Lake Chad Regional Economic Memorandum  |  Development for Peace Technical Paper 5. Conflict and Climate Change in the Lake Chad Region Peter Fisker (University of Copenhagen) 243 Lake Chad Regional Economic Memorandum  |  Development for Peace 6.1 Introduction Peace and security are basic conditions for economic is somehow physiological, as humans are generally and social development. Conflict, on the other hand, shown, in the medical literature, to be more aggressive can reverse years of economic growth and induce long- when temperatures are higher. Harari and Ferrara [2018] term harm on almost all aspects of development. For the explore the ’income channel’ and find that part of the past decade, the Lake Chad region has been the setting of variation in conflict can be explained by a drought index conflicts between government forces and armed groups, when dis-aggregated to the growing period of the main most notably the Boko Haram. Although the intensity of crops across Africa. However, whether the results would fighting has petered off in recent years, the conflict has hold without temperatures as an input to the SPEI is spread from Northern Nigeria and now affects all four unclear. countries of the region. This paper attempts to shed light on the geographical Due to the paramount importance of avoiding armed distribution of conflict and its climatic determinants conflict, a large economic literature exists that seeks in the Lake Chad region following a sub-national to find explanations for the onset and prevalence of approach where readily available spatial data is conflict in developing countries. Blattman and Miguel employed at two different units of aggregation:  Firstly, [2010] list some of the most common theories of conflict 90 second level administrative areas, and secondly, around including competition for resources, economic grievances, 5,318 grid cells covering the same region. Exposure to and the possibility of looting. conflict is here defined as the intensity (for districts) or incidence (for cells) of conflict in a given unit in a More recently, a strand of literature focuses more given year. Parts of the population may not be directly on geographic and climatic root causes of conflict. exposed by this definition, but since the units of analysis For instance, in a meta-analysis of 55 studies, Burke are relatively small, most will be affected in some ways, et al. [2015] find that higher temperatures is the most for instance by safety concerns when visiting the nearest important climatic factor leading to more interpersonal towns to trade or by the general economic consequences. and intergroup conflict. With a specific focus on civil war in Africa, Burke et al. [2009] warned that projected The results of the analysis suggest, in line with increases in temperatures could lead to 54 percent the literature mentioned above, that temperature increase in armed conflicts by 2030. However, both anomalies do have a positive impact on conflict studies conclude that more research is needed in order across districts, cells and years. It also shows that to understand the mechanisms behind this relationship negative NDVI (Normalised Difference Vegetation as well as investigating the potential adverse effects of Index) anomalies are associated with more conflict— climate change. More recently, Eberle et al. [2020] found especially in cropland zones and during growing seasons. that a 1 degree increase in temperatures is associated Rainfall anomalies as well as the SPEI (Standardized with a 54 percent increase in conflict probability in Precipitation-Evapotranspiration Index) do not exhibit areas that are home to both herders and farmers and a the same effect on conflict. This could be an indication 17 percent increase in other areas of Africa. A central of measurement errors in these variables—or it could question is whether the effect goes through an ’income indicate that temperatures and rainfall have different channel’ where conflict is ultimately caused by economic effects on conflict rather than the often-mentioned downturns due to lower agricultural productivity in drought-income channel. periods of warmer temperatures—or whether the effect 244 6.1 Introduction Technical Paper 5. Conflict and Climate in the Lake Chad Region 6.2 Data At the core of the analysis lies the geographical Table A6.1 in the appendix contains a list of indicators delimitation of the Lake Chad region. It comprises included in the analysis, their sources, as well as the 4 countries and a total of 90 districts (2nd level spatial and temporal resolution of the raw data. Except administrative units). for conflict, all data sets included here are originally raster format, but are, for the purpose of the analysis, Map 6.1 shows the extent of the Lake Chad region aggregated to the units of analysis, either using the sum within the four countries of Niger, Chad, Nigeria, (population, conflict event, and fatalities) or mean (Share and Cameroon. It also shows the units of analysis of this of cropland, travel time, rainfall, temperatures, and study, namely the districts within the lake region on the greenness). While NDVI and temperature data is based left panel and the grid cells on the right. These units are on pure (processed) satellite images, data on population, chosen from a practical and methodological perspective. travel time, and precipitation are drawn from secondary Firstly, they are large enough to cover a meaningful sources where the pixel values of the raster data sets are number of satellite data pixels, while small enough for the generated from combining various sources of raw data. total number of units to be useful in regression analyses. Conflict risk numbers stand out in this list as it is based Secondly, many policies are implemented at this level, so on geo-referenced point data from the ACLED database policy-makers will be interested in being able to compare and aggregated to the second level administrative units distributions of key variables at district level. The directly from the recorded latitudes and longitudes of the grid cell level is chosen to accompany the district level conflict events. analysis since it allows for much more variation and more observations due to a higher resolution. Furthermore, Data on conflict as well as climate come with a time since the cells represent little squares, there is no concern dimension as well. Here, values are summarized to about endogenous border locations. individual calendar years from 2001 to 2018. Map 6.1: Extent and units of analysis 6.2 Data 245 Lake Chad Regional Economic Memorandum  |  Development for Peace 6.2.1 Conflict conflict event types, battles and violence against civilians have followed a largely similar pattern over the years while The conflict data used in this study comes from riots and protests are not as commonly reported, but still the Armed Conflict Location and Event Database growing in later years. The sum of conflict fatalities in (ACLED). In this database, conflict events are registered the region is generally high and volatile, but saw a peak based mainly on local media reports, and geo-referenced. around 2014 and 2015 to around 1,000 per year before It distinguishes between various types of conflict events, dropping again later. Note the logarithmic scale of the most notably battles, riots, protests, and violence against vertical axis. civilians. For each conflict event, the number of fatalities is also reported. In this analysis, both the number of Table 6.1:  Number of conflict events and fatalities 2001–2018 events and the number of fatalities by district-year are used. These two measures of conflict exposure are central Cameroon Chad Niger Nigeria Total outcome variables in the regression analyses presented in Conflict 1,861 692 620 12,702 15,875 the next section. events Battles 636 332 256 3,171 4,395 Table 6.1 includes summary statistics of key conflict Protests 145 56 83 3,101 3,385 variables in the Lake Chad region during the years Riots 87 24 61 1,448 1,620 2001 to 2018. Of conflict events, battles and violence Violence 621 193 133 3,813 4,760 against civilians are the most widespread types. Nigeria Fatalities 6,124 6,234 2,550 60,925 75,833 has seen by far the largest number of actual battles, and also the largest number of fatalities. Cameroon is in second place in terms of events, but with a distribution 6.2.2 Climate of events leaning more towards acts of violence against civilians. Niger is the most peaceful country in the region Climate and climate change are often mentioned over the period. among the most important factors for peace and development in the Lake Chad region. For instance, Figure 6.1 shows the development of conflict over time the lake itself has provided livelihoods for the people in the entirety of the Lake Chad Region. Of the four surrounding it for centuries, but shrank dramatically Figure 6.1: Conflict events and fatalities over time Number of events, 2001–2018 100,000– 10,000– 1,000– 100– 1– 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 ▬ Battles ▬ Riots ▬ Protests ▬ Violence … Fatalities Source: Fisker 2021. 246 6.2 Data Technical Paper 5. Conflict and Climate in the Lake Chad Region in size during the 1970’s and 80’s before gradually other parts of the region despite relatively low levels regenerating in recent decades. It now covers 56 percent of rainfall. Map 6.2 shows the distribution of rainfall, of its 1973 extent, although much of the surface is now temperatures, and greenness across the 90 sub-national also covered by vegetation [Vivekananda et al., 2019]. units of the region. While rainfall and greenness show a Land-degradation, over-exploitation, and climate change clear latitudinal gradient, temperatures are also mediated are often mentioned as possible causes for this. by altitude, and thus generally higher in the Eastern parts of the region. Furthermore, the region around Lake Chad is by no means uniform in terms of climate and weather. To the While large geographical variations exist as shown South, the climate is more humid, and the landscape is in Map 6.2, another in- teresting perspective is greener, whereas the Northern parts are drier, less green, the variation over time. Figure 6.2 demonstrates the and with a larger difference between day and night- district-level average anomalies of NDVI, rainfall, and time temperatures. The large areas that were historically temperatures over the period 2001–2018. The large submerged by the lake are still greener and cooler than positive daytime temperature anomaly towards the end  verage rainfall, temperatures, and NDVI Map 6.2: A Source: Fisker 2021. Figure 6.2: NDVI, rainfall, and temperature anomalies over time 1.0– 0.8– 0.6– 0.4– 0.2– 0– -0.2– -0.4– -0.6– -0.8– -1.0– 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 ▬ NDVI ▬ Rainfall ▬ Temperature (day) ▬ Temperature (night) 6.2 Data 247 Lake Chad Regional Economic Memorandum  |  Development for Peace of the period stands out while two other interesting number is probably needed in order for law enforcement observations is a general decline in NDVI throughout and other societal institutions to be efficient. Likewise, the period as well as an upward trend in temperatures. infrastructure can be considered to play a roll in the Rainfall generally fluctuates around the mean. spread of conflict, since an efficient road system allows armed groups to move between locations. Again, more In this paper, drought is measured in three different desolate areas may also provide opportunities to hide ways: F  irstly by anomalies in rainfall and temperatures from government forces, thus enabling local militias measured respectively by the Chirps (Climate Hazards to form and grow. Finally, economic activity—often Group InfraRed Precipitation with Station) data and measured by the intensity of night lights—can affect the Modis Terra, which would correspond to the notion risk of conflict; either because richer areas contain more of meteorological droughts; secondly by NDVI opportunity for looting, or because poorer areas may be anomalies—a more direct proxy for agricultural drought, easier to capture. Map 6.3 displays the distribution by and finally by the SPEI drought index which combines district of the number of conflict fatalities during the long time series of rainfall and temperatures to calculate period 2015–2019, population in 2020, and average the difference between precipiation and potential travel time to nearest urban centre in 2015. While the evapotransipiration. Despite several shortcomings, the distributions of the latter two indicators look similar, latter is used extensively in the economics literature, for they measure slightly different aspects of economic instance by Harari and Ferrara [2018]. development: For any given population density, travel time indicates how easy it is to move around the district. 6.2.3 Other explanatory factors Table 6.2 shows mean values of the different indicators split by country. In terms of average rainfall, the districts Obviously, conflict depends on other factors than the of the region belonging to Niger are the driest and climatic: demographics, infrastructure, and economic Cameroon the wettest. Temperatures are highest in Chad development, to mention a few. A larger population (28.5 degrees Celsius) while the other three countries density means more potential for disagreement and are all around one degree cooler. The largest increases in more competition for limited resources, cf Blattman and temperatures due to climate change are projected to take Miguel [2010]. On the other hand, a certain population place in Niger, followed by Chad. Niger is also the country  onflict intensity, population density and travel time Map 6.3: C 248 6.2 Data Technical Paper 5. Conflict and Climate in the Lake Chad Region Table 6.2: Summary statistics: Mean values of climate and control variables Cameroon Chad Niger Nigeria Total NDVI 0.360 0.287 0.189 0.347 0.327 Temp. (daytime) 15.42 15.51 15.48 15.40 15.42 Temp. (nighttime) 14.69 14.69 14.61 14.66 14.66 Rainfall (chirps) 61.45 36.75 24.39 63.39 56.84 Projected rainfall change 22.00 30.44 30.90 22.63 24.19 Projected temp. change 6.045 6.430 7.498 6.364 6.451 Population 2000 (1000s) 443.6 142.3 290.6 121.3 160.2 Population 2020 (1000s) 788.2 276.4 769.5 228.0 318.8 Travel time 73.27 282.1 603.4 73.19 143.5 Distance to border 25.98 96.79 97.96 72.79 74.57 with the lowest projected increase in rainfall, pointing populations, smaller travel times to urban centres, towards even more difficult conditions for farmers and and also in areas with a larger share of cropland areas. pastoralists there. NDVI values are generally much larger Regarding the latter two indicators, the correlations are in Cameroon and Nigeria, and lower in Niger. reversed when observing districts compared to cells. This is likely caused by the fact that some districts are  verage population numbers for districts vary between A geographically large (especially in Niger and Chad) and 228,000 (Nigeria) and 788,000 (Cameroon), with these have lower shares of cropland area as well as larger the districts belonging to Niger having seen the largest travel distances while also less conflict. percentage increase between 2000 and 2020.  airwise correlations between conflict and Table 6.3: P explanatory variables in pooled data  ravel times to urban areas are generally low in Nigeria T and Cameroon, while large distances exist in Niger and Any event log(events) (cells) (districts) to a lesser extent in Chad, the main reason being that districts in these countries stretch far into desert areas. Temp anom. 0.0873 0.241 Average distance to an international border is by far Rainfall anom 0.0171 0.0345 lowest in Cameroon. NDVI anom -0.0822 -0.1697 SPEI 0.0912 0.1496 log(population) 0.1409 0.5557 6.2.4 Correlations log(travel time) -0.1648 0.0057 Cropland share 0.1278 -0.0598 In order to provide an overview of how conflict Observations 96,642 1,692 incidence and intensity is correlated with the factors that form part of the analysis,  Table 6.3 consists of pairwise correlations between conflict and each of the other variables when all cross-sections that form the panel are pooled. While the effects of the climatic variables are studied in more detail in the next section, it is interesting to note here that conflict is more likely in areas with larger 6.2 Data 249 Lake Chad Regional Economic Memorandum  |  Development for Peace 6.3 Empirical strategy This section lays out the approach to analysing the index in the third. In the analysis, these three models climatic determinants of conflict in the Lake Chad are used because they represent three different ways of region. The analysis investigates the effects of climate measuring climatic impacts. The first model, which on conflict from various perspectives: in the main includes rainfall and temperature anomalies, is the most specification, district-year conflict intensity and cell-year direct way of linking climate shocks to conflict intensity/ conflict incidence are explained by anomalies (z-scores incidence. The second model (with NDVI anomalies) calculated each month in the 19-year period where the compares the outcome of climate variations (i.e. the value represents standard deviations from the long- conditions of the vegetation) to conflict, while the third term mean within the unit and month) in temperature, approach refers to a drought index (SPEI) that combines rainfall, and greenness as well as a 6-month SPEI drought rainfall and temperatures into a measure that informs index in a fixed effects set-up. Since both conflict and about agricultural potential. climate are spatially dynamic processes, the regressions are based upon assumptions of spatially correlated error W * conflictit is the spatial lag of the dependent terms, and in some specifications including a spatially variable. It measures the average number of conflict lagged dependent variable. This takes into account the events or fatalities in neighboring districts in the same year, fact that conflict events tend to spread from a point of i.e. districts or cells that share a border with the unit in origin to neighboring areas. question. It does not distinguish between within-country borders and country borders in this set-up. This term is Due to the differences in size between second-level included separately as a check to whether controlling for administrative units and cells of approx. 10 km * 10 km, the auto-regressive nature of conflict alters the results. ε is two different dependent variables are considered, that the random error term that allows for spatial correlation. best exploit the variation in the data: the former case employs the logarithm of the number of conflict events The climatic variables included in the baseline (i.e. the intensity of conflict) while in the latter case, a specification described by equation 1 are all averages dummy variable indicating whether a conflict event has for full calendar years. However, as argued by Harari taken place in a given cell in a given year is used (i.e. and Ferrara [2018] among others, if the mechanism that incidence of conflict). links climate anomalies and conflict is economic hardship induced by agricultural drought, only the anomalies Equation 1 d  escribes the fixed effects model of conflict observed during the agricultural growing season should intensity/incidence and its climatic predictors at the matter. district/cell level: A related concern is that not all units of observation Conflictit = β1Cit + β2W * conflictit + εit  (1) are areas of agricultural activity. If a drought-income- conflict relationship is expected, it is likely to be more  here Conflict is either the logarithm of the number w directly impacting cropland areas than desert or pastoral of conflict events in a district (i) or an indicator of the areas. In order to capture the differential effect of climate presence of conflict in a cell (also i) in given year (t). C is anomalies on conflict, in equation 2, each climate variable a vector of climate anomalies observed in unit i in year t: in C is therefore interacted with the share of cropland in rainfall as well as daytime temperature in the first model, each unit of observation. Another potential source of NDVI (greenness) in the second, and the 6-month SPEI heterogeneity in impacts is the population of a given 250 6.3 Empirical strategy Technical Paper 5. Conflict and Climate in the Lake Chad Region unit. More people means more potential for conflict and thus a larger effect of climate shocks could be expected. Equation 2 describes the model with heterogeneous effects, which is similar to equation 1 in all other aspects: Conflictit = β1C(GP)it + β2C(GP)it * Xi + β3W * conflictit + εit  (2)  here C(GP)it is a vector of climate anomalies calculated w only for the growing period months i the specific locations before aggregating to years, districts and cells and X is the share of cropland in cell/district i in the year 2000. Finally, in order to test whether the relationships depend on population density,  a model is run at the cell level where conflict incidence is interacted with a dummy variable taking the value one if a cell belongs to the upper half of the population distribution. 6.3 Empirical strategy 251 Lake Chad Regional Economic Memorandum  |  Development for Peace 6.4 Results 6.4.1 Baseline results usual years, districts or cells are more likely to experience conflict activity. Furthermore, and perhaps surprisingly, Table 6.4 includes the results of applying a fixed effects positive rainfall anomalies, i.e. years where rainfall levels estimator to equation 1 where the units of observation are above the mean are also associated with more conflict are districts and the dependent variable the logarithm measured at both district and cells. Turning to measures of conflict events in a given year. Column 1–3 contain of drought, NDVI anomalies have the expected sign, the results of regressions where the error terms are meaning that worse growing conditions are correlated assumed to be spatially correlated whereas column with more conflict. The SPEI, on the other hand, shows 4–6 assume spatial auto-correlation and thus include a the opposite correlation, namely that drought-years (a spatially lagged dependent variable. negative value by this measure) tend to be aligned with less widespread conflict. Each column represents a specific way of measuring impacts of climate variation: Column 1 and 4 focus Adding numbers to the results, a positive temperature on the direct relationship between weather anomalies anomaly of one standard deviation is associated with a (rainfall and temperature) and conflict intensity. 17.6 percentage points increase in the yearly number of Column 2 and 5 use the observed NDVI-anomalies as an conflict events taking place in a given district. At the cell observable proxy for drought conditions whereas column level, a similar temperature anomaly adds 0.8 percentage 3 and 6 show the effects of a common drought-index points to the likelihood of a cell experiencing any conflict that combines long-term information on rainfall and events in that year. A negative NDVI anomaly of one temperatures, namely the 6-month SPEI. standard deviation leads to an increase in the number of conflict events of 8.9 percentage points at the district The results for districts and cells are qualitatively level whereas the likelihood of experiencing a conflict at comparable: Temperature anomalies (both daytime the cell level increases by 0.7 percentage points. and night-time) show a positive effect on conflict intensity and incidence. In other words, in hotter-than- Table 6.4: Baseline results, Districts (1) (2) (3) (4) (5) (6) 0.176*** 0.099*** Temp. (0.028) 0.015) 0.011 0.013 Rainfall (0.027) (0.014) -0.089*** -0.057*** NDVI (0.028) (0.014) 0.165*** 0.080*** SPEI (0.048) (0.022) Spat. lag 0.784*** 0.796*** 0.811*** Conf. events (0.020) (0.020) (0.019) N 1,692 1,598 1,692 1,692 1,598 1,692 Pseudo-r2 0.059 0.029 0.022 0.067 0.037 0.028 Note: Spatially correlated standard errors in parentheses. Fixed effects. Z-scores. * p<0.10, ** p<0.05, *** p<0.01 252 6.4 Results Technical Paper 5. Conflict and Climate in the Lake Chad Region Table 6.5: Baseline results, Cells (1) (2) (3) (4) (5) (6) 0.008*** 0.006*** Temp. (0.001) (0.000) 0.002*** 0.001*** Rainfall (0.001) (0.000) -0.007*** -0.005*** NDVI (0.001) (0.000) 0.011*** 0.006*** SPEI (0.001) (0.001) 0.443*** 0.446*** 0.450*** Sp_lag (0.005) (0.005) (0.005) N 91,273 91,273 91,273 91,273 91,273 91,273 psudo-r2 0.01 0.01 0.01 0.01 0.01 0.01 Note: Spatially correlated standard errors in parentheses. Fixed effects. Z-scores. * p<0.10, ** p<0.05, *** p<0.01 All results are robust to controlling for the geographical economic hardship. In order to investigate that, the next spillover of conflicts. Column 4–6 of Table 6.4 and 6.5 set of results will include climate anomalies calculated add a spatially lagged version of the dependent variable on basis of growing season months only, and further that measures the average conflict incidence or number introduce interaction terms between each variable and of conflict events in neighboring units (i.e. districts the share of cropland within a unit of observation. This or cells that share a border with the district or cell in largely follows the approach of Harari and Ferrara [2018] question). This variable generally has a large contribution who found an effect growing season SPEI on conflict to explaining conflict intensity while point estimates on incidence across all of Sub-Saharan Africa, albeit with the explanatory variables tend to drop slightly. much larger units of observations. Table A6.2 and A6.3 in the appendix show results Table 6.6 and 6.7 show the effects of growing season- of estimating a model including temporal lags of specific climate anomalies on conflict in districts and the explanatory variables. It is demonstrated that cells respectively. Both tables further include interactions temperature anomalies are significant predictors of the between these and the share of cropland within each unit. number of conflict events at the district level up to three years into the future. NDVI anomalies, on the other T  emperature anomalies are still positively associated with hand, only predict conflict with statistical significance conflict; especially in areas with more cropland. in the same year as the conflicts. This fits well with the notion of NDVI anomalies being a more direct proxy of For drought measured by NDVI anomalies, the vegetation conditions on the ground than other climatic negative effect observed in the baseline model also variables. persists. Additionally it should be noted that the effect is weaker in areas of no cropland and larger, the larger the share of a unit is considered cropland. This is in line with 6.4.2 Exploring heterogeneous effects expectations that bad harvests can lead to more conflict through an income channel. A central question that remains to be addressed is whether the results found in table 6.4 and 6.5 are Turning to rainfall and SPEI—the two variables where caused by a drought-income-channel where conflict is results opposite to the expectations were found in the more likely in places where farmers are suffering from baseline analysis, a few interesting observations are 6.4 Results 253 Lake Chad Regional Economic Memorandum  |  Development for Peace Table 6.6: Heterogeneous effects, Districts Table 6.7: Heterogeneous croplands effects, Cells (1) (2) (3) (1) (2) (3) 0.021 0.002* GP Temp. anom. GP Temp. anom. (0.047) (0.001) 0.002 0.001 GP Rainfall anom. GP Rainfall anom. (0.043) (0.001) Cropland*GP Temp. 0.478*** Cropland*GP Temp. 0.046*** anom. (0.112) anom. (0.001) -0.086* Cropland*GP 0.004 GP NDVI anom. (0.051) Rainfall anom. (0.003) Cropland*GP NDVI -0.219* -0.001 GP NDVI anom. anom. (0.117) (0.001) -0.173** Cropland*GP NDVI -0.036*** GP SPEI (0.070) anom. (0.003) 0.464*** 0.001 Cropland*GP SPEI GP SPEI (0.159) (0.002) 0.025*** N 1,692 1,692 1,692 Cropland*GP SPEI (0.004) N 61,013 61,013 61,013 pseudo-r2 0.07 0.05 0.01 pseudo-r2 0.02 0.01 0.00 Note: Standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01 Note: Standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01 noted: F  irstly, the positive correlation between rainfall Table 6.8: Heterogeneous population effects, Cells and conflict disappears when only considering the (1) (2) (3) growing seasons. The same is true for SPEI, however, 0.002*** when focusing on the agricultural areas, the positive (and Temp. anom. (0.001) somewhat contradictory) relationship re-emerges. 0.000 Rainfall anom. (0.001) Another potential mediating factor is population Temp. anom. 1.288*** density. In Table 6.8, Urban refers to a situation where (day)*Urban (0.095) the population of a cell is larger than the median of the Rainfall 0.385*** anom*Urban (0.095) distribution, which serves as a crude way of distinguishing -0.002*** between urban and rural areas. What is evident is that NDVI anom. (0.001) the effects of climate anomalies on conflict events are -1.251*** NDVI anom*Urban largely driven by areas witha population density above (0.093) the median. In all cases the point estimates retain their 0.003** SPEI direction, but become much more significant (statistically (0.001) and economically) when adding the urban interaction 1.404*** SPEI*Urban (0.165) terms. N 91,273 91,273 91,273 pseudo-r2 0.01 0.01 0.00  esults including heterogeneous effects related to market R Note: Standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01 access (travel time to nearest urban area) are not included as they are similar to those where population is used as interaction term. 254 6.4 Results Technical Paper 5. Conflict and Climate in the Lake Chad Region 6.5 Conclusion In conclusion, t his study finds that the distribution of Based on this it is therefore not possible to conclude conflict events across time and space in the Lake Chad which of the explanations is more likely. The fact that region is correlated with climatic factors. NDVI anomalies show expected signs and the effect is more pronounced in croplands during growing season Higher-than-usual temperatures leads to an increase points toward the measurement error explanation. in conflict activity both measured at the district level However, this relation could to some extent also be and the more detailed grid cell level. The same is true spuriously driven by temperatures affecting both NDVI for observed greenness anomalies, an effect that becomes and conflict. More precise rainfall data, for instance from stronger when focusing on anomalies during the growing the Global Precipitation Measurement Mission (GPM) season in cropland areas. However, rainfall (CHIRPS) might shed more light on this puzzle. and SPEI are not showing similar relationships with conflict. Two possible explanations for these apparently contradictory findings stand out: The first possibility is that conflict in the Lake Chad region is, in fact, affected much more by temperature anomalies than rainfall anomalies. This would be in line with the hypothesis that there the channel through which the relationship operates is more physiological than depending on agricultural income. A second possible explanation for the seemingly opposite results could simply be measurement errors in the SPEI and CHIRPS data sets. Both of these data sources are (partly) interpolated from weather station observations, and the distance to the nearest weather station is sometimes large. Figure A.1 in the appendix shows the distribution of weather stations used by CHIRPS and CRU (the database behind SPEI) respectively in the Lake Chad Region. There are around 12 (CHIRPS in 2010) and 7 (CRU in all years 2000-2014) weather stations in the region with observations that feed into the Chirps and SPEI data sets. This compares to 90 districts and 5,369 cells. So for a large majority of the observations in this analysis, rainfall and the SPEI will be based entirely on interpolations. On the contrary, the spatial resolution of NDVI and Temperatures is higher than the cells used, so in that case, the observed values are more valid. Likewise, conflict data is aggregated from high precision geographical coordinates, so there is also high confidence that the conflicts actually took place in the recorded locations. 6.5 Conclusion 255 Lake Chad Regional Economic Memorandum  |  Development for Peace References Christopher Blattman and Edward Miguel. Civil war. Journal of Economic Literature, 48(1):3–57, March 2010. doi: 10.1257/jel.48.1.3. URL https://www.aeaweb.org/articles?id=10.1257/jel.48.1.3. Marshall Burke, Solomon M. Hsiang, and Edward Miguel. Climate and conflict. Annual Review of Economics, 7(1):577–617, 2015. doi: 10. 1146/annurev-economics-080614-115430. URL https://doi.org/10.1146/ annurev-economics-080614-115430. Marshall B. Burke, Edward Miguel, Shanker Satyanath, John A. Dykema, and David B. Lobell. Warming increases the risk of civil war in africa. Proceedings of the National Academy of Sciences, 106(49):20670–20674, 2009. ISSN 0027- 8424. doi: 10.1073/pnas.0907998106. URL https://www.pnas.org/content/ 106/49/20670. Ulrich J. Eberle, Dominic Rohner, and Mathias Thoenig. Heat and Hate: Climate Security and Farmer-Herder Conflicts in Africa. CEPR Discussion Papers 15542, C.E.P.R. Discussion Papers, December 2020. URL https://ideas.repec.org/p/cpr/ceprdp/15542.html. Mariaflavia Harari and Eliana La Ferrara. Conflict, climate, and cells: A disaggregated analysis. The Review of Economics and Statistics, 100(4):594– 608, 2018. doi: 10.1162/rest\_a\_00730. URL https://doi.org/10.1162/ rest_a_00730. Janani Vivekananda, Dr Martin Wall, Dr Florence Sylvestre, and Chitra Nagara- jan. Shoring up stability: Addressing climate fragility risks in the Lake Chad Region. adelphi research gemeinnützige GmbH, 2019. 256 References Technical Paper 5. Conflict and Climate in the Lake Chad Region Appendix Table A6.1: Data sources Indicators Data format Spatial resolution Temporal coverage Source Number of people Raster (tiff) 30 arc seconds 2000, 2020 World Pop Population per cell (~1 km) Intensity of Raster (tiff) 500 m pixels Monthly - Visible Infrared Night-time lights here April Imaging (average radiance) 2012 (earliest Radiometer available) and Suite (VIIRS) Infrastructure April 2019 Accessibility to Raster (tiff) 1 km 2015 (update and Malaria Atlas cities (travel time improvement to Project to nearest urban 2000 dataset) center) Precipitation Raster (tiff) 2.5 arc minutes Monthly, Chirps (~4 km) 2000–2018 Climate Greenness (NDV) Raster (HDF) 0.05 degrees Monthly, Modis Terra, and temperature (~5 km) 2000–2018 mod13c2 Projected Raster (tiff) 2.5 arc minute 2014–2060 Worldclim temperature and (~4 km) Climate Change precipitation (CMIP6, SSP2.5) Number of events Geo- GPS points 2015–2019 and ACLED + Fatalities referenced aggregated to change between Conflict (Battles, protests, event (point) district level 1014 and 1519 riots, violence data against civilians) Appendix 257 Lake Chad Regional Economic Memorandum  |  Development for Peace  aseline results with time lags, district Table A6.2: B  aseline results with time lags, district Table A6.3: B level level (1) (1) (2) 0.192*** -0.168*** Temp. anom. (day) NDVI anom. (0.034) (0.025) 0.132*** -0.055 L.Temp. anom. (day) L.NDVI anom. (0.031) (0.067) 0.201*** -0.058 L2.Temp. anom. (day) L2.NDVI anom. (0.047) (0.048) 0.065* -0.118 L3.Temp. anom. (day) L3.NDVI anom. (0.038) (0.076) -0.022 0.321*** Rainfall anom. (mean) spei06 (0.026) (0.043) -0.080*** 0.206*** L.Rainfall anom. L.(mean) spei06 (0.027) (0.034) -0.018 0.206*** L2.Rainfall anom. L2.(mean) spei06 (0.026) (0.030) -0.086*** 0.110** L3.Rainfall anom. L3.(mean) spei06 (0.030) (0.045) 0.769*** 0.673*** 0.942*** Constant Constant (0.028) (0.032) (0.044) N 1410 N 1,316 1,410 r2 0.17 r2 0.06 0.11 Note: Standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01 Note: Standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01 258 Appendix Technical Paper 5. Conflict and Climate in the Lake Chad Region Figure A6.1:  Distribution of weather stations in the Lake Chad region Appendix 259