Modeling regional supply responses using farm-level economic data and a biophysical model: a case study on Brazilian land-use change
by Samuel Ferreira Balieiro
Date of Examination:2021-10-18
Date of issue:2021-11-04
Advisor:Prof. Dr. Folkhard Isermeyer
Referee:Prof. Dr. Folkhard Isermeyer
Referee:Prof. Dr. Achim Spiller
Files in this item
Name:eDiss_FerreiraBalieiro_Oct2021.pdf
Size:8.56Mb
Format:PDF
Description:Dissertation
Abstract
English
Estimating farmers’ supply responses to changes in framework conditions is important to inform decision-makers on the expected impacts on production volume as well as the resulting land-use shifts. Existing agricultural supply response models generally require either larger databases with farm-level data for microregional analysis or are implemented with a coarse resolution (e.g., country level) due to the lack of data. While such approaches are suitable for regions with abundancy of data or for global-scale analysis, there is a need for an alternative for micro-level analysis in countries with low data availability. In addition, it is important to include the spatial component in the regional supply response analysis, allowing not only the quantification of the overall change in output but also the likely spatial land-use change. Against this background, this dissertation aims to answer the research question whether a combination of a biophysical model with farm-level economic data can be used to estimate farm-level profitability of individual crops and respective cropping systems and thereby simulate farmers’ supply responses in countries with limited data availability. To answer this question, a new modeling approach called Profitability Assessment Model (PAM) is developed, tested and validated. This new modeling approach follows the principles of minimum data, focusing on delivering timely and quantitative analyses with satisfactory accuracy to inform decision-makers. That is an important feature since the overall goal of the concept is to limit the data required by the model to a minimum, allowing quick implementation while accepting moderate accuracy. The PAM is a spatially explicit model with simulation units’ size of spatial resolution grid varying between 5 and 30 arcmin (10x10 to 50x50 km in area), following that used by the Global Biosphere Management Model (GLOBIOM). PAM estimates the profitability of each farming alternative at the simulation unit level and allocates the land to maximize farmers’ return to land. The PAM model is developed and calibrated for the Brazilian agricultural sector. Using Brazil as the case study is interesting due to its overall importance in the global production of agricultural commodities as well as the environmental impact of land-use changes. For this case study, four production system are represented in the PAM model: (a) double cropping of soybeans and maize, (b) soybeans with a cover crop, (c) sugarcane monoculture and (d) beef production. While the profitability of the arable crops is endogenously estimated, beef is considered as an opt-out option, which is modeled based on exogenous return-to-land information. Since soybean, maize and sugarcane production accounts for 84% of the total seeded area in Brazil, the current version of the PAM model represents the most important cropping alternatives to farmers in Brazil, but not all. An important methodological contribution of the dissertation is the development of routines for the extrapolation of each production cost component from the known typical farms’ data to all regions in the country. These routines are based on local expertise as well as existing information on yield levels, prevailing production systems and farming conditions. Each cost component is analyzed individually and, based on theoretical discussions, specific cost functions are proposed following the expected behavior of each cost item – e.g., linear relationship with yields or fixed per ha. That should improve the accuracy of the model in estimating production costs (and finally profitability) while also allowing the model to be adapted to simulate changes in framework conditions that may affect only selected cost items (e.g., a significant increase in fuel prices). In addition, the PAM model improves on existing models because it accounts for specific cost components such as the transport of sugarcane from farm to mill, which is required due to the perishability of the crop. Besides the important impact of inbound transport cost on the overall profitability of sugarcane production, the endogenous simulation of this cost item allows the model to spatially differentiate among regions depending on the current availability of mills. A major constraint for regional profitability analysis is the lack of information regarding farm input and output prices. To overcome this problem, the PAM model provides an interesting alternative by endogenously estimating prices via the transport module. By considering the different transportation costs of each crop and basing the distance estimation on the actual availability of roads, the model allows a straightforward conversion of reference prices to farm-gate prices. The ability to endogenously simulate transport cost is a useful feature for the simulation of scenarios based on price shocks. Apart from the development of the modeling approach, this dissertation focuses on the quantitative model validation as a key step to identify strengths and limitations of the concept. Projected yields are validated against regional statistics and production cost estimates are benchmarked against the two available datasets, with a suitable number of primary typical-farm data. Furthermore, the resulting land-use maps are evaluated against two simplified validation maps representing current land use. In the business-as-usual scenario, the PAM model estimates a national weighted average of returns to land of 248 USD/ha for double cropping and 188 USD/ha for sugarcane. This relationship, however, is different in the states of Sao Paulo and Minas Gerais, where, on average, sugarcane has a higher return to land than double cropping. Benchmarking PAM’s production cost estimates with observed local data shows a satisfactory model accuracy with a relative mean absolute error (rMAE) lower than 14%. The lowest error found in the production cost estimation is in sugarcane (rMAE of 8.7%) and the highest in second-crop maize (rMAE of 14%). The validation of the business-as-usual land-use map shows that the PAM model is able to satisfactorily reproduce the current land use in Brazil. The visual and quantitative validation results show a strong correlation between the available land-use maps, with PAM allocating the same crop as observed in 86% of total arable land. To test the ability of the PAM model to predict land-use and output changes due to changing framework conditions, a scenario analysis is carried out: What will happen in case yields of key crops change significantly as a consequence of climate change? Due to the strong reduction in the returns to land for grains (i.e., maize and soybeans) in the tropical region more than 24% of the current arable land is simulated to move from grains to sugarcane production. These results, however, vary significantly in the different regions, where the most affected states are Goiás, Paraná and Mato Grosso, jointly accounting for more than 55% of the total land-use change. This dissertation contributes to the overall development of regional farmers’ supply response models for countries with limited data availability, showing that it is feasible to combine a biophysical model and farm-level economic data as the basis for the profitability estimation in a high spatial resolution. The ability to estimate individual cost components separately gives the model the required flexibility for the simulation of market- and policy-related questions, providing timely and accurate information for decision-makers. The bottom-up approach based on local expertise is an important strength of the PAM model, avoiding unrealistic parametrization and ensuring that the majority of local features of production systems are included in the estimation. Finally, considering the overall goal of using minimum data, the model accuracy indicates a strong potential of the model to answer research questions, with additional parametrization and integration expected to further improve its performance.
Keywords: Brazil; Land-use change; Biophysical model; Production costs; Supply analysis; Soybeans; Maize; Sugarcane; Climate change; Profitability