Understanding the risk of armed conflict is essential for promoting peace. Although the relationship between climate variability and armed conflict has been studied by the research community for decades with quantitative and qualitative methods at different spatial and temporal scales, causal linkages at a global scale remain poorly understood. Here we adopt a quantitative modelling framework based on machine learning to infer potential causal linkages from high-frequency time-series data and simulate the risk of armed conflict worldwide from 2000–2015.
Research on the climate–conflict connection covers a wide range of climate phenomena as well as conflict dimensions which makes the diverse outcomes of the different studies difficult to compare. Although scientists have yet to agree on the causal climate–conflict connections, there is an increasing acceptance that climate change or changes in climate variability increase the risk of armed conflict in certain circumstances. Exploring the causal climate–conflict linkages at the global scale is still a challenging task. In recent years, simulation- and data-driven approaches (named machine learning) have been proven to have the potential to solve many complex problems based on large amounts of data, including climate-conflict linkages. Therefore, we propose the potentially tractable question of whether the machine-learning approach could be used to discover patterns between conflict risk and high-dimensional covariates.
In this research, we combined a machine-learning approach with high-frequency time-series data to model armed conflict risk under climate change. We proposed a hypothesis that where such patterns exist, a machine-learning model fitted from a single-year dataset should have a certain ability in predicting armed conflict risk in other years with the pattern we capture. We adopted a time-cross validation method to prove our hypothesis at a more detailed scale. We showed that the risk of armed conflict is primarily influenced by stable background contexts with complex patterns, followed by climate deviations related covariates. We further revealed that positive temperature deviations or precipitation extremes are associated with increased risk of armed conflict worldwide. We also simulated the risk of armed conflict worldwide from 2000 to 2015. Thus, this study provides a novel insight into understanding the climate–conflict link at the global scale.
The main data are as follows: (1)Armed conflict database: GED 17.1 version data are taken from the UCDP website, which is an openly available armed conflict dataset with georeferenced information4. This dataset records three types of armed conflict events (state-based conflict, non-state conflict and one-sided violence) with at least 1 direct death at a specific location and with specific data. The maximum spatial resolution of the UCDP GED 17.1 version is the individual village or town. Therefore, we can localize armed conflict to 0.1° × 0.1° grid based on latitude and longitude coordinates. (2)The Climate Research Unit TS4.0 global dataset downloaded from the Climatic Research Unit (CRU) of University of East Anglia was used to construct a monthly gridded land surface precipitation dataset on 0.5° × 0.5° grids for the period from 1901 to 2015. Then the monthly precipitation dataset was used to generate long-term (1970–1999) mean precipitation distribution data, one-year and two-year standardized precipitation index (2000–2015). To match other covariates, we resampled these data to 0.1° × 0.1° grids. (3)From the CRU website, we acquired the monthly mean temperature dataset that is arithmetically derived from the Climate Research Unit TS4.0 global dataset at each 0.5° latitude/longitude grid cell across the global land surface. Then long-term (1970–1999) mean temperature distribution data, and the one-year and two-year standardized temperature index (2000–2015) were produced and resampled to 0.1° × 0.1° grids.
We assume that a machine learning model should be able to infer potential patterns between armed conflict and climate variability based on established facts, which may help to simulate the risk of armed conflict. The potential patterns may be complex. To capture complex responses, the boosted regression tree (BRT) modelling framework was adopted based on the R version 3.3.3 64-bit statistical computing platform. In the present study, independent variables are classified into two categories: stable background contexts and climate deviation related factors. The former is used to reflect various meteorological, geographical, political and socioeconomic contexts, while the latter is adopted to depict the extent of climate change. Based on UCDP GED, two binary dependent variables, including armed conflict incidence and armed conflict onset, were defined for each 0.1° × 0.1° grid on a yearly basis to represent armed conflict risk. If there are one or more instances of armed conflict event in one grid in a single year, the armed conflict incidence indicator is coded as one (high-risk) for the grid. In addition, if a new armed conflict event outbreak occurs after one calendar year of inactivity in one grid, armed conflict onset is assigned the value of one for the grid. Both binary dependent variables are otherwise assigned the value of zero (low-risk). For each year, an equivalent amount of low-risk samples and high-risk samples are randomly selected to construct the one-year samples and to train the BRT models.
The results reveal that the risk of armed conflict is primarily influenced by stable background contexts with complex patterns, followed by climate deviations related covariates. The inferred patterns show that positive temperature deviations or precipitation extremes are associated with increased risk of armed conflict worldwide. Our findings indicate that a better understanding of climate-conflict linkages at the global scale enhances the spatiotemporal modelling capacity for the risk of armed conflict.
Sources:
NASA
Provided by the IKCEST Disaster Risk Reduction Knowledge Service System
Comment list ( 0 )
