1. Least Cost Control of Agricultural Nutrient Contributions to the Gulf of Mexico Hypoxic Zone Sergey Rabotyagov, Todd Campbell, Manoj Jha, Hongli Feng, Philip W. Gassman, Lyubov Kurkalova, Silvia Secchi, and Catherine L. Kling. Center for Agricultural and Rural Development, Iowa State University October 2008

2. 2 How do we solve this complex problem? Enormous number of farm fields/decision makers Enormous number of farm fields/decision makers Each can have one or more land use/conservation practice Each can have one or more land use/conservation practice Retire land (e.g., CRP) Retire land (e.g., CRP) Reduced, mulch, or no till Reduced, mulch, or no till Terraces Terraces Contouring Contouring Grassed Waterways Grassed Waterways Nutrient management: reduce fertilizer, better timing, etc. Nutrient management: reduce fertilizer, better timing, etc. Costs and effectiveness of practices can vary across locations Costs and effectiveness of practices can vary across locations

3. 3 CARD-UMRB Model: Economics, Land use, and Water Quality Models can help answer questions, evaluate alternatives, play out scenarios that real world cannot Model of Upper Mississippi River Basin – water quality and land use Unit of analysis: National Resources Inventory “point” in the UMRB as unit of analysis (field) 110,000 total “points” and expansion factors, 37,500 cropland observations Data from many sources to describe each location: land use, weather, crop history, tillage, slope, HEL, existing conservation practices, costs, prices, yields, etc. SWAT component A hydrologic and water quality model developed by USDA-ARS A hydrologic and water quality model developed by USDA-ARS Watershed-scale simulation model, operates on a daily time step, assess the impact of different management practices on water quality Watershed-scale simulation model, operates on a daily time step, assess the impact of different management practices on water quality Gassman et al. (2007) identifies over 250 publications using SWAT Gassman et al. (2007) identifies over 250 publications using SWAT Economics component Economics component Cost of adoption practices Cost of adoption practices Profits and revenues from alternative crops Profits and revenues from alternative crops

4. 4 189,000 square miles in seven states, 189,000 square miles in seven states, dominated by agriculture: 67% of total area, dominated by agriculture: 67% of total area, > 1200 stream segments and lakes on EPAs impaired waters list, > 1200 stream segments and lakes on EPAs impaired waters list, SAB Report: 43% of N and 41% of P delivered to Gulf SAB Report: 43% of N and 41% of P delivered to Gulf The Upper Mississippi River Basin

5. 5 Using Models to inform Policy Using a water quality model, Evaluate water quality effects of a configuration of conservation practices Estimate the costs of the set of practices But how to choose which set of practices is best (and what policies would be needed to get those changes in place)? But how to choose which set of practices is best (and what policies would be needed to get those changes in place)? Could evaluate lots of different alternatives to find most cost-efficient Could evaluate lots of different alternatives to find most cost-efficient Using water quality model, analyze all the feasible scenarios, picking cost-efficient solutions Using water quality model, analyze all the feasible scenarios, picking cost-efficient solutions But, if there are N conservation practices possible for adoption on each field and there are F fields, this implies a total of possible N F configurations to compare But, if there are N conservation practices possible for adoption on each field and there are F fields, this implies a total of possible N F configurations to compare 30 fields, 2 options  over 1 billion possible scenarios 30 fields, 2 options  over 1 billion possible scenarios

6. 6 One possible watershed configuration a d b a b c a d a b a a a 13 Fields 4 conservation practices 13 4 =28561 possible configurations Genetic Algorithm lingo Field = gene Practice options =allele set watershed configuration = individual (described by set of genes) Population = set of configurations

7. 7 Algorithm flow diagram Individual = watershed configuration = specific assignment of practices to fields Population = set of watershed configurations

8. 8 Fitness assignment example Strength S(i)= # of individuals i dominates Strength S(i)= # of individuals i dominates Raw fitness R(i)= sum of strengths of individuals that dominate i Raw fitness R(i)= sum of strengths of individuals that dominate i

9. 9 Pareto frontier: UMRB

10. 10 Selection of individuals for a 30% reduction in N or P

11. 11 Consequences of seeking a 30% reduction in NO 3 Conservation and Land use to achieve reduction Conservation and Land use to achieve reduction N fertilizer reductions N fertilizer reductions grassed waterways (extensive) grassed waterways (extensive) terraces (combined with N fertilizer reductions) terraces (combined with N fertilizer reductions) additional (substantial) land retirement additional (substantial) land retirement A 30% reduction in outlet NO 3 automatically leads to a 35% reduction in outlet P A 30% reduction in outlet NO 3 automatically leads to a 35% reduction in outlet P The annual additional cost is estimated to be The annual additional cost is estimated to be $ 1.4 billion (more than quadrupling baseline cost)

12. 12 Final Remarks CARD-UMRB model can help quantify tradeoffs between: CARD-UMRB model can help quantify tradeoffs between: cost and pollution reductions cost and pollution reductions different pollutants different pollutants Many assumptions and caveats, but Many assumptions and caveats, but The model is flexible and amenable to improvement The model is flexible and amenable to improvement Need to keep in mind purpose of modeling “All models are wrong, some are useful,” George Box Need to keep in mind purpose of modeling “All models are wrong, some are useful,” George Box Policy Role: Can we really set policy based with modeling results?? Yes and no Policy Role: Can we really set policy based with modeling results?? Yes and no