Mekelle: 15 May 2024 (Tigray Herald)
War on cropland in Tigray causes famine
(15-05-2024) Stanford and Ghent University researchers investigated the abandonment of cropland during the Tigray war in northern Ethiopia. They found a direct causal link between the war and the famine there.
The researchers focused on the well-cultivated land area in the highlands of Tigray (above 1,200 m) in 2021. Using satellite images, they detected a significant net loss of 543 km² of well-cultivated land, which remained fallow. Field observations confirmed this. Moreover, they found a strong causal relationship between the number of conflict incidents (battles, mass killings, bombings) and net loss of cropland area. When considering only conflict incidents that occured during the growing season, the relationship is even stronger.
Moreover, the team, which is strongly rooted in Tigray, identified vast uncultivated croplands in districts from which large numbers of internally displaced persons (IDPs) had left. The researchers also counted the kilocalories lost due to the loss of productive croplands: the missed volume of staple crops could otherwise have fed at least 90% of the IDPs in Tigray.
This unique study illuminates the impact of the Tigray war on arable land, correlating the loss of cultivated land with the intensity of conflict incidents and the proportion of IDPs.
The war in Tigray, Ethiopia has triggered a massive humanitarian crisis, displacing millions. Yet, the impact on cultivated land and local food production remains poorly understood, impeding effective aid. Leveraging Sentinel-2 satellite imagery and a decision tree algorithm with Normalized Difference Vegetation Index (NDVI) time series, we developed a model to map well-cultivated cropland, defined as fields judged by field surveyors to have satisfactory to optimal crop condition in 2021-2022 field observations. Assessing satellite estimated well-cultivated land in highland croplands ( 1200 m a.s.l), we found a net loss of 543 sq. km (95% CI: 81 sq. km) well-cultivated land in highland croplands equivalent to ≈ 8% of the average total surveyed cropland estimate from Central Statistical Agency between 2015 and 2019 (ESS, 2023a) in potential highland cropland. The net loss was positively associated with the density of recorded conflict incidents and sub-regions with high numbers of internally displaced peoples (IDPs), consistent with a causal effect of the conflict on cultivated land.
Employing a two-way fixed effect model causal analysis with rainfall covariates, we quantified the impact of conflict incidents on cultivated land during the pre-war (2019/20) and in-war (2021) periods. Results indicated a ≈ 6.17 sq. km (SE: 2.06) additional loss per unit increase in conflict incidents during the growing season (June to October), eight times higher than total incidents occurring throughout the entire study period. We estimated the kilocalories lost due to loss of well-cultivated croplands in 2021 could have supported at least 90% of all recorded IDPs in Tigray as of June 2021, discounting for Western Tigray. Our study showcases the utility of satellite data, coupled with local agricultural knowledge, for timely and cost-effective information crucial for aid agencies and long-term rehabilitation initiatives in smallholder farming contexts.
promoted expansion of cropland in relative safer zones and hindered cultivation in other areas in Syria (Li et al., 2022). Although the context-specific variation in the relationship between conflict and cropland abandonment is evident, their individual respective impact on local food supply (Witmer, 2008; Skakun et al., 2019; Yin et al., 2019) and the repercussion of conflict induced cropland abandonment in the affected regions is poorly understood (Olsen et al., 2021). This is even more important to understand in nations dominated by smallholder farmers where more than 70% of food calories directly consumed is provided by smallholder farmers as compared to 55% globally (Samberg et al., 2016). Furthermore, the socio-economic implication of cropland abandonment is detrimental to the livelihood of the most vulnerable and often rural smallholder farming communities (Blair et al., 2018; Lambin et al., 2001; Shackleton et al., 2001). Unfortunately, abandonment of fragmented cropland in small scale subsistence farming systems are generally understudied, in part due to insufficient spatial resolution in past satellite imagery (Segarra et al., 2020; Yin et al., 2020; Mutanga et al., 2017), and in part due to a lack of sufficient ground agricultural survey data with which to test estimates. Thus, further work is needed to improve quantification of conflict’s impact on cropland abandonment and food security, especially in smallholder farming systems (Olsen et al., 2021).
In this study, we use remotely sensed satellite imagery to assess the impact of the Tigray conflict on cultivated land dynamics and consequent local crop production in 2021 as compared to 2019/20. Tigray is dominated by subsistence smallholder farming and was kept under a complete communication blackout. We used time series of Sentinel-2 images and derived Normalized Difference Vegetation Index (NDVI) data to map well-cultivated land at 10 m resolution over the study period validating with field observations from the growing seasons in 2021 and 2022; well-cultivated land is qualitatively classified as satisfactory to optimally cultivated land using growth features such as plant height, greenness and density, ear length and homogeneity in crop stand based on local standards (Ghebreyohannes et al., 2022c). We then combined the estimated changes in cultivation status with data on conflict locations (Figure 1) and IDP in Tigray to investigate three research questions: (i) Can we detect changes in well-cultivated land using NDVI criteria?, (ii) Is there a causal relationship between changes in well-cultivated land and the density of conflict incidence and recorded IDP? and (iii) What are the implications of changes in well-cultivated land on the total calorie supply in the region?. Our study showcases the potential of satellite monitoring of conflict-driven loss of well-cultivated land, particularly in smallholder dominated environments, and illustrates the heavy toll of the conflict on recent food availability in the region.
262 field sample plots collected in 2022 included 59 plots that were good cultivated, 63 medium cultivated, 100 poor cultivated, 18 fallowed and 22 fallowed with weeds. About 84% in 2021 and 87% in 2022 of all cultivated sample plots were cereals.
2.1.5. Conflict incidence data
Spatial distribution of reported conflict incidents from November 2020 to December 2021 were originally collected by Map Ethiopia (Map Ethiopia, 2022) and obtained from the Tigray Atlas (Annys et al., 2021) for this study. Conflict incidents include battles, ambushes, air strikes, drone attacks and shelling (Annys et al., 2021). A total of 245 conflict incidents were reported from November 2020 to October 2021 (Annys et al., 2021). The highest number of conflict incidence occurred in the Central zone with total of 62 recorded incidents and the second highest in the Northwestern zone with 49 incidents. In most zones, conflict incidents occurred primarily before the growing season. A total of 102 and 101 conflict incidents were reported in the pre-agricultural activity season, and land preparation season, respectively while only 42 incidents were reported in the growing season (Figure 2).
output of our model with 89% overall accuracy for the test data; the omission error for the test data was 3% for the well-cultivated and 16% for the non-cultivated class. Employing additional surface reflectance variables or vegetation indices and compositing on bi-weekly or weekly basis did not improve our predictions (see details in section S1.3). Upon further investigation, we found it helpful to include Max NDVI month less or equal to 9 as an additional feature to our model based on small number of cases in both years. Tables S1 and S2 in the supplements show performance of the combined criteria per cultivation status category for 2021 and 2022. The combined criteria (NDVImax of greater or equal to 0.5, (NDVImax – (NDVImaxmonth+2)) of 0.2 or higher and NDVImaxmonth less than 9) do well in identifying well-cultivated land from not cultivated lands. However, there is not a clear dichotomy between cultivated and non-cultivated land due to the presence of some very poorly cultivated fields which limits our assessment to change in well-cultivated land only. We believe such fields can effectively be considered as uncultivated for the purposes of food security.
2.2.2. Change in well cultivated land
Change in well-cultivated land was assessed in the Tigray region specifically on land classified as either cropland or grassland according to ESA land cover classification (Zanaga et al., 2021). Grassland use was considered after visual inspection of ESA map via Google satellite base map where we observed actual croplands were incorrectly classified as grassland especially in Northern areas in Northwestern zone and Southern areas in Central zone. The combined crop and grass land classes cover about 67% (crop: 25%, grass: 42%) of land cover in Tigray based on ESA. In this study, we refer to combined cropland and grassland ESA land cover classification as potential cropland in Tigray.
Since lowlands characterized by hot arid and hot semi-arid agro-climatic zones were not represented in the ground truth data collected in both 2021 and 2022, we lack a good understanding of our NDVI criteria performance in lowlands in general. Preliminary NDVI time series inspection of lowlands below 1200 m a.s.l. using known land cover status from 2020 suggests our model cannot distinguish between late peaking crops and fallow with weeds. Consequently, we only consider change in cultivation status for areas above 1200 m a.s.l. which reduces total Tigray’s crop and grass land cover (according to ESA 2020 classification) from 34,932 to 23,896 km2 which unfortunately excludes the majority of Western and some parts of Northwestern Tigray. Croplands in Tigray are estimated to produce cereals with a dominant share of 82%, oil seeds (13%) and pulses (3%) on average from 2014 to 2019 according to the land utilization survey (ESS, 2023a). Western Tigray shares a total of 21% cropland half of which comes from oil seeds. On the other hand, Northwestern Tigray shares a total of 24% cropland where 20% belonged to cereal cropland according to land utilization survey done in 2015 (ESS, 2023b). Using 5-year (2014-2019) surveyed average cropland share per zone and potential cropland (crop and grass land) percentage below 1200 m, we estimate we are excluding about 29% of total cropland share and 23% of cereal cropland
We employed the NDVI combined criteria discussed in Section 2.2.1 on potential cropland area that are above 1200 m a.s.l at 10 m resolution, we refer to this area as potential highland cropland hereafter in our study. We predicted cultivation status for both the pre-war period (2019-2020) (see supplementary section S1.2 for aggregation details) and in-war period (2021) to detect change in well-cultivated land status due to war. Our well-cultivated land classification process is summarized in Figure 4. To measure the uncertainty in the results, we tested various combination of minimum NDVImax (ranging between 0.4 to 0.6) and minimum NDVImax – NDVImaxmonth+2 (ranging from 0.15 to 0.25) that achieve similar accuracies using 2021 ground truth plots (see Table S9 in the supplements). We applied the methods outlined in Olofsson et al. (2013), to our pixel counting-based area estimation, producing error-adjusted area estimates and confidence intervals for both the pre-war and in-war periods (see details in section S1.4 in the supplements).
all major cereal types produced in the region and their respective cropland share to calculate the yield lost. Thus, our estimates do not take into account differences in crop yield productivity between plots. The crop land share data was used to estimate the portion of lost cultivated land per crop type. Energy lost in kilocalories was then estimated using USDA’s crop specific nutrient data from Food Data Central (USDA, 2019). We assessed the number of people that could have been supported for a year using total energy lost in kcal and the minimum daily recommended calorie intake requirement, 2100 kcal per adult equivalent per day (Asrat and Anteneh, 2020) (see section 3.4).
3. Results and discussion
3.1. Change in well-cultivated land between 2019 and 2021
Out of Tigray’s total area of 52,751 km2, well-cultivated land assessment based on our NDVI criteria was applied on 23,883 km2 area which was classified as either crop or grassland based on ESA (Zanaga et al., 2021) and above 1200 m a.s.l. at 10 m resolution. Figure 5A shows spatial distribution of change in well-cultivation status in this study. We observed 543 km2 in net loss (gain-loss) of what qualifies as well-cultivated land when comparing pre-war time (2019/2020) to in-war time (2021) (see details of cultivated land area proportion for both periods in Table S5 in the supplements). The net loss is equivalent to 8% of the average total surveyed cropland estimate from Central Statistical Agency between 2015 and 2019 (ESS, 2023a) in potential highland cropland. Net loss of well-cultivated land reached up to 25% of total potential highland cropland area in a 10×10 km grid cell (see Figure 5B).
Figure 5. Spatial distribution of estimated change in cultivation status from 2019/20 to 2021 (A) and estimated net (gain minus loss) cultivated cropland from 2019/20 to 2021 using 10x10km grid (B) in Tigray, Ethiopia as derived from Sentinel-2 satellite imagery
The 543 km2 net loss of well-cultivated land is overall modest estimate given the severity of the war. This may signal smallholder farmers’ resilience under occupying forces targeted killings, looting and siege (Nyssen et al., 2022). Our estimate is in line with earlier findings that reported most lands were expected to be cultivated in 2021 even under difficult conditions based on interviews and remote assessment of ploughing status on 16 sample rain-fed and 14 irrigated perimeters spread over Tigray (Nyssen et al., 2022). Although confidence interval for the estimated net loss in well-cultivated land is challenging, we estimate ranges of 193 to 983 km2 using NDVI criteria thresholds that achieve similar accuracies based on sampled field plots from 2021 (see supplementary section S1.5). We recognize that this is not a comprehensive confidence interval for estimated total net loss (gain-loss) in well-cultivated land, but rather an estimate of the sensitivity of results to alternative classification rules, which gives one measure of the uncertainty in the results.
3.2. Causal relationship between conflict incidence and loss of cultivation
We found that the area of well-cultivated land was negatively correlated to the density of conflict incidents overall. More importantly, the loss of well-cultivated land due to a conflict incidence occurring during the growing season (June to October) was about eight times more than a conflict incidence occurring overall period in this study. Specifically, two-way fixed effect model causal analysis performed using conflict incidence during growing season and overall period separately resulted in ≈ -6.17 (SE: 2.06) and ≈ -0.73 (SE: 0.5) additional well-cultivated land in sq. km for every 1 unit increase in conflict incidents, respectively
measure of conflict stress than whether any conflict is within a certain distance.
3.3. Loss of well cultivated land and internally displaced people
The Tigray conflict has led to a severe humanitarian crisis displacing close to 2 million people internally and more than 48,000 refugees in neighboring Sudan by August 2021 (Annys et al., 2021). Given that 80% of the population in Tigray depend for their livelihoods on agriculture (Tigray Bureau, 2022), one might expect a strong relationship between reported IDPs and satellite-estimated loss of cultivated land in our study. We used the statistics of 331,976 IDPs out of the 2 million (excluding Western Tigray) with known wereda (district) of origin (Annys et al., 2021) to compare the estimated loss of cultivated land in potential highland cropland due to abandonment by IDPs in 2021 (LossIDP) with that of satellite estimated loss using our NDVI based method (LossNDVI). The high number of IDPs reported without zone of origin may lead us to significantly underestimate loss of cultivated land due to abandonment of IDPs in 2021.
Nonetheless, we found central and Northwestern zones with high estimated loss of cultivated land due to abandonment of IDPs also showed high satellite-estimated loss of well-cultivated land, with deviation of 47 and 128 km2, respectively (Figure 6). On the other hand, areas with lower estimated loss of cultivated land due to abandonment of IDPs (namely the Eastern, South Eastern and Southern zones) also resulted in lower satellite-estimated loss of well-cultivated land with deviation ranging 12 to 76 km2 (Figure 6). In addition, it is possible for idle land owned by IDP to be cultivated by close relative left behind with permission to do so; we do not consider such scenario in our study.
3.4. Estimated loss of energy from loss of well-cultivated land
The loss of 543 km2 of well-cultivated land during harvest season in 2021 reduced the ability of local farmers to meet the needs of food production within Tigray; this is excluding 23% of cereal cropland share from Western and portion of Northwestern zones (see details in Table 1) where the highest recorded IDP originated from (Annys et al., 2021). Tables 3 and 4 show details for calculating estimated loss of energy from the 543 km2 net loss of well-cultivated land in this study and the consequent number of people that could have been supported by lost crop land. Using a minimum daily recommended calorie intake requirement of 2100 kilocalorie per adult equivalent per day (Asrat and Anteneh, 2020), this production loss could have met the food energy needs of 527,123 (95% CI: 78,797) people in Tigray which represents 90% (95% CI: 13%) of all recorded IDPs as of June 2021, discounting for Western Tigray. We note that this estimate does not consider energy losses from livestock either looted or killed nor the decrease in productivity of farm lands sown during the study period nor accessibility of kilocalories that may be affected by factors such as logistics or market prices.
4. Conclusion
The goal of our study was to evaluate the impact of the war in Tigray on cultivation status of croplands using satellite observations. We found that the net loss of well cultivated land in the Tigray highland croplands exhibited spatial heterogeneity and positively correlated to the density of recorded conflict incidents, consistent with a causal effect of the conflict on cultivated land. Administrative zones with high estimated abandonment of cultivated land based on reported IDP also show high satellite-estimated loss of well cultivated land, further corroborating the conflict as a cause of loss in cultivated land. Furthermore, our study estimated about 90% of the recorded IDPs by June 2021 (outside of those originating from Western Tigray) could have been supported by the calorie lost due to abandonment of highland croplands in Tigray. Our findings show that the use of satellite imagery data in a subsistence smallholder farming context, along with proper local understanding of agricultural practices, may provide timely and pertinent information to humanitarian aid agencies and long-term rehabilitation projects. Moreover, the use of these methodologies can go well beyond conflict assessment to aid in successful land use, agricultural intervention and conservation strategies.
5. Code availability
The code used to perform analyses in this study is available on GitHub repository at: https://github.com/LobellLab/TigrayWarPaper2023
Liya Weldegebriel: Writing – review & editing, Writing – original draft, Visualization, Validation, Methodology, Formal analysis, Data curation, Conceptualization. Emnet Negash: Writing – review & editing, Visualization, Resources. Jan Nyssen: Writing – review & editing, Data curation. David Lobell: Writing – review & editing, Supervision, Methodology, Funding acquisition, Conceptualization
Declaration of Competing Interest
☒ The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
☐ The authors declare the following financial interests/personal relationships which may be considered as potential competing interests:
Acknowledgment
This work was supported by NASA Harvest Consortium (NASA Applied 787 Sciences Grant No. 80NSSC17K0652, sub-award 54308-Z6059203 to DBL). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the NASA. We are grateful to the team of geographers from Mekelle University in Tigray, Ethiopia for the field data collection.
Source፡ Science Direct
