Impacts of Climate Change and Extreme Weather on U.S. Agricultural Productivity Sun Ling Wang, Eldon Ball, Richard Nehring, Ryan Williams (Economic Research Service, USDA) Truong Chau (Pennsylvania State University) Prepared for presentation at the OECD meeting Agricultural Total Factor productivity and the Environment Paris, France, May 23-24, 2017
Background (I) TFP growth varied across regions (1960-2004), with some states growing faster than the others. Long-run TFP growth is mainly affected by R&D, Extension, and public infrastructure. TFP growth moves closely with output growth and fluctuates from year to year mainly in response to transitory events, such as adverse weather. 2
Background (II) Literature has shown that Climate change and extreme weather have affected crops and livestock production. Crops: higher variance in climate conditions leads to lower average crop yields and greater yield variability; temperature increase beyond a certain level can reduce crop yields; weather extremes can cause disease outbreaks and impact agricultural production. Livestock: when animals thermal environment is altered it can affect animal health and reproduction; the feed conversion rate can also be affected by changes in thermal condition; an increase in the temperature humidity index (THI) can result in technical inefficiency in livestock production. 3
Issues The frequency of adverse weather events has increased since 1980s (Parry et al. 2007; Hatfield et al., 2014). Measured productivity growth fluctuates dramatically from time to time, reflecting a drop or slower growth in agricultural output due to transitory events, such as adverse weather. Empirical studies have been focused on individual commodity responses to climate change. Little attention has been paid to aggregate level and sector-wide analysis in agriculture sector 4
Literature on Climatic Effects There are three major streams of literature studying the relationship between climate change/weather effect and economic activities. One focuses on biophysical impacts through examining the relationship between climatic factors and individual commodity production/productivity, such as crop yield or livestock production One focusses on adaptive response at the individual level through evaluating how an individual farm/firm/person reacts to climatic impacts The third stream of literature addresses impacts at a regional/ national/sectoral scale, considering both biophysical effects and adaptation. They are usually done by quantifying the effects of climate/weather changes on aggregate economic performance using country/regional level data or sectoral data 5
Objectives We examine how climate has changed in the U.S., using 1940-1970 historical weather data (mean and variation) as the norm. We use state panel data to study the impacts of climate change and extreme weather on U.S. agricultural productivity (technical inefficiency) empirically, from the entire farm sector aspect We compare the effects of different climate variables in explaining state technical inefficiency We project regional productivity performance in 2030-2040, using 2000-2010 as the reference period, based on climate change and extreme weather scenarios 6
Method (I) Stochastic Production Function Following Aigner, Lovell, and Schmidt (1977), Meeusen and van den Brocck (1977), and Bassete and Coelli (1995) we form the stochastic frontier production model as: lny it = ß 0 + K k=1 (4) J ß k ln x kit + ß t t+ j=1 N ß j D j + m=1 ß m D m + v it u it y is an implicit quantity of total output, x s are inputs t is a time trend, D j s are state dummy variables, and D m s are time dummy variables to capture the unobserved heterogeneity across regions and over time. The deviations (ε it ) from the frontier are composed of a two-sided random error (v it ~N(0, ơ v 2 )), and a one-sided error term (u it >=0). u it is assumed to be half-normally and independently distributed, with a truncation at zero of the normal distribution~(z it γ, ơ u 2 ); z it is a vector of exogenous variables, and γ is a vector of parameters to be estimated. 7
Method (II) Inefficiency measurement If u it =0, then state i is at the production frontier and is technical efficient. If u it > 0 then state i is deviated from the frontier, and is technical inefficient. Following Battese and Coelli (1995) and Alvarez et al. (2006) we estimate an inefficiency variance regression model simultaneously with equation (4), i.e. 2 lnơ uit = γ 0 + N n=1 γ n z nit + ω it ; ω it ~N(0, σ 2 ω ) (5) z s include climate variables, irrigated area ratio, and other local public goods R&D, Extension, and roads in alternative model specifications. The stochastic frontier is estimated by a maximum likelihood (ML) procedure. 8
Weather variables(i) We employ two different weather variables to capture climate s impacts on livestock and crops respectively, i.e. THI load and an aridity index Oury index. Following Thom (1958), St-Pierre, Cobanov and Schnitkey (2003); Zimbelman, et. al. (2009), Key and Sneeringer (2014) THI is defined as: THI=(dry bulb temperature o C) + (0.36*dew point temperature o C) + 41.2 THI load is the number of hours that the location has a THI above the threshold. Following Oury (1965), and Zhang and Carter (1997) the Oury index is defined as W s = P s 1.07 T s where W represents the aridity index (Oury idex), s is the month (s=1 12), Ps is the total precipitation for month s in millimeters; and Ts is the mean temperature for month s in degrees centigrade. 9
Weather variables (II) We construct THI norm and Oury norm estimates using weather data between 1941 and 1970. We also construct annual THI load means/standard deviations and Oury index means/standard deviations for each year and each state between 1961 and 2004. In addition to THI load and Oury index we also construct THI shock and Oury shock variables, which refer to the units of standard deviations from their historical norms. One standard deviation is referred to as one unit of shock. 10
Weather Data We draw monthly temperature and precipitation data at the county level from a weather dataset produced by Oregon State University s PRISM climate group (Daly et al. 2008). We construct state level weather variables using county level weather data and county livestock weights for THI estimates and cropland weights for Oury index. The livestock weights were developed using data from the Census of Agriculture. The cropland weights were developed by summing the cropland pixels from the National Land Cover Database 2006 (NLCD 2006) 11
Other Data Sources Quantity of agricultural output and inputs variables are drawn from USDA-ERS s U.S. agricultural productivity accounts. Irrigated area ratio is measured as irrigated land area divided by crop land area (USDA-NASS). We employ a cubic spline technique to interpolate the information between census years. R&D stock is constructed using public R&D investment data and following a trapezoidal-weight pattern (Huffman, 2009) Extension is a measure of extension capacity calculated as total full time equivalent (FTE) extension staff (NIFA) divided by the land area Road infrastructure is a road density index constructed using total road miles excluding local (e.g. city street) miles for each state divided by total land area (DOT) 12
Climate Conditions for Crops (I) Lower Oury index indicates much drier situation. If the Oury index is lower than 20, it indicates drought conditions, and if the Oury index is less than 10, the area is desert like. Aridity (Oury) Index varied across regions and over time. 13
Climate Conditions for Crops (II) Many eastern states experienced negative Oury index shocks in 1983 while other states in the southwest experienced a better climate condition for crops in 1983. Oury index shocks in 1995 show states within the Corn Belt region experienced drier conditions than their historical norms. 14
Climate Conditions for Livestock (I) A higher THI load index indicates a more intensive heat stress condition that can hinder livestock productivity. Different THI conditions could result in state-specific livestock production profiles. In general, variations of THI loads from year to year were less than variations in the Oury indexes, on average. 15
Climate Conditions for Livestock (II) Based on estimated THI shocks, many states experienced worse heat stress compared to their historical norms in 1983 than in 1995. 16
variables Results: Stochastic Production Frontier Estimates production function coefficient t-ratio coefficient t-ratio coefficient t-ratio coefficient t-ratio tech 0.0009 6.86 *** 0.0010 7.49 *** 0.0010 7.23 *** 0.0010 7.73 *** lncap 0.0813 4.10 *** 0.0775 3.96 *** 0.0705 3.59 *** 0.0785 4.11 *** lnmat 0.5959 44.35 *** 0.5952 45.24 *** 0.5920 45.49 *** 0.5879 45.42 *** lnlab 0.0982 10.60 *** 0.0998 10.98 *** 0.1089 11.57 *** 0.1079 11.66 *** lnland 0.1124 6.55 *** 0.1055 6.25 *** 0.1083 6.23 *** 0.0995 5.80 *** lnơ v 2 (noise) constant -5.8828-59.40 *** -5.8048-52.68 *** -5.8232-71.97 *** lnơ u 2 (inefficiency) Model 1 Model 2 Model 3 Model 4 constant -4.5181-26.02 *** -5.2706-25.05 *** -2.4305-1.14-2.7825-1.52 THI load 0.00002 1.31 0.00006 3.38 *** Oury index -0.0257-4.29 *** -0.0201-3.06 *** THI load shock 0.3087 5.40 *** 0.3073 5.25 *** Oury index shock -0.1831-2.15 *** -0.1831-2.15 *** Irrigation ratio -1.6170-2.89 *** -1.4210-1.93 *** -2.8771-3.45 *** -2.2217-3.01 *** LnR&D -0.3867-2.86 *** -0.3314-2.67 *** LnExtension -0.6245-3.71 *** -0.4787-2.69 *** LnRoad -0.8779-3.68 *** -0.7994-3.78 *** state fixed effects yes yes yes yes time fixed effects yes yes yes yes log-likelihood 2678.5 2697.9 2712.7 2725.6 X 2 (95) 16400000 prob>x 2 =0 11700000 prob>x 2 =0 15900000 prob>x 2 =0 14600000 prob>x 2 =0 observations 2112 2112 2112 2112
Conclusions State production data and climate information show noticeable variations across and within production regions. Higher THI loads and positive THI shocks can drive state production away from its best performance. Higher Oury index, and positive Oury index shocks can reduce state inefficiency. Higher irrigation ratio, and more abundant local R&D, Extension, and higher road density can help reduce production inefficiency. The unexpected climate shock variables seem to have more robust and consistent impacts than mean weather variables on technical inefficiencies. A certain change in the level of temperature or precipitation could result in different impacts on productivity in different regions or states. 18
Q&A Thank you! http://www.ers.usda.gov/data-products/agriculturalproductivity-in-the-us.aspx slwang@ers.usda.gov 19