precipitation Patterns
背景：降水模式趋势

The world is heating up. Although the most recent years on average have been marginally cooler than they were at the turn of the century, 2010 tied 2005 as the warmest year on record.

^{9}

Simultaneously, we have witnessed a dramatic decrease in seasonal Arctic sea ice extent, and the global rate of hydrometeorological natural disasters has increased exponentially over the last few decades, even if it is hard to quantify the exact contribution of global warming.

^{10}

世界正在升温。虽然最近几年的平均气温比本世纪初略低，但 2010 年已追平 2005 年，成为有记录以来最热的一年。

^{9}

与此同时，我们目睹了北极季节性海冰面积的急剧下降，全球水文气象自然灾害的发生率在过去几十年中成倍增加，尽管很难量化全球变暖的确切影响。

^{10}

Climate change is not felt equally in all corners of the world, however. Warming occurs disproportionately in the Northern Hemisphere, and particularly at higher latitudes. Precipitation patterns display similar, though not always overlapping, changes, with considerable interregional and interannual variations. According to the IPCC Fourth Assessment Report, precipitation will increase in higher latitudes and decrease in the subtropics, where droughts will become more frequent and more intense.

^{11}

然而，气候变化在世界各地的影响并不相同。北半球，尤其是高纬度地区的气候变暖尤为严重。降水模式也显示出类似的变化，尽管并不总是相互重叠，但区域间和年际间的变化相当大。根据政府间气候变化专门委员会第四次评估报告，高纬度地区的降水量将增加，而亚热带地区的降水量将减少，干旱将变得更加频繁和剧烈。

^{11}

Just as climate change differs among regions, so does vulnerability to shift-
正如气候变化因地区而异，易受气候变化影响的程度也不尽相同。

ing environmental conditions. Africa is projected to be hit first and most extensively by a less hospitable climate, given its economic dependence on rain-fed agriculture, high environmental vulnerability, and weak institutional coping capacity.

^{12}

Only 4 percent of arable land in Sub-Saharan Africa is irrigated, making the predominantly agricultural African economies poorly suited to withstand increasing drought. The result might be substantial vegetation dieoff in exposed regions, with negative implications for agricultural productivity and food security.

^{13}

One-third of the African population lives in drought-prone areas today, and almost all Sub-Saharan countries are projected to be in a state of water stress by 2025.

^{14}

Moreover, two-thirds of the workforce in SubSaharan Africa is employed in the rural sector, making this region especially sensitive to future climate changes.

^{15}

环境条件的变化。由于非洲在经济上依赖雨水灌溉的农业，环境高度脆弱，机构应对能力薄弱，因此预计非洲将首先受到不太适宜的气候的影响，而且影响范围最广。

^{12}

撒哈拉以南非洲地区只有4%的耕地得到灌溉，这使得以农业为主的非洲经济难以承受日益严重的干旱。其结果可能是暴露地区的植被大量死亡，对农业生产率和粮食安全造成负面影响。

^{13}

目前，三分之一的非洲人口生活在干旱易发地区，预计到 2025 年，几乎所有撒哈拉以南非洲国家都将处于水资源紧张状态。

^{14}

此外，撒哈拉以南非洲三分之二的劳动力在农村就业，这使得该地区对未来的气候变化尤为敏感。

^{15}

The direction and extent of climate change vary also within regions. As shown in figure 1, parts of East Africa and the Horn have become considerably wetter over the last half century while other areas, notably central and western Africa, have seen a decline in annual precipitation. Results from global circulation models of various IPCC scenarios suggest that these intracontinental trends will continue.

^{16}

It has to be noted, however, that there is considerable uncertainty in these projections, especially at the subregional level.
各地区气候变化的方向和程度也不尽相同。如图 1 所示，在过去半个世纪里，东非和非洲之角的部分地区变得相当湿润，而其他地区，特别是非洲中部和西部，年降水量有所下降。政府间气候变化专门委员会各种方案的全球环流模型的结果表明，这些洲内趋势将继续下去。

^{16}

然而，必须指出的是，这些预测存在相当大的不确定性，尤其是在次区域一级。

Linking Drought to Civil War
将干旱与内战联系起来

The general expectation that failing precipitation could cause armed conflict draws on an influential body of work commonly referred to as the environmental security literature. Thomas Homer-Dixon’s well-known typology of
人们普遍认为降水不足会引发武装冲突，这种看法借鉴了通常被称为环境安全文献的有影响力的研究成果。托马斯-霍默-迪克森（Thomas Homer-Dixon）提出了著名的

Figure 1. Changes in Mean Annual Precipitation from 1952-61 to 1995-2004
图 1.1952-61 年至 1995-2004 年年平均降水量的变化

NOTE: The map shows changes in mean annual precipitation from the first to the most recent decade of precipitation data (1952-61 and 1995-2004, respectively). White areas are deserts with mean annual precipitation of less than 100 millimeters. See data and measurements section for further details.
注：地图显示了从第一个十年降水数据到最近十年降水数据（分别为 1952-61 年和 1995-2004 年）的年平均降水量变化。白色区域为年平均降水量小于 100 毫米的沙漠。更多详情，请参阅 "数据和测量 "部分。
supply- and demand-induced and structural scarcities of renewable resources has informed much research in this area.

^{17}

According to this school of thought, less developed countries are most predisposed to endemic resource shortages because they lack the necessary knowledge and capabilities to overcome persistent resource pressure. Likely consequences include reduced economic productivity, migration, sharpening of group demarcations, and disruption of political institutions. These problems would indicate a higher risk of armed conflict, and the environmental security literature is not short on reports of conflicts in resource-poor regions. The models, explanations, and particularly the assumptions of environmental security scholarship have been heavily criticized from a political ecology point of view.

^{18}

在这一领域，许多研究都参考了由供需引起的可再生资源稀缺和结构性稀缺的理论。

^{17}

根据这一学派的观点，欠发达国家最容易出现地方性资源短缺，因为它们缺乏必要的知识和能力来克服持续的资源压力。可能产生的后果包括经济生产力下降、人口迁移、群体划分加剧和政治体制混乱。这些问题表明发生武装冲突的风险较高，而环境安全文献中也不乏关于资源匮乏地区冲突的报道。从政治生态学的角度来看，环境安全学术研究的模型、解释，尤其是假设，一直饱受批评。

^{18}

Quantitative cross-national empirical research has still not converged on a systematic and robust connection between resource scarcity and civil war,

^{19}

although disaggregated studies of single countries appear somewhat more supportive.

^{20}

Yet, even if endemic scarcities do not generally cause civil war, a sudden drop in resource supply might. For example, subsistence-based populations may be forced to relocate when environmental shocks eradicate their crops or livestock. Migration may instigate violent conflict in several ways: through resource competition between newcomers and natives; through increasing polarization and hardening of socioeconomic cleavages among identity groups; and through antagonism between host authorities and the migrants’ home government caused by suspicion over the real cause or motive for displacement.

^{21}

Indeed, migration is viewed with much concern in the emerging discourse on climate security.

^{22}

Although there is some statistical evidence for a link between transnational refugee flows and the outbreak of armed conflict,

^{23}

it is far from obvious that environment-induced migration (to the extent migration can be considered monocausal) will have the same security implications. Because of a lack of conceptual clarity and a complex web of migration drivers, no empirical study has been able to explore the general consequences of “environmental migration” across multiple cases. Besides, the temporal aspect of the migration-conflict link is underdeveloped and vague, as epitomized by claims that the Sahelian drought in the 1980s caused the war in Darfur more than a decade later.

^{24}

尽管对单个国家的分类研究似乎在某种程度上更支持资源稀缺与内战之间的联系，但定量的跨国实证研究仍未就资源稀缺与内战之间系统而稳健的联系达成一致。

^{20}

然而，即使地方性的资源匮乏一般不会引发内战，但资源供应的突然减少可能会引发内战。例如，当环境冲击使农作物或牲畜绝收时，自给自足的人口可能会被迫迁移。移民可能会通过以下几种方式引发暴力冲突：新移民与本地人之间的资源竞争；身份认同群体之间的两极分化加剧和社会经济裂痕加深；以及由于对流离失所的真正原因或动机的怀疑而导致东道国当局与移民原籍国政府之间的对立。

^{21}

事实上，在新出现的气候安全讨论中，人们对移民问题非常关注。

^{22}

虽然有一些统计证据表明，跨国难民潮与武装冲突的爆发之间存在联系，但

^{23}

由环境引起的移民（在移民可被视为单一原因的情况下）会产生同样的安全影响，这一点远非显而易见。由于概念不清，移民驱动因素错综复杂，还没有任何实证研究能够探讨 "环境移民 "在多种情况下的一般后果。此外，关于移民与冲突之间联系的时间方面的研究也不够充分和模糊，例如，有人声称 20 世纪 80 年代萨赫勒地区的干旱导致了十多年后达尔富尔的战争。

^{24}

The recent economic literature on civil war provides an alternative explanation for a scarcity-conflict connection. In rain-fed agrarian societies, significant
最近关于内战的经济文献为稀缺与冲突之间的联系提供了另一种解释。在雨水充沛的农业社会，大量的

deviations from normal precipitation levels will have a negative impact on agricultural output, thereby reducing state income from taxation and exports.

^{25}

Such economic shocks are argued to heighten the risk of conflict in two ways. First, they can reduce the government’s counterinsurgent capacity and its ability to deliver public goods, thereby increasing the opportunities and incentives for dissident organizations to take up arms. Second, they can reduce individuals’ economic opportunity cost of becoming a rebel soldier. The latter argument is part of a more general theory of rebellion as individual criminal behavior, where the decision to rebel is based on calculations of expected private economic gains.

^{26}

偏离正常降水量会对农业产出产生负面影响，从而减少国家的税收和出口收入。

^{25}

这种经济冲击被认为会从两个方面增加冲突的风险。首先，它们会削弱政府的反叛乱能力和提供公共产品的能力，从而增加持不同政见组织拿起武器的机会和动机。其次，它们可以降低个人成为叛军士兵的经济机会成本。后一种论点是作为个人犯罪行为的叛乱这一更为普遍的理论的一部分，在这种理论中，叛乱的决定是基于对预期私人经济收益的计算。

^{26}

Unlike the complex causal chains normally portrayed in the environmental security literature, the economic opportunity argument yields falsifiable statistical predictions in which climatic anomalies constitute a short-term trigger effect that should exhibit a general correlational pattern with the outbreak of armed conflict. At the same time, the state-centric national economy explanation is vague on actors and motives, so it remains unclear which groups of society would be more likely to challenge the government in response to a climate-induced economic shock. In what follows, we outline a theoretical framework of the scarcity-breeds-conflict proposition by explicitly incorporating local actors and their political context. By doing so, we provide a more refined and appropriate specification of the “most likely” scenario than what has been presented in earlier empirical research.
与环境安全文献中通常描述的复杂因果链不同，经济机会论证可得出可证伪的统计预测，即气候异常构成短期触发效应，应与武装冲突的爆发呈现一般相关模式。与此同时，以国家为中心的国民经济解释在行为主体和动机方面含糊不清，因此仍不清楚哪些社会群体更有可能在气候引发的经济冲击下挑战政府。在下文中，我们将明确纳入地方行动者及其政治背景，从而勾勒出稀缺--滋生--冲突命题的理论框架。通过这样做，我们对 "最有可能 "发生的情况提供了比先前的经验研究更精细、更恰当的描述。

POLITICAL EXCLUSION AS A FACILITATING FACTOR
政治排斥是一个促进因素

Civil war is not a random process that occurs in response to adverse environmental shocks, isolated from geopolitical and cultural realms. A drought cannot cause an armed rebellion by itself, but it could trigger a latent conflict under certain conditions. If the stories portrayed in the environmental security literature represent a widespread phenomenon, the weak empirical support from the large-

N

literature might be attributed to overly aggregated research designs and a failure to specify the conditions under which environmental scarcities lead to violence.
内战不是一个孤立于地缘政治和文化领域的随机过程，它是对不利环境冲击的反应。干旱本身不会引起武装叛乱，但在某些条件下，它可能会引发潜在的冲突。如果环境安全文献中描述的故事代表了一种普遍现象，那么来自大型

N

文献的微弱实证支持可能归因于过于综合的研究设计以及未能明确环境匮乏导致暴力的条件。
Most reports of scarcity-induced conflicts concern politically and environmentally marginalized groups.

^{27}

Members of social groups without influence
关于稀缺引发冲突的大多数报告都涉及政治和环境边缘化群体。

^{27}

没有影响力的社会群体成员

over national politics share both a grievance and an identity that can be utilized in overcoming collective action problems related to mobilization.

^{28}

Colin Kahl makes this connection explicitly, arguing that the potential for demographic and environmental stress to produce violent conflict is highly contingent on the combination of exclusive political institutions and high salience of ethnic cleavages.

^{29}

Groups that are denied access to central decisionmaking processes have few peaceful means of addressing their concerns.

^{30}

Geographic concentration of the marginalized further facilitates coordination, mobilization, and conduct of violent opposition.

^{31}

Among the many economically and politically disadvantaged groups that reportedly have reacted violently in response to worsening environmental conditions are the Tuaregs in Mali, the indigenous peasants of Chiapas, Mexico, and the Bougainvilleans of Papua New Guinea.

^{32}

^{28}

科林-卡尔（Colin Kahl）明确提出了这一联系，他认为人口和环境压力产生暴力冲突的可能性在很大程度上取决于排他性政治体制和高度突出性的结合。

^{28}

科林-卡尔（Colin Kahl）明确提出了这种联系，他认为人口和环境压力产生暴力冲突的可能性在很大程度上取决于排他性政治体制和高度突出的种族裂痕的结合。

^{29}

无法参与中央决策过程的群体很少有和平手段来解决他们所关心的问题。

^{30}

边缘化群体的地理集中进一步促进了暴力反对的协调、动员和实施。

^{31}

据报道，马里的图阿雷格人、墨西哥恰帕斯州的土著农民和巴布亚新几内亚的布干维尔人等许多经济和政治上处于不利地位的群体都对日益恶化的环境状况做出了暴力反应。

^{32}

Crucially in this context, asymmetric societal structures tend to be accentuated during times of environmental crisis. A poor harvest or loss of pasture has immediate implications for local subsistence-based communities, particularly if nonagricultural sources of income are unavailable. Poor households often lack access to nonfarming economic activities and are forced to sell assets such as livestock in the face of drought.

^{33}

State-based discrimination puts further strain on a group’s resources and might have a detrimental impact on its ability to cope with a climatic shock.

^{34}

In agrarian countries, a failing harvest sea-
最重要的是，在这种情况下，不对称的社会结构在环境危机时期往往会更加突出。收成不好或失去牧场会对当地自给自足的社区造成直接影响，尤其是在没有非农业收入来源的情况下。贫困家庭往往缺乏从事非农业经济活动的机会，在干旱面前被迫出售牲畜等资产。

^{33}

基于国家的歧视进一步加重了一个群体的资源压力，并可能对其应对气候冲击的能力产生不利影响。

^{34}

在农业国家，歉收会给农民带来巨大的经济损失。

son also affects the national income and reduces the amount of funding available for redistribution. The result is often a widening wealth gap between the haves and the have-nots, between those on whom the regime depends for sustained control and those outside the sphere of political influence.

^{35}

Accordingly, acute water shortages and drought should carry the largest potential for violent dispute when they affect marginalized populations that lack resources to pursue alternative modes of living and at the same time are likely to be ignored by the central government.
儿子也会影响国民收入，减少可用于再分配的资金。结果往往是富人和穷人之间、政权赖以持续控制的人群和政治影响范围之外的人群之间的贫富差距不断扩大。

^{35}

因此，当严重缺水和干旱影响到那些缺乏资源来追求其他生活方式，同时又很可能被中央政府忽视的边缘化人群时，就最有可能引发暴力争端。
The famine in northern Ethiopia in 1984-85 illustrates how the social and political effects of drought depend on political structures and government policies. At the time of the 1984 drought, the uprising of the Tigrayan People’s Liberation Front (TPLF) had been active for nearly a decade. An essential reason why the drought turned into a famine was the Derg regime’s ruthless counterinsurgency policies, which included large-scale military offensives in surplus-producing areas, aerial bombardments of markets, and strong restrictions on trade and migration. The Derg’s policy backfired, however; the famine made people turn to the TPLF in unprecedented numbers. From 1984 to 1987, the TPLF army increased nearly threefold, and the front was turning away volunteers. This occurred mainly because ordinary Tigrayans became deeply alienated from the government, believing that they would never be free from famine as long as the junta was in power. The famine clearly made it easier for the TPLF to persuade local Tigrayans to turn against the Amharadominated state, and it also lowered the opportunity cost of joining the rebellion by making ordinary life insecure and miserable.

^{36}

1984-85 年埃塞俄比亚北部的饥荒说明了干旱的社会和政治影响如何取决于政治结构和政府政策。1984 年发生旱灾时，蒂格雷人民解放阵线（蒂格雷人阵）的起义已经持续了近十年。干旱演变成饥荒的一个重要原因是德尔格政权无情的戡乱政策，其中包括在过剩产区发动大规模军事进攻、对市场进行空中轰炸以及对贸易和移民实行严格限制。然而，德尔格的政策适得其反；饥荒使人们以前所未有的数量转向蒂格雷人民解放阵线。从 1984 年到 1987 年，蒂格雷人民解放阵线的军队增加了近三倍，而前线却将志愿者拒之门外。出现这种情况的主要原因是普通的提格雷人对政府产生了深深的隔阂，他们认为只要军政府还在掌权，他们就永远无法摆脱饥荒。饥荒显然使蒂格雷人民解放阵线更容易说服当地的蒂格雷人反对阿姆哈拉人主导的国家，而且饥荒还使普通人的生活变得不安全和悲惨，从而降低了参加叛乱的机会成本。

^{36}

Ethnopolitical exclusion is certainly not the only contextual factor that might determine whether water shortages will spark political unrest. Resilient and highly adaptive societies, which are characteristic of most developed countries, are generally well prepared to handle excessive climate variability and unexpected scarcities. Although the adaptive capacity of small farmers and pastoralists in African drylands is often underestimated,

^{37}

there are limits to experience-based knowledge in the face of more dramatic and permanent changes. For this reason, developing countries directly dependent on subsistence resources are considered particularly vulnerable to climate-induced risks.

^{38}

This is reflected in the recent empirical literature on environmental
种族政治排斥当然不是决定水资源短缺是否会引发政治动荡的唯一背景因素。大多数发达国家所特有的具有复原力和高度适应性的社会一般都做好了应对过度气候变异和意外缺水的准备。虽然非洲干旱地区小农和牧民的适应能力往往被低估，但

^{37}

面对更加剧烈和持久的变化，以经验为基础的知识是有限的。因此，直接依赖生存资源的发展中国家被认为特别容易受到气候引发的风险的影响。

^{38}

这反映在最近关于环境问题的经验文献中。

change and civil war, which focuses almost exclusively on Africa.

^{39}

Another potential catalyst of armed conflict is access to preexisting networks, such as political parties or other interest groups.

^{40}

Indeed, desperately poor people in search of earning opportunities are unlikely to be able to muster a potent rebel army without a significant contribution from entrepreneurs with organizational skills and ample funds.

^{39}

《变革与内战》一书几乎只关注非洲。

^{39}

武装冲突的另一个潜在催化剂是利用现有网络，如政党或其他利益集团。

^{40}

事实上，如果没有拥有组织技能和充足资金的企业家的大力协助，寻找赚钱机会的赤贫者不太可能组建起一支强大的反叛军。
To summarize, we argue that any measurable impact of drought on civil conflict risk is especially likely to materialize within the context of a vulnerable agricultural economy and ethnopolitical marginalization, in which the affected population is disproportionately harmed, yet deprived of nonviolent channels of political protest. In this sense, a drought might constitute the final straw leading people to rebel. A possible counterargument is that a drought could further weaken marginalized groups, thereby constituting a temporary obstacle to mobilization. Although this reasoning might have some merit, rebellion tends to be staged by aspiring elites that are well protected from sudden resource scarcities and are able to provide funds and arms whenever the timing is considered right.

^{41}

Whether rebel entrepreneurs share the underlying grievances with the affected group at large is subordinate in this context. When agrarian livelihoods dry up, prospective rebel leaders may seize the opportunity, expecting that commoners’ willingness to rebel increases with deteriorating living conditions and a lowered economic opportunity cost of fighting.
总之，我们认为，干旱对国内冲突风险的任何可衡量的影响都特别有可能在脆弱的农业经济和种族政治边缘化的背景下实现，在这种背景下，受灾人口受到不成比例的伤害，但却被剥夺了非暴力的政治抗议渠道。从这个意义上说，干旱可能是导致人们反抗的最后一根稻草。一个可能的反驳理由是，干旱可能会进一步削弱边缘化群体，从而构成动员的暂时障碍。虽然这种推论可能有一定道理，但叛乱往往是由有抱负的精英发动的，他们能够很好地保护自己免受突如其来的资源匮乏的影响，而且只要时机合适，他们就能够提供资金和武器。

^{41}

在这种情况下，反叛企业主是否与广大受影响群体有共同的基本不满情绪是次要的。当农业生计枯竭时，潜在的反叛领导人可能会抓住机会，因为他们预计平民的反叛意愿会随着生活条件的恶化和战斗的经济机会成本降低而增加。

WHY LOCATION MATTERS 为什么地点很重要

Before we move on to operational matters, a further specification of the theoretical framework is necessary. Above, we have argued that climate patterns and societies’ ability to mitigate or adapt to changes in these patterns often vary greatly within countries. A drought might not have the same short-term impact on industrialized urban centers as it has on rain-dependent subsistence communities. Similarly, when a disaster strikes, politically important segments of society are much more likely to receive aid and compensation early on than are minorities and pastoralist groups. The logical implication of the refined environmental security proposition is straightforward; a drought constitutes a threat to security and peace primarily when it overlaps with other conflict-
在我们继续讨论操作问题之前，有必要进一步说明理论框架。在上文，我们已经论证了气候模式和社会减缓或适应这些模式变化的能力在国家内部往往存在很大差异。干旱对工业化城市中心的短期影响可能与对依赖雨水生存的社区的短期影响不同。同样，当灾难发生时，社会中政治上重要的阶层比少数民族和牧民群体更有可能尽早获得援助和补偿。改良后的环境安全命题的逻辑含义很简单：干旱主要是在与其他冲突重叠时对安全与和平构成威胁。

inducing features, notably, political exclusion, economic marginalization, and cultural discrimination.
这些特征主要是政治排斥、经济边缘化和文化歧视。

Yet, a crucial question remains: Is any resulting violence likely to occur in the areas exposed to the drought? Strategic considerations might dictate that the locus of the initial riot or attack is set to maximize visibility or damage inflicted on the government. At the same time, a rebellion depends on local support to secure recruitment, funding, and shelter, as well as to establish a credible “just” cause. Consequently, popular uprisings and insurgencies are remarkably predictable in spatial terms. Even if rebel leaders occasionally emerge from cities outside the core insurgency area, rebel organizations usually issue demands that refer to local grievances and concerns as a justification for their actions.

^{42}

Besides, rebels at the outset tend to be rather few and poorly equipped, and therefore lack the capability to conduct raids in distant locations. Often, conflicts start by rebel attacks on local police or army installations and nearby communications facilities such as radio and television stations. In Chiapas, the first military encounter between the local Zapatistas and state police forces occurred in San Cristóbal, the second-largest town in Chiapas. Similarly, the 1990 Tuareg uprising in Mali began with rebel attacks on government buildings in Gao Region, the heart of Tuareg territory, whereas the longstanding conflict in West Papua started with Papuan militias attacking Indonesian army headquarters in Manokwari District, West Papua.
然而，一个关键问题依然存在：由此引发的暴力事件是否可能发生在受干旱影响的地区？出于战略考虑，最初暴乱或袭击的地点可能会尽量选择在能见度最高或对政府造成最大破坏的地方。与此同时，叛乱需要当地人的支持才能获得招募、资金和庇护，并建立起可信的 "正义 "事业。因此，民众起义和叛乱在空间上是非常可预测的。即使叛乱领导人偶尔会出现在叛乱核心地区以外的城市，叛乱组织通常也会提出要求，将当地人的不满和担忧作为行动的理由。

^{42}

此外，叛乱分子一开始往往人数较少，装备较差，因此缺乏在遥远地区发动袭击的能力。冲突的起因往往是叛军袭击了当地的警察或军队设施以及附近的通讯设施，如电台和电视台。在恰帕斯州，当地的萨帕塔主义者与州警察部队的第一次军事交锋发生在恰帕斯州第二大城镇圣克里斯托瓦尔（San Cristóbal）。同样，1990 年马里图阿雷格人起义的起因是叛军袭击了图阿雷格人领地中心加奥地区的政府大楼，而西巴布亚长期冲突的起因则是巴布亚民兵袭击了西巴布亚马诺夸里区的印尼军队总部。

That said, we do not disregard possible insecurity effects of drought that work through national-level mechanisms and that might have nationwide implications.

^{43}

Rather, weakening of the state economy and institutions, disintegration of national unity, and, potentially, state collapse should be seen as consequences of more immediate, local developments. This is evident in the current climate-security discourse, in which most articulated warnings and speculations center on the vulnerability and resilience of local populations.

^{44}

尽管如此，我们并不忽视干旱可能通过国家层面的机制产生的不安全影响，这些影响可能会波及全国。

^{43}

相反，国家经济和机构的削弱、民族团结的瓦解以及潜在的国家崩溃应被视为更直接的地方发展的后果。这一点在当前的气候安全讨论中显而易见，其中大多数明确提出的警告和猜测都以当地居民的脆弱性和复原力为中心。

^{44}

HYPOTHESES 假设

The general argument outlined above implies that increasing water scarcity and drought should have the most immediate impact on local communities.
上文概述的一般论点意味着，日益严重的缺水和干旱应该对当地社区产生最直接的影响。

Moreover, a careful reading of relevant narratives suggests that a conflictinducing effect of drought should be most pronounced when the drought strikes a marginalized, peripheral population. This yields the following testable propositions:
此外，对相关叙述的仔细解读表明，当干旱袭击边缘化的边缘人口时，干旱的冲突诱发效应应该最为明显。由此可以得出以下可检验的命题：
H1. Drought increases the local risk of civil war. H2. The effect of drought on civil war risk is larger in politically marginalized areas.
H1.干旱会增加当地发生内战的风险。H2.在政治边缘化地区，干旱对内战风险的影响更大。

Research Design, Data, and Measures
研究设计、数据和措施

Extant large-

N

studies of the general scarcity-conflict nexus offer inconclusive results.

^{45}

Recently, a new wave of research addresses the climate-conflict link explicitly by studying short-term implications of climatic (precipitation, temperature) anomalies for civil war either directly

^{46}

or via poor economic growth

^{47}

Although these studies reach widely different conclusions, they represent a significant step forward because they focus on rapid environmental changes rather than on static conditions that shaped the first generation of relevant comparative research.

^{48}

At the same time, they rely on a habitual country-level design and hence suffer from the same aggregation problem as earlier work. Such analyses may capture possible effects of drought that work through state institutions or have nationwide impacts (e.g., via displacement), but they are likely to miss or discount more immediate effects at the local level. Both climate and underlying factors that affect the latent risk of conflict (e.g., ethnopolitical exclusion, poverty, economic discrimination, population size, and settlement patterns) often display considerable variation within countries, so aggregating data by country may introduce substantial measurement error and, possibly, ecological fallacy.

^{49}

N

对一般匮乏与冲突之间关系的现有大型研究没有得出结论。

^{48}

同时，这些研究依赖于惯常的国家级设计，因此存在与早期研究相同的汇总问题。这些分析可能会捕捉到干旱可能通过国家机构产生的影响或对全国产生的影响（如通过流离失所），但它们很可能会忽略或不考虑在地方层面产生的更为直接的影响。气候和影响冲突潜在风险的潜在因素（如种族政治排斥、贫困、经济歧视、人口规模和定居模式）在国家内部往往存在很大差异，因此按国家汇总数据可能会带来很大的测量误差，甚至可能导致生态谬误。

^{49}

To remedy this shortcoming, we developed a spatially disaggregated dataset
为了弥补这一不足，我们开发了一个空间分类数据集

for Africa, 1960-2004, derived from the PRIO-GRID database.

^{50}

African economies are largely dependent on rain-fed agriculture, and the continent’s overall high environmental vulnerability, weak institutional coping capacity, a predominantly rural population, and colonial legacy imply that our analytical design implicitly incorporates a number of important scope conditions.

^{51}

The employed grid structure comes with a resolution of

0.5 \times 0.5

decimal degrees, which corresponds to a cell size of roughly

50 \times 50

kilometers. This resolution is high enough to ensure that even small countries such as Rwanda and Djibouti are represented by multiple cells, and sufficiently coarse to keep the size of the overall dataset at a manageable level. The African grid contains more than 10,000 unique cells per cross section. Each cell was observed once per year since 1960 or the year of independence of the country to which the cell belongs.

^{52}

We excluded grid cells with little or no measurable amount of precipitation in a normal year (less than 100 millimeters); these areas are mostly uninhabited and, by definition, never experience drought.

^{53}

The complete panel dataset includes about 350,000 valid observations.
从 PRIO-GRID 数据库中提取的非洲 1960-2004 年的数据。

^{50}

非洲经济在很大程度上依赖于雨水灌溉的农业，非洲大陆整体环境的高度脆弱性、薄弱的机构应对能力、以农村人口为主以及殖民时期的遗留问题，都意味着我们的分析设计隐含着许多重要的范围条件。

^{51}

采用的网格结构的分辨率为

0.5 \times 0.5

十进制度，相当于大约

50 \times 50

公里的单元格大小。这一分辨率足以确保卢旺达和吉布提这样的小国也能通过多个单元格来表示，同时也足够粗略，可将整个数据集的大小保持在可管理的水平。非洲网格每个横截面包含 10,000 多个独特的单元。自 1960 年或单元格所属国家独立以来，每年对每个单元格观测一次。

^{52}

我们排除了在正常年份降水量很少或没有可测量降水量（少于 100 毫米）的网格单元；这些地区大多无人居住，顾名思义，从未经历过干旱。

^{53}

完整的面板数据集包括约 350,000 个有效观测值。
The disaggregated research design offers a number of improvements over conventional country-level analyses; yet, there are alternative approaches to the grid structure. Perhaps the most appealing option would be to construct the data around politically relevant social groups.

^{54}

After all, organized political violence is conducted by groups of people, not by individuals or artificial grid cells. At the same time, to capture the great within-country variation in precipitation, one would have to choose spatially distinct groups. Although there are good georeferenced ethnicity data available (see below), we decided
与传统的国家级分析相比，分类研究设计有许多改进之处；然而，网格结构也有其他方法可供选择。最有吸引力的方案或许是围绕与政治相关的社会群体构建数据。

^{54}

毕竟，有组织的政治暴力是由群体而非个人或人为网格单元实施的。同时，为了捕捉国家内部降水量的巨大差异，我们必须选择空间上不同的群体。虽然有很好的地理参照种族数据（见下文），但我们决定
‘Disaggregating Civil War,’" Journal of Conflict Resolution, Vol. 53. No. 4 (August 2009), pp. 487-495, and subsequent articles in the same issue.
内战分解"，《解决冲突期刊》，第 53 卷，第 4 期（2009 年 8 月），第 487-495 页。4 (August 2009), pp.
50. Andreas Forø Tollefsen, Håvard Strand, and Halvard Buhaug, “PRIO-GRID: A Unified Spatial Data Structure,” Journal of Peace Research, Vol. 49, No. 2 (March 2012).
50.Andreas Forø Tollefsen、Håvard Strand 和 Halvard Buhaug，"PRIO-GRID：统一的空间数据结构"，《和平研究期刊》，第 49 卷，第 2 期（2012 年 3 月）。
51. In slightly more technical jargon, the research design assumes a fixed-effects approximation by implicitly accounting for a number of (near) continent-wide features that may plausibly affect the environment’s impact on civil war risk.
51.用稍微专业一点的术语来说，研究设计假定了一种固定效应近似值，隐含地考虑了一些（接近）全大陆的特征，这些特征可能会合理地影响环境对内战风险的影响。
52. To allow attaching country-specific features to the grid cells, we assigned each cell to one country only. Cells overlapping international borders are assigned to the country with the largest geographic share. For earlier grid-based analyses of civil war, see Halvard Buhaug and Jan Ketil Rød, “Local Determinants of African Civil Wars, 1970-2001,” Political Geography, Vol. 25, No. 3 (March 2006), pp. 315-335; and Clionadh Raleigh and Henrik Urdal, “Climate Change, Environmental Change, and Armed Conflict,” Political Geography, Vol. 26, No. 6 (August 2007), pp. 674694.
52.为了在网格单元中附加国家特征，我们将每个单元只分配给一个国家。与国际边界重叠的单元格分配给地理份额最大的国家。有关早期基于网格的内战分析，请参见 Halvard Buhaug 和 Jan Ketil Rød，"1970-2001 年非洲内战的地方决定因素"，《政治地理学》，第 25 卷，第 3 期（2006 年 3 月），第 315-335 页；以及 Clionadh Raleigh 和 Henrik Urdal，"气候变化、环境变化和武装冲突"，《政治地理学》，第 26 卷，第 6 期（2007 年 8 月），第 674694 页。
53. A desert is commonly defined as an arid climate with less than 250 millimeters annual precipitation. See the Köppen climate classification as described in Tom L. McKnight and Darrel Hess, Physical Geography: A Landscape Appreciation, 8th ed. (Upper Saddle River, N.J.: Prentice Hall, 2004). 54. See, for instance, Halvard Buhaug, Lars-Erik Cederman, and Jan Ketil Rød, “Disaggregating Ethno-Nationalist Civil Wars: A Dyadic Test of Exclusion Theory,” International Organization, Vol. 62, No. 3 (July 2008), pp. 531-551.
53.沙漠通常被定义为年降水量少于 250 毫米的干旱气候。参见 Tom L. McKnight 和 Darrel Hess 合著的《自然地理》第 8 版中描述的柯本气候分类：(Upper Saddle River, N.J.: Prentice Hall, 2004)。54.例如，见 Halvard Buhaug、Lars-Erik Cederman 和 Jan Ketil Rød，"Disaggregating Ethno-Nationalist Civil Wars: A Dyadic Test of Exclusion Theory，"International Organization，Vol.62，No.3 (July 2008)，pp.531-551。
not to let ethnic groups constitute our units of analysis, for two reasons. First, the size of ethnic group territories varies greatly, and some are large enough to introduce some of the aggregation problems that we seek to avoid. Second, focusing on ethnic groups would limit our dependent variable to ethnically based civil wars. An alternative approach would be to focus on first- or second-order administrative entities.

^{55}

Again, the heterogeneity in size is problematic. Besides, administrative boundaries may change over time, and their function varies among countries-sometimes in response to past conflict or anticipation of future unrest. To our knowledge, there is no publicly available time-series data on subnational political entities. A seeming drawback of the grid structure-its complete ignorance of sociopolitical entities and boundaries-is also a significant strength: the grid cells do not change over time; their size and outline are exogenous to past conflict and the political context; they may be aggregated or further disaggregated in a consistent manner; and they are immediately comparable across countries.
不以族群为分析单位有两个原因。首先，族群领地的大小差异很大，有些族群领地大到足以带来一些我们试图避免的汇总问题。其次，将重点放在族群上会使我们的因变量局限于以族群为基础的内战。另一种方法是关注一阶或二阶行政实体。

^{55}

同样，规模的异质性也是个问题。此外，行政边界可能会随着时间的推移而改变，其功能也因国家而异--有时是对过去冲突的回应，有时是对未来动乱的预期。据我们所知，目前还没有关于国家以下各级政治实体的公开时间序列数据。网格结构的一个看似缺点--对社会政治实体和边界的完全忽略--也是一个重要的优点：网格单元不会随时间而改变；它们的大小和轮廓与过去的冲突和政治背景无关；它们可以以一致的方式汇总或进一步分解；它们在不同国家之间具有直接可比性。

The dependent variable in all model specifications is onset of civil armed conflict, as defined by the UCDP/PRIO Armed Conflict Dataset.

^{56}

We excluded coups d’état from the reported models; intragovernment conflicts are distinct from disputes pitting state against nonstate actors, and they are less relevant from a theoretical point of view.

^{57}

To facilitate the spatially disaggregated analysis, we pinpointed all conflicts to the exact onset location (i.e., the locality of fighting on the first day of the conflict) and assigned corresponding geographical coordinates. The onset cells are coded 1 in the initial year of a new armed conflict, as well as in the first year of conflict recurrence after a peaceful intermittency of at least two calendar years. The explanatory factor of interest in this study is drought, operationalized as negative precipitation deviation. In Africa, annual rainfall is a good proxy for freshwater availability (as mentioned earlier, only 4 percent of arable land is irrigated), and precipitationgenerated measures of drought are attractive given their exogenous relationship with human/societal behavior (in contrast to agricultural output, food prices, calls for international assistance, and other nonmeteorological proxies for drought).

^{58}

Georeferenced precipitation data were derived from the Global Precipitation Climatology Centre of the UN-sponsored World Meteorological
所有模型中的因变量都是 UCDP/PRIO 武装冲突数据集定义的国内武装冲突的爆发。

^{56}

我们在报告的模型中排除了政变；政府内部冲突有别于国家与非国家行为者之间的争端，而且从理论角度来看，它们的相关性较低。

^{57}

为了便于进行空间分类分析，我们将所有冲突精确定位到冲突开始的地点（即冲突第一天发生战斗的地点），并分配了相应的地理坐标。在新的武装冲突发生的第一年，以及在和平间歇至少两年后冲突再次发生的第一年，起始点单元的编码均为 1。本研究关注的解释因素是干旱，即降水量的负偏差。在非洲，年降水量是淡水可用性的良好替代指标（如前所述，只有 4% 的耕地得到灌溉），降水量产生的干旱指标与人类/社会行为的外生关系（与农业产出、粮食价格、国际援助呼吁和其他非气象干旱替代指标相反），因此具有吸引力。

^{58}

地理参考降水数据来自联合国赞助的世界气象组织的全球降水气候学中心。

Organization (GPCC)

^{59}

The GPCC data contain annual gauge-based estimates of total precipitation (millimeter) for the global land surface at

0.5 \times 0.5

degree resolution for all years, 1951-2004. From the GPCC data, we constructed several alternative measures of drought, all calculated specifically for each cell year. The simplest indicator is a measure of proportional change, or growth, in annual rainfall

(R)

since the previous year

(R_{i t} - R_{i t - 1}) / R_{i t - 1})^{60}

We also included a variable that captures rainfall anomaly, or the relative deviation (%) from the long-term average annual precipitation in the cell. In addition, we tested a binary drought indicator from the United Nations Environment Programme (UNEP/GRID Europe), based on precipitation data from the Climate Research Unit at the University of East Anglia. The UNEP data are available only for the period 1980-2001.

^{59}

GPCC 数据包含 1951-2004 年全球陆地表面总降水量（毫米）的年度估算值，分辨率为

0.5 \times 0.5

度。根据 GPCC 数据，我们构建了几种可供选择的干旱度量指标，这些指标都是专门为每个单元年计算的。最简单的指标是衡量自前一年

(R_{i t} - R_{i t - 1}) / R_{i t - 1})^{60}

以来年降雨量

(R)

的比例变化或增长情况，我们还加入了一个变量来捕捉降雨异常，即与小区长期平均年降雨量的相对偏差（%）。此外，我们还测试了联合国环境规划署（UNEP/GRID Europe）的二元干旱指标，该指标基于东英吉利大学气候研究室的降水数据。联合国环境规划署仅提供 1980-2001 年期间的数据。

Aggregated data on annual rainfall correlate well with severe drought events; yet, such statistics could potentially mask drought events caused by the rain falling at the wrong time of the year. Hence, our preferred operationalization of drought is a more comprehensive measure that captures intraannual deviations from normal precipitation. The Standard Precipitation Index, SPI6, is the measure that corresponds best to nonmeteorological drought statistics, such as the EM-DAT database by the Centre for Research on the Epidemiology of Disasters. For each month, the SPI6 index measures negative deviation from normal rainfall during the six preceding months. The values are standardized where SPI6 estimates below 1 indicate near normal rainfall (or wetter); 1 to 1.49 indicate moderately dry; 1.5 to 1.99 indicate severely dry; and values in excess of 2 standard deviations indicate extremely dry. The monthly data are then aggregated to a yearly format and categorized to indicate drought events. The annualized SPI6 is coded 1 if there were at least three consecutive months with SPI6

\geq 1

in the given grid cell during the year (moderate drought); 1.5 if the SPI6 was

\geq 1.5

for at least two consecutive months (severe drought); and 2.5 if both of the above criteria are met (extreme drought). If no drought event occurred during the year, the SPI6 is assigned the value 0 . To ease interpretation and because the categorized SPI6 variable is an ordinal rather than interval scale, we employ a dichotomous variant of the SPI6 where all positive values are recoded 1. All drought measures are created
年降雨量的综合数据与严重干旱事件有很好的相关性；然而，这些统计数据可能会掩盖因降雨时间不对而导致的干旱事件。因此，我们更倾向于用一种更全面的方法来衡量干旱，这种方法能捕捉到年内降水量与正常降水量的偏差。标准降水指数 SPI6 是最符合非气象干旱统计数据的衡量标准，例如灾害流行病学研究中心的 EM-DAT 数据库。SPI6 指数衡量每个月前 6 个月与正常降雨量的负偏差。数值经过标准化处理，SPI6 估计值低于 1 表示降雨量接近正常值（或较湿）；1 至 1.49 表示中度干旱；1.5 至 1.99 表示严重干旱；数值超过 2 个标准差表示极端干旱。然后将月度数据汇总为年度格式，并进行分类，以显示干旱事件。如果给定网格单元中的 SPI6 在年内至少连续三个月

\geq 1

，则年化 SPI6 的编码为 1（中度干旱）；如果 SPI6 至少连续两个月

\geq 1.5

，则编码为 1.5（严重干旱）；如果同时满足上述两个标准，则编码为 2.5（极端干旱）。如果当年没有发生干旱事件，则 SPI6 值为 0。为了便于解释，而且由于分类 SPI6 变量是一个序数而非区间量表，我们采用了 SPI6 的二分变量，将所有正值重新编码为 1。

in multiple variants to account for various time lags, reference periods, and cumulative scores. In addition, we include a spatially lagged SPI term that gives the cell-specific distance

(\log km)

to the nearest drought event to capture potential displacement effects.

^{61}

有多种变体，以考虑不同的时间滞后、参照期和累积分数。此外，我们还加入了一个空间滞后的 SPI 项，它给出了小区到最近干旱事件的距离

(\log km)

，以捕捉潜在的位移效应。

^{61}

Spatial data on political marginalization of ethnic groups is generated from two recent and compatible data projects, the Geo-Referencing of Ethnic Groups (GREG) dataset and the Ethnic Power Relations (EPR) dataset.

^{62}

The GREG project stores information on all significant ethnic groups in a geographic information systems (GIS) format, where each group is represented by one or more polygons that denote their settlement area. To obtain groupspecific information on political status, we then linked the group polygons to the political status data from EPR. Note that EPR, in contrast to the GREG project, identifies politically relevant ethnic groups, so in some instances a GREG group was dropped or had to be merged with other groups to match the more inclusive delineation of the EPR data.
有关族群政治边缘化的空间数据来自两个最新且兼容的数据项目，即族群地理参照（GREG）数据集和族群权力关系（EPR）数据集。

^{62}

GREG项目以地理信息系统（GIS）的格式存储了所有重要族群的信息，每个族群由一个或多个多边形表示，这些多边形表示他们的居住区。为了获得特定群体的政治地位信息，我们将群体多边形与 EPR 的政治地位数据进行了链接。需要注意的是，EPR 与 GREG 项目不同，它确定的是与政治相关的族群，因此在某些情况下，GREG 族群被删除或必须与其他族群合并，以便与 EPR 数据中更具包容性的划分相匹配。

When adapted to the grid structure, each cell is coded in accordance with the status of the majority group within the corresponding geographic area. The variable for marginalized ethnic groups (MEG) takes on the value 1 in cellyears where the majority group in the cell is excluded from central governmental positions.

^{63}

Reflecting the salience of ethnicity in African politics, a majority of observations contain excluded populations; only about one-third (36 percent) of the cell-years host ethnic groups in power (EGIPs). The MEG indicator is time-variant and captures important shifts in the national political structure of countries. In contrast to the strictly exogenous drought measures, however, the political status of ethnic groups may to some extent be a function of past power struggles as well as the regime’s anticipation of future aspirations. The nature of the EPR data does not allow us to distinguish among marginalized groups based on the reason for their exclusion. We do, however,
在与网格结构相适应时，每个单元格都根据相应地理区域内多数群体的地位进行编码。如果单元格中的多数族群被排除在中央政府职位之外，则边缘化族群变量（MEG）的值为 1。

^{63}

这反映了种族问题在非洲政治中的突出地位，大多数观察结果都包含了被排斥的人口；只有大约三分之一（36%）的单元年包含了掌权的种族群体（EGIPs）。MEG 指标是随时间变化的，它捕捉了国家政治结构的重要变化。然而，与严格的外生干旱指标相比，族群的政治地位在某种程度上可能是过去权力斗争以及政权对未来期望的预期的函数。由于 EPR 数据的性质，我们无法根据边缘化群体被排斥的原因对其进行区分。不过，我们可以这样做、
61. Recent case research suggests that migration in response to rapid environmental change tends to take the form of short-distance displacement, often to the nearest urban center. See contributions to the special issue of Population and Environment (December 2010) on “Human Migration and the Environment.”
61.最近的案例研究表明，为应对快速环境变化而进行的移徙往往采取短距离迁移的形式，通常是迁移到最近的城市中心。参见为《人口与环境》特刊（2010 年 12 月）"人类迁徙与环境 "撰写的文章。
62. A presentation of the GREG data can be found at Nils B. Weidmann, Jan Ketil Rød, and LarsErik Cederman, “Representing Ethnic Groups in Space: A New Dataset,” Journal of Peace Research, Vol. 47, No. 4 (July 2010), pp. 491-499. The EPR data are described in Andreas Wimmer, Lars-Erik Cederman, and Brian Min, “Ethnic Politics and Armed Conflict: A Configurational Analysis of a New Global Dataset,” American Sociological Review, Vol. 74, No. 2 (April 2009), pp. 316-337. Note that since work on this article was concluded, a georeferenced version of the EPR data has been released: Julian Wucherpfennig, Nils B. Weidmann, Luc Girardin, Lars-Erik Cederman, and Andreas Wimmer, “Politically Relevant Ethnic Groups across Space and Time,” Conflict Management and Peace Science, in press.
关于 GREG 数据的介绍见 Nils B. Weidmann, Jan Ketil Rød, and LarsErik Cederman, "Representing Ethnic Groups in Space：新数据集》，《和平研究杂志》，第 47 卷，第 4 期（2010 年 7 月），第 491-499 页。EPR 数据在 Andreas Wimmer、Lars-Erik Cederman 和 Brian Min 合著的 "Ethnic Politics and Armed Conflict：美国社会学评论》，第 74 卷，第 2 期（2009 年 4 月），第 316-337 页。请注意，自本文完成后，EPR 数据的地理参照版本已经发布：Julian Wucherpfennig, Nils B. Weidmann, Luc Girardin, Lars-Erik Cederman, and Andreas Wimmer, "Politically Relevant Ethnic Groups across Space and Time," Conflict Management and Peace Science, in press.
63. For a similar GIS-based application of the ethnicity data, see Buhaug et al., “Disaggregating Ethno-Nationalist Civil Wars.”
63.类似的基于地理信息系统的种族数据应用，见 Buhaug 等人，"Disaggregating Ethno-Nationalist Civil Wars"。
apply a one-year time lag to the MEG indicator to limit the potential for reverse causality, and as a sensitivity test, we exclude subsequent civil wars for areas with a history of conflict, with little discernible impact on the results.
作为敏感性测试，我们排除了有冲突历史的地区随后发生的内战，但对结果影响不大。

Contextual factors are controlled for by the following: (1) cell- and yearspecific estimates of (log) population size, derived from the Gridded Population of the World, version 3, data;

^{64}

(2) a dummy marking off cells that contain the capital city; (3) distance to the nearest international border (log km); as well as country-level controls for (4) democracy, represented by the Polity 2 index of the Polity IV project;

^{65}

(5) infant mortality rate from the UN’s Population Division, supplemented by data from Henrik Urdal;

^{66}

and (6) conflict history, represented by a decay function of the time since the previous conflict in the country, using a half-life parameter of two years.

^{67}

环境因素由以下因素控制：(

^{64}

(2)标记包含首都的单元的虚拟值；(3)到最近的国际边界的距离（对数公里）；以及(4)民主的国家级控制，由 Polity IV 项目的 Polity 2 指数表示；

^{65}

(5)联合国人口司的婴儿死亡率，由 Henrik Urdal 提供的数据作为补充；

^{66}

(6)冲突历史，由该国上次冲突以来时间的衰减函数表示，使用的半衰期参数为两年。

^{67}

All results reported below are based on regressions on a reduced sample to limit computation time and to avoid spatial autocorrelation.

^{68}

The subsample contains all onset cell-years and a randomly drawn selection of 5 percent of the nonconflict observations, 1960-2004. The case control design is an efficient procedure to study excessively large datasets where most observations carry little information, and the logit estimator is unaffected by the relative share of 1 s and 0 s in the sample. For sensitivity tests, we employed rare events logit (relogit) and estimated the models on the full sample and on alternative, randomly drawn subsamples, all of which yielded virtually identical results to those reported here. Lastly, to better account for unit heterogeneity and correct
以下报告的所有结果都是基于缩小样本的回归结果，以限制计算时间并避免空间自相关性。

^{68}

子样本包含所有的发病单元年和随机抽取的 5%的非冲突观测值（1960-2004 年）。病例控制设计是一种有效的方法，可用于研究大多数观测值信息量很小的超大数据集，而且对数估计器不受样本中 1 s 和 0 s 的相对比例的影响。为了进行敏感性测试，我们采用了罕见事件 logit（relogit），并在全部样本和随机抽取的其他子样本上对模型进行了估计，所有结果与本文报告的结果几乎完全相同。最后，为了更好地考虑单位异质性并纠正

Figure 2. Average Annual Precipitation in Africa, 1952-2004
图 2.1952-2004 年非洲年平均降水量

NOTE: The map shows the spatial distribution of precipitation across Africa, based on highresolution data from the Global Precipitation Climatology Centre of the UN-sponsored World Meteorological Organization. White areas are deserts with less than 100 millimeters precipitation per year, whereas the wettest parts of Africa recieve more than 3,500 millimeters of rain in a normal year.
注：根据联合国赞助的世界气象组织全球降水气候学中心提供的高分辨率数据，该地图显示了整个非洲的降水空间分布情况。白色区域为沙漠，年降水量不足 100 毫米，而非洲最潮湿的地区正常年份的降水量超过 3500 毫米。
for possible omitted variable bias, we reestimated all models with country fixed effects. The results do not change.

^{69}

考虑到可能存在遗漏变量偏差，我们用国家固定效应对所有模型进行了重新估计。结果没有变化。

^{69}

Results 成果

Climatic conditions, including rainfall patterns and temperature, vary considerably within most countries, and Africa is no exception. Figure 2 illustrates the distribution of normal annual precipitation across the continent. Evidently, large parts of the Sahara and Namib deserts have virtually no precipitation.
大多数国家的气候条件，包括降雨模式和温度，都有很大差异，非洲也不例外。图 2 展示了非洲大陆正常年降水量的分布情况。显而易见，撒哈拉沙漠和纳米布沙漠的大部分地区几乎没有降水。

Other regions are characterized by a tropical climate with distinct wet and dry seasons and where any drought is relative. More generally, figure 2 illustrates how climate zones tend to cross, rather than follow, national boundaries, thereby effectively pointing to the limitations of using aggregate statistics of country climate.
其他地区则属于热带气候，干湿季节分明，任何干旱都是相对的。更广泛地说，图 2 说明了气候区往往跨越而不是沿国界分布，从而有效地指出了使用国家气候综合统计数据的局限性。

An initial assessment of the hypotheses offers little support for the propositions. Table 1 shows a trivariate cross-tabulation of drought, ethnopolitical status, and onset of civil war. With such a high-resolution research design and point data for conflict outbreak, there is no surprise that civil war is a rare event indeed. Yet, conflict is not distributed completely at random. About onethird of all observations are represented by ethnic groups in power, whereas only 9 of 59 conflicts ( 15 percent) break out in these areas. Drought exhibits a much weaker covariation with civil war. The conflict rate among the drought observations is 0.014 percent; the corresponding rate for nondrought observations is 0.018 percent (i.e., slightly higher). Still, the lower right cell, which represents the interaction between exclusion and drought, provides tentative, if weak, support for hypothesis 2. The relative increase in the proportion of onsets between EGIP observations and MEG observations is twice as large for drought observations as for non-drought observations. The limited number of conflict onsets, however, implies that these differences are not statistically significant with a reasonable margin of error.
对假设的初步评估显示，这些命题几乎没有得到支持。表 1 显示了干旱、民族政治地位和内战爆发的三变量交叉表。有了如此高分辨率的研究设计和冲突爆发的点数据，内战确实是罕见事件也就不足为奇了。然而，冲突并不是完全随机分布的。大约三分之一的观察结果是由少数民族掌权，而 59 起冲突中只有 9 起（15%）是在这些地区爆发的。干旱与内战的协变关系要弱得多。干旱观测值的冲突发生率为 0.014%；非干旱观测值的相应发生率为 0.018%（即略高）。尽管如此，代表排斥与干旱之间交互作用的右下方单元格还是为假设 2 提供了初步支持（尽管很微弱）。在 EGIP 观测值与 MEG 观测值之间，干旱观测值的冲突发生率相对增加的比例是非干旱观测值的两倍。然而，由于冲突发生的数量有限，这些差异在合理误差范围内并不具有统计意义。

Next, we ran a series of multivariate regression models. Table 2 presents three models: the baseline model with controls only (model 1), a model that adds a measure of local drought (model 2), and a model with the interaction between political exclusion and drought (model 3). In line with earlier research, model 1 shows that conflict is significantly more likely in areas dominated by marginalized ethnic groups. Proximity to the border, the capital city, high population density, and high infant mortality rate are other factors associated with a significantly higher risk of civil war onset. Although the result for local population density would seem to support notions of population pressure and demographic stress,

^{70}

we interpret this finding as a consequence of two phenomena: the tendency for local governmental bodies and assetstypical targets of initial rebel assaults-to be located in urban centers such as towns and regional capitals, and the “urban bias” in media coverage. The coefficient for democracy is small and not significantly different from zero.
接下来，我们运行了一系列多元回归模型。表 2 列出了三个模型：仅包含控制因素的基线模型（模型 1）、增加了当地干旱衡量指标的模型（模型 2）以及政治排斥与干旱交互作用的模型（模型 3）。与之前的研究结果一致，模型 1 显示，在边缘化族群占主导地位的地区，冲突发生的可能性明显增大。靠近边境、首都、高人口密度和高婴儿死亡率等因素也与内战爆发的风险显著增加有关。尽管当地人口密度的结果似乎支持人口压力和人口紧张的概念，

^{70}

我们将这一结果解释为两种现象的结果：当地政府机构和资产通常是叛军最初攻击的目标，这些机构和资产往往位于城镇和地区首府等城市中心，以及媒体报道中的 "城市偏见"。民主系数较小，与零无显著差异。

Model 2 introduces our preferred drought measure, the SPI dummy that captures within-year as well as between-year precipitation anomalies. Con-
模型 2 引入了我们首选的干旱衡量指标 SPI 虚拟指标，该指标可捕捉年内和年际降水异常。Con-

Table 1. Cross Tabulation of Drought, Political Exclusion, and Civil War Onset
表 1.干旱、政治排斥和内战爆发的交叉表

	Ethnic Group in Power (EGIP) 掌权族群（EGIP）		Marginalized Ethnic Group (MEG) 边缘化族裔群体（MEG）
	No Drought 没有干旱	Drought 干旱	No Drought 没有干旱	Drought 干旱
No onset 没有发病	98,277	22,844	186,079	42,855
No onset 没有发病	$(99.992 %)$	$(99.997 %)$	$(99.977 %)$	$(99.981 %)$
Onset 开始	8	1	42	8
	$(0.008 %)$	$(0.003 %)$	$(0.023 %)$	$(0.019 %)$
Total 总计	98,285	22,845	186,121	42,863
	$(100 %)$	$(100 %)$	$(100 %)$	$(100 %)$

NOTE: Relative frequencies by column are given in parentheses. EGIP represents cells where the local population is part of the national government; MEG observations are excluded from the national government. Drought is represented by the annualized, dichotomous SPI 6 statistic. SPI stands for Standardized Precipitation Index. See data section for further details.
注：括号内为各栏的相对频率。EGIP 表示当地人口是国家政府一部分的单元格；MEG 观察值不包括国家政府。干旱以年化二分法 SPI 6 统计量表示。SPI 指标准化降水指数。详见数据部分。
trary to hypothesis 1, the SPI parameter estimate has a negative sign, suggesting a slightly lower risk of conflict onset in areas with drought. The coefficient comes with a large standard error, however, so one should be careful about drawing any conclusion regarding the direction of the relationship. Other testimonies to the trivial impact of drought on conflict risk are the virtually unaffected log likelihood and the similar coefficients for the other variables in model 2 compared to the baseline model.

^{71}

与假设 1 相反，SPI 参数估计为负号，表明干旱地区发生冲突的风险略低。不过，该系数的标准误差较大，因此在对两者关系的方向下结论时应小心谨慎。干旱对冲突风险影响微不足道的其他证明是，与基线模型相比，模型 2 中其他变量的对数可能性几乎不受影响，系数也相似。

^{71}

Model 3 offers an empirical evaluation of the proposed conditional effect of drought (hypothesis 2) by including an interaction term between SPI and the dichotomous measure of political exclusion, MEG. Again, the results counter the notion that acute water shortages increase the local risk of violent conflict. Instead, the location of civil war in Africa can be explained by generic political, socioeconomic, and geographic factors: a politically marginalized local population, high local population density, the country capital, proximity to an international border, and high infant mortality.
模型 3 通过加入 SPI 与政治排斥的二分衡量指标 MEG 之间的交互项，对所提出的干旱条件效应（假设 2）进行了实证评估。结果再次反驳了严重缺水会增加当地暴力冲突风险的观点。相反，非洲内战的发生地可以用一般的政治、社会经济和地理因素来解释：政治边缘化的当地人口、当地人口密度高、国家首都、靠近国际边境以及婴儿死亡率高。

Overall, the results presented in table 2 provide little evidence that popular narratives of drought-induced conflict represent a widespread phenomenon. Even though the analysis has been designed to capture the conditions most conducive to violence (agriculture-dependent poor, local political exclusion) and thereby constitutes a more appropriate large-

N

test of environmental security arguments than earlier investigations, there are nevertheless reasons to caution against a hasty rejection of the hypotheses. Most critically, the results
总体而言，表 2 中的结果没有提供多少证据表明干旱引发冲突的流行说法是一种普遍现象。尽管该分析旨在捕捉最容易引发暴力的条件（依赖农业的贫困人口、地方政治排斥），因此与之前的调查相比，它更适合对环境安全论点进行大规模

N

检验，但仍有理由告诫我们不要草率地否定这些假设。最关键的是，结果

Table 2. Logit Regression of Local Civil War Onset, 1960-2004
表 2.1960-2004 年地方内战爆发的对数回归

Variables 变量	Model 1 模型 1	Model 2 模型 2	Model 3 模型 3
Drought (SPI) 干旱（SPI）		-0.303	-0.419
		$(0.333)$	$1.095)$
Marginalized ethnic 边缘化族裔	${1.394}^{*}$	${1.407}^{*}$	${1.390}^{*}$
group 组	$(0.426)$	$(0.428)$	$(0.460)$
Marginalized ethnic 边缘化族裔			0.133
group $\times$ Drought 组 $\times$ 干旱		$(1.148)$
Distance to border (In) 到边界的距离（英寸）	$- {0.268}^{*}$	$- {0.268}^{*}$
	$(0.100)$	$(0.099)$
Capital city 首都	${1.947}^{*}$	$({0.099}^{*}$	${1.971}^{*}$

Climate Wars? 气候战争？

Climate change is hot. 气候变化是热门话题。

Backdrop: Trends in Precipitation Patterns背景：降水模式趋势

Linking Drought to Civil War将干旱与内战联系起来

POLITICAL EXCLUSION AS A FACILITATING FACTOR政治排斥是一个促进因素