Climate Change, Drought, and Agricultural Production in Brazil

OVERVIEW. Climate change is a major challenge facing humanity, and understanding its myriad effects is important for learning how to lessen its negative consequences. Climate change is expected to alter the frequency and magnitude of natural disasters. This paper is drawn from a larger research project that studies how climate change has affected one type of natural disaster, drought, and how this has impacted agricultural production, productivity and poverty in Brazil. The research project studies: 1) whether climate change is altering the frequency, duration, and severity of droughts over more than a century, 2) how droughts have affected agricultural production in the past 50 years, 3) whether droughts affect total factor productivity (TFP) in agriculture, 4) how different dimensions of drought cause poverty, and 5) to what extent droughts affect poverty through the causal channel of TFP in agriculture. This current paper draws from the first two topics. 1. MEASUREMENT AND ANALYSIS OF DROUGHT. The first part of the paper constructs a host of short- and long-run measures of drought that all incorporate potential evapotranspiration. Some prominent studies have analyzed the frequency, duration, and severity of drought using indicators that depend mainly on the rainfall regime or vegetation index. However, despite the increase in temperature in recent decades, changes in rainfall have not exhibited a clear trend, while evapotranspiration shows the same rising trend as temperature. Drought impacts are likely underestimated without considering all factors that are influenced by global warming. This section of the paper studies whether drought has increased in Brazil in several dimensions, checking the robustness to alternative measures. Our preferred measure is a Standardized Precipitation Evapotranspiration Index (SPEI) that is constructed to measure short-run droughts (between 3 and 12 months), and the frequency, duration, severity and extension of longer-run droughts (measured in 5- or 10-year windows). The data on temperature, precipitation, and potential evapotranspiration used to construct the drought indicators from 1901 to 2020 are drawn from the Climate Research Unit at the University of East Anglia (CRU-UEA). They provide monthly information at a 0.5 grid level, representing approximately 55 km2. Among the many interesting descriptive findings are: a) the severity of drought increased in level and variability in the second half of the 20th century, and the level more than doubled in the most recent decade; b) drought severity has increased much more in the North and Center-West Regions of the country; c) the duration, but not frequency, of drought has followed the same pattern as the severity; and d) a decomposition of the increase in drought severity reveals that it has largely been caused by rising evapotranspiration, not by changes in precipitation. 2. IMPACTS OF DROUGHT ON AGRICULTURAL PRODUCTION. The second part of the paper estimates the effects of drought on Brazilian agricultural production between 1974 and 2020. This section aims to understand how different types of drought measures based on the SPEI—such as long-run drought severity or duration, annual drought, and quarterly drought—differentially impact agricultural production and productivity. Because drought is a rare event, we calculate the distribution of impacts across municipalities. The data used to measure agricultural production—calculated with a Fischer quantity index—are drawn from the survey Municipal Agricultural Production (PAM) from the Brazilian Institute of Geography and Statistics (IBGE). The data cover the 69 principal crops in Brazil—33 annual and 36 perennial—by year and municipality from 1974 to 2020. The data used to construct the drought measures were described above. We explore a variety of models to control for time-invariant and time-varying local unobservables, as well as the lagged effects of drought. After testing, we settle on a model with municipal fixed effects and state level quadratic trends. Under the assumption that droughts are exogenous to agricultural producers at the municipal level, and that unobservables are adequately controlled for, we estimate the causal impacts of drought on municipal level agricultural production. We use the estimated coefficients to calculate the distribution of drought impacts across municipalities and years. We also explore heterogeneity in the distribution of impacts by crop type, biome and sub-period, as well as decompose the impact on production into productivity and area effects. Among the many interesting results, we highlight: a) droughts that take place in the first two quarters of the year have much stronger negative effects than droughts that happen later in the year; b) as expected, droughts impact annual crops much more than perennial crops; c) across biomes, and relevant to poverty, droughts that happen in the Caatinga (semi-arid Northeast) have much more severe effects in percentage terms than in the Mata Altantica (2nd) or Cerrado (3rd); d) a drought at about the 50th percentile of impacts in the Caatinga biome has about the same impact (-20%) on production as a drought in the 1st percentile of impacts in the Cerrado biome; e) in terms of the volume of output, droughts have the largest effect in the Cerrado; and f) the impact of droughts has been increasing over time. 3. SIMULATED IMPACTS OF DROUGHT FROM 2021 to 2100 . The third part of the paper, which is in progress, conducts simulation exercises. We rely on climatological models from the Coupled Model Intercomparison Project version 6 (CMIP6) for alternative scenarios of what might happen to precipitation and evapotranspiration, and thus our SPEI drought measures, to simulate the impacts of drought on agricultural production in Brazil throughout the 21st century. 4. CONCLUSIONS AND POLICY IMPLICATIONS In progress.