PREDICTING FUTURE LAND DEGRADATION AND ITS ECONOMIC EFFECTS -Evidence From an Econometric ApproachBonn, April 10, 2013 UNCCD 2nd Scientific Conference Alex De Pinto Akiko Haruna Tingju Zhu International Food Policy Research Institute Purpose of the study “Land degradation is silent emerging process that increases the risk for the livelihood of millions” “Prevention is less costly than restoration” Create a tool that allows to “reasonably” predict land degradation and to prioritize action. Establish a link between LD, climate change and food security. Major drivers of land degradation What the literature says: Physical drivers Socioeconomic drivers Climatic factors Domestic Policy + -Rainfalls + Institutional capacity + -Rainfall intensity - Land tenure/property rights +/- -Wind - Technology +/- Slope - Information + Vegetation cover + Population density -/+ Soil type Fire Economy - Selected source: Huber et al. 2011; Begue et al. 2011; Zhao et al. 2010; Safriel and Adeel 2005; Ravi et al. 2010; Le et al. 2012; Sonneveld and Keyzer 2000; Vogt et al., 2011; Pardini et al 2004; Eswaran et al. 2001; Young, 2001; Mitchell 2004; Kassam et al. 2009; Jansen et al. 2006; K.J. Wessels et al. 2007; Tesfey, 2006; Geist and Lambin 2004; Pender, Place and Ehui, 2006, Hagos and Holden, 2006; Mulvaney, Khan, and Ellsworth 2009; Nkonya et al. 2004; Boyd and Slaymaker 2000, Pretty et al. 2011; Benin et al 2007; Bai et al. 2008; Vlek et al. 2010; Nachtergaele et al. 2010; Moti Jaleta, Menale Kassie, 2012; Zimmerman et al., 2003; Li and Reuveney, 2006 -Market access +/- -Livelihood diversification +/- -Poverty -/+ -Economic growth + + : beneficial for prevention of LD - : drive LD + / - : ambiguous NDVI: Proxy for LD and Land Carrying Capacity • Advantage of NDVI • Global coverage • Single index with readily available dataset • Excellent temporal and spatial extensions • Weakness of NDVI • Coarse resolution (for non-global analyses) • Accuracy of observations • Differentiation from land cover and land use and other human interventions • Truly representative of land degradation? Approach • Dependent variable: max NDVI (average 2002-2006) • We use cross-section data • Exploit spatial rather than temporal variability • Increases the richness of the data set • Explanatory variables include major drivers of land degradation • Model (OLS, quadratic form): • ܻ ൌ ࢼ ࢼଵ ࢄ ࢼଶ ࢄଶ ߳ • Where ࢄ is a location specific vector of climatic, geophysical, and socioeconomic variables and ࢼ a vector of parameter estimates Model and data details • Dataset: global level • Covariates: Climatic, Geophysical and socioeconomic variables • Control on other ecological factors: 6 AEZ-LPG (length of plant growth) dummies • Control for irrigations: Irrigated vs. cultivated land ratio using IFPRI’s Spatial Production Allocation Model (SPAM) • Control for potential spatial correlation: regular sampling method (3x3 grids) Variables for model estimation Variable Resolution Period Max. NDVI 0.083o x 0.083o 2002–2006 Avg. Precipitation 0.54o x 0.54o 2002–2006 Rainfall intensity (# of events 1 S.D. above mean) 0.54o x 0.54o 2002–2006 Avg. Temperature 0.54o x 0.54o 2002–2006 Slope Soil Organic Carbon Population density Access to market 0.008o x 0.008o 0.5o x 0.5o 0.5o x 0.5o 0.008o x 0.008o Data source GIMMS-AVHRR dataset (Global Land Cover Facility) Climate Research Unit (CRU), University of East Anglia Climate Research Unit (CRU), University of East Anglia Climate Research Unit (CRU), University of East Anglia GMTED2010, USGS 2000 2000 Avg. GDP growth rate Country level 2002-2006 Avg. input usage Country level 2002-2006 Avg. Rules of Law Country level 2002–2006 Infant Mortality Rate Regional level 2000 Hiederer et al (2012) CIESIN Uchida and Nelson (2009) UNSTAT constant 2005 prices The World Development Indicators Worldwide Governance Indicators CIESIN OLS regression coefficients Dependent variable: max NDVI – Range [-1, 1] Expl. VARIABLES Precipitation Above mean and 1 S.D. of precipitation Temperature Slope Soil Organic Carbon Population density Access to market Infant Mortality Rate GDP growth rate Rule of law Input usage Irrigated area AEZ dummies Coefficient 0.0002** -0.008** -0.005** -0.007** 0.001** -4.49e-05** -2.63e-05** -0.001** 0.009** -0.011** -0.0002** -0.045** 0.217**; 0.262**; 0.295**; 0.291**; 0.307** Observations 211,332 R-squared 0.69 ** p<0.01, * p<0.05, + p<0.1 OLS regression coefficients Dependent var: NDVI maximum VARIABLES Precipitation Above mean and 1 S.D. of precipitation Temperature Slope Soil Organic Carbon Population density Access to market Infant Mortality Rate GDP growth rate Rule of law Input usage Irrigated area Observations R-squared Coefficient 0.0002** -0.008** -0.005** -0.007** 0.001** -4.49e-05** -2.63e-05** -0.001** 0.009** -0.011** -0.0002** -0.045** 211,332 0.689 VARIABLES Precipitation2 Above mean and 1 S.D of precipitation2 Temperature2 Slope2 SOC2 Pop density2 Access2 IMR2 GDP growth rate2 Rule of law2 Input usage2 Irrigated area2 Coefficient -3.54e-08** -0.0003** 3.98e-05** 0.0001** -2.61e-06** 1.30e-09** 5.22e-10** 6.37e-06** -0.0006** 0.010** 6.69e-08** 0.066** AEZ dummies Constant 0.394** Future scenarios • We use our ࢼ estimates to project NDVI into 2050 (Logic: if predicted NDVI ↓ land degradation ↑) • Year 2050; IPCC SRES A1B scenario; 2 GCMs (CSIRO-Mk3.0 and MIROC 3.2) Variable Resolution Period Data source Precipitation 0.54o x 0.54o 2050 Jones et al. (2009), Downscaled IPCC-AR4 GCMs Rainfall intensity (1 S.D. above mean precipitation) 0.54o x 0.54o 2050 Jones et al. (2009), Downscaled IPCC-AR4 GCMs Temperature 0.54o x 0.54o 2050 Jones et al. (2009), Downscaled IPCC-AR4 GCMs Population density 0.5o x 0.5o 2050 UN World Population Prospects GDP growth rate Country level 2050 IMPACT pessimistic scenario Input usage (fertilizers) Country level 2050 (Wood et al, 2004) Mortality rate Regional level 2050 Derived from IMPACT malnutrition estimate Slope, SOC, Access to market, Rule of law Constant from the 2000’s Areas with declines in NDVI 2000 – 2050 Socioeconomic variables only Climatic variables have an impact MIROC pessimistic scenario Climatic variables have an impact CSIRO pessimistic scenario Implication for food security • Estimation of global calorie production: • 16 major food crops (wheat, rice, maize, barley, millet, sorghum, potato, sweet potato, cassava, banana and plantain, soybean, other beans, other pulse, sugar cane, sugar beet, ground nuts) • Yield and harvest area: 0.083o x 0.083o spatial dataset (SPAM) • Calorie per unit of product (USDA, FAO) • Calorie production per pixel = ∑ {Calorie per unit of product X yield per product per area X harvest area per product per pixel } • Large NDVI decline with major food production area • Areas with NDVI change below mean of all negative NDVI changes • Areas with calorie production above mean of all calorie productions Food security implication: MIROC Identified areas with below mean of negative NDVI changes and areas with above mean of calorie production MIROC pessimistic scenario: 116 million ha of above-average production cropland affected (current output: 65 billion USD per year) Food security implication: CSIRO Identified areas with below mean of negative NDVI changes and areas with above mean of calorie production CSIRO pessimistic scenario: 105 million ha of above-average production cropland affected (current output: 54 billion USD per year) Food security implication: Dark Red: Areas with below mean of negative NDVI changes and areas with above mean of calorie production MIROC pessimistic scenario: 116 million ha of above-average production cropland affected (current output: 65 billion USD per year) Food security implication: Dark Red: Areas with below mean of negative NDVI changes and areas with above mean of calorie production CSIRO pessimistic scenario: 105 million ha of above-average production cropland affected (current output: 54 billion USD per year) Food security implication: Dark Red: Areas with negative changes in NDVI greater than 10% and areas with above mean of calorie production CSIRO pessimistic scenario: 13 million ha of above-average production cropland affected (current output: 11 billion USD per year) Food security implication: Dark Red: Areas with negative changes in NDVI greater than 10% and areas with above mean of calorie production MIROC pessimistic scenario: 15 million ha of above-average production cropland affected (current output: 13 billion USD per year) Other possible important implications: Biodiversity Hotspots: at least 1,500 species of vascular plants as endemics, and it has to have lost at least 70 percent of its original habitat Source: Conservation International Other possible important implications: Predicted LD hotspots (MIROC) and biodiversity hotspots Conclusions • One step towards a predictive tool for LD and the inclusion of climate change effects • Substantial food production areas potentially affected by LD • Climate change appears to exacerbate LD in certain areas • More work on linking NDVI to reality on the ground (recent work of Bao, Vleck, and others) • A call for a close collaboration among scientists from different disciplines