A dynamic linkage between greenhouse gas (GHG) emissions and agricultural productivity: evidence from Ethiopia

Climate models project significant impacts of climate change on agricultural productivity, threatening global food security via higher average temperatures, water resource depletion, and greater climate variability. Agriculture-related food systems contribute roughly one-third of global GHG emissions. Africa, despite low per-capita emissions, is highly vulnerable; agricultural and land-use emissions in Africa rose markedly between 2000 and 2018. In Ethiopia, agriculture employs 80–85% of the population, contributes 36.7% of GDP, and generates 88.8% of export revenue; it is also the main source of national GHGs (~80%). Agricultural sector GHG emissions increased from 19,586.06 Gg CO2e (1994) to 226,157.6 Gg CO2e (2022). Prior studies in Ethiopia have largely examined micro-level effects and often omit time dynamics of agricultural GHGs on productivity. This study addresses these gaps by assessing how GHG emissions (CO₂, CH₄, N₂O) and related factors affect agricultural productivity over time, using grain yield per hectare as a proxy. The core research question is: If there is a relationship between agricultural GHG emissions and agricultural productivity (grain yield per hectare), is it positive or negative in the long run?

Global literature attributes climate change to human-induced GHG emissions, with agriculture the second-largest source after energy. While developing countries contribute less globally, emissions are rising and climate impacts threaten agricultural welfare through reduced yields, higher food prices, and increased hunger risk (e.g., Parry et al., 2004; Edoja et al., 2016). Studies suggest closing yield gaps can mitigate emissions growth (Valin et al., 2013). Country evidence (e.g., Nigeria, Cameroon, Tanzania) links climate change to reduced agricultural productivity. In Ethiopia, agricultural and land-use changes contribute significantly to emissions; the government’s CRGE strategy targets limiting emissions growth while pursuing development. Empirical findings for Ethiopia indicate CO₂ emissions can negatively affect productivity and household welfare, and climate change shortens crop maturation, reduces yields, and affects livestock and forage. The literature identifies a gap: most Ethiopian studies are micro-level or cross-sectional, with insufficient time-series analyses capturing dynamic relationships under CRGE. This study complements the literature by providing macro-level, time-series evidence on GHG–productivity linkages.

Study design: Time-series econometric analysis for Ethiopia using annual data (primarily 1990–2022) from the World Bank IBRD/IDA database. Agricultural productivity is proxied by cereal yields (kg/ha). Explanatory variables include fertilizer consumption (kg/ha of arable land), arable land (hectares per person), CO₂ emissions (kt), agricultural methane emissions (CH₄; thousand metric tons CO₂e), agricultural nitrous oxide emissions (N₂O; thousand metric tons CO₂e), rural population (total), agricultural employment (% of total employment), and total land for cereal production (hectares). All variables are log-transformed. Econometric framework: An unstructured VAR is specified in logs to mitigate heteroskedasticity and interpret elasticities. Stationarity is tested using Augmented Dickey–Fuller (ADF) and Phillips–Perron (PP) tests. Lag length is selected via AIC and FPE (recommended lag = 2). Cointegration is assessed using Johansen trace and maximum eigenvalue tests. Given evidence of cointegration, a Vector Error Correction Model (VECM) is estimated to capture short-run dynamics and the speed of adjustment to long-run equilibrium. Diagnostics include residual normality (Jarque–Bera), heteroskedasticity tests, serial correlation tests (Correlogram-Q, Breusch–Godfrey LM), stability tests (inverse roots of AR characteristic polynomial; recursive residuals), Ramsey RESET for specification, variance decomposition, and Granger causality tests to assess predictive causality directions. Variance decomposition quantifies the relative contribution of shocks to forecast error variance over a 5-year horizon.

- Stationarity and integration: Most series are I(1) per ADF and PP tests; rural population is I(0). Residuals of the cointegrating relationship are stationary, indicating a long-run equilibrium relationship. - Lag selection: AIC/FPE suggest VAR lag = 2; cointegration tests conducted accordingly (lag-1 in VECM due to differencing). - Cointegration: Johansen tests (trace and max-eigen) indicate 4–6 cointegrating vectors at the 5% level, evidencing long-run relationships among agricultural productivity and regressors. - Long-run elasticities (normalized cointegrating vector): • Fertilizer consumption (FeC): +0.276 (SE 0.043) – A 1% increase raises agricultural productivity by 0.276%. • Arable land (A_Land): +2.094 (SE 0.142) – 1% increase raises productivity by 2.094%. • CO₂ emissions: −2.749 (SE 0.083) – 1% increase reduces productivity by 2.749%. • Agricultural methane emissions (AME, CH₄): −17.110 (SE 0.727) – 1% increase reduces productivity by 17.11%. • Agricultural nitrous oxide emissions (ANOE, N₂O): +15.916 (SE 0.774) – 1% increase raises productivity by 15.916%. • Rural population (R_PoP): +5.326 (SE 0.240) – 1% increase raises productivity by 5.326%. • Total land for cereal production (Land_CP): +1.314 (SE 0.043) – 1% increase raises productivity by 1.314%. • Agricultural employment (A_Empt): −5.818 (SE 0.400) – 1% increase reduces productivity by 5.818%. Findings indicate elastic long-run relationships for all regressors except fertilizer consumption (inelastic). - Short-run dynamics (VECM): The error-correction (speed of adjustment) coefficient for Δlog(AGP) is −0.744 (significant), implying 74.4% annual adjustment toward long-run equilibrium; convergence to steady state ≈ 1 year 4 months. Short-run coefficients indicate: negative and significant association of lagged fertilizer use, arable land, CO₂ emissions, AME, and rural population with Δlog(AGP); positive and significant effects of lagged ANOE, Land_CP, and A_Empt on Δlog(AGP). Authors note short-run ECM signs can be counterintuitive. - Variance decomposition (Δlog(AGP) over 5 periods): By year 5, AGP’s forecast error variance is explained by its own shocks (59.05%), fertilizer consumption (22.13%), agricultural land (5.96%), CO₂ emissions (~0.14% per table narrative; table values list small shares), AME (4.48%), ANOE (0.59–4.5% per text/table), rural population (~0.92%), Land_CP (~0.92%), and A_Empt (~0.0014%). - Granger causality (lags=3): Bidirectional causality between AGP and FeC; AGP and CO₂; AGP and AME. Unidirectional causality involving AGP with ANOE, R_PoP, and A_Empt. Several additional causal links among regressors are reported at 1–10% significance. - Diagnostics and stability: Inverse roots lie within the unit circle (VECM stability). Serial correlation tests indicate no residual autocorrelation (p>0.10). Ramsey RESET p>0.10 supports correct specification. Recursive residuals remain within critical bounds, indicating stability.

The study answers the research question by demonstrating statistically significant long-run relationships between GHG emissions and agricultural productivity in Ethiopia. CO₂ and CH₄ emissions are negatively associated with productivity, consistent with broader literature on adverse climate impacts on yields and food security. N₂O emissions, largely tied to fertilizer application and soil nitrogen processes, show a positive long-run association with productivity, reflecting the productivity-enhancing role of nitrogen inputs while underscoring the need for efficient management to mitigate environmental costs. Positive long-run elasticities for arable land, rural population, and cereal land allocation suggest scale, land availability, and labor-related factors can enhance productivity, whereas higher agricultural employment shares relate negatively to productivity, potentially reflecting labor underemployment or inefficiencies in labor utilization relative to land and inputs. The strong adjustment speed indicates rapid reversion to equilibrium after shocks. Causality tests confirm dynamic interdependence, with bidirectional predictive relations between AGP and key emission variables (CO₂, CH₄) and fertilizer use, highlighting feedbacks between productivity and emissions. These findings underscore the importance of climate-smart agriculture, improved fertilizer efficiency, and land-use strategies to sustain productivity while reducing harmful emissions.

GHG emissions present a critical challenge for Ethiopia’s agriculture. Time-series evidence shows that CO₂ and CH₄ increases are associated with significant long-run productivity losses, while N₂O (linked to nitrogen use), fertilizer consumption, arable land, rural population, and cereal land allocation are positively related to productivity. The study contributes macro-level, dynamic evidence to the debate on farm size and productivity, finding a positive, elastic long-run relationship between farm size (arable land per person) and productivity. Policy implications include implementing climate adaptation and mitigation via climate-smart practices, enhancing fertilizer and nitrogen-use efficiency, expanding and managing cropland sustainably (including dry-season/irrigated farming), and developing carbon sinks (e.g., afforestation). Coordinated institutional, technical, and financial strategies are necessary to lower agricultural emissions while maintaining or raising yields, thereby improving food security and supporting a just, environmentally sustainable development path.

- Scope and data: Macro-level national time series (primarily 1990–2022) may mask regional heterogeneity across Ethiopia’s agroecological zones; reliance on secondary data (World Bank) limits variable granularity (e.g., technology adoption specifics, management practices). - Model interpretation: Short-run ECM coefficients can sometimes display counterintuitive signs, complicating interpretation; residual normality tests yield mixed indications in the text, though overall stability diagnostics are supportive. - External validity: Findings are specific to Ethiopia’s structural context; generalizations to small versus medium/large farms require more data and nuanced analysis of local conditions. - Emissions–productivity mechanisms: Positive long-run elasticity for N₂O likely reflects nitrogen input effects; disentangling agronomic benefits from environmental externalities requires richer micro-level and process-based data. - Omitted factors: Potential influences such as irrigation intensity, technology adoption detail, input quality, and policy shocks are not explicitly modeled.