Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Using machine learning to predict species diversity for agri-environmental results-based schemes.
Agri-environmental schemes are important to promote biodiversity-friendly farming practices. This thesis aims to propose a result-based scheme that uses machine learning to predict species diversity.
Background
We experience currently severe loss of our biodiversity. This biodiversity loss is also documented in our agricultural landscapes. Agri-environmental schemes can support farmers to implement practices that promote biodiversity. Result-based schemes are one alternative for policy makers to implement agri-environmental schemes (Mack et al. 2020). Result-based schemes compensate farmers, when specific pre-defined results are reached. The assessment of the results needs monitoring and indices that measure biodiversity at the landscape, farm, or plot level (Matzdorf et al. 2008, Tasser et al 2019). Recently, Bartkowski et al (2021) proposed an agri-environmental scheme based on modeled results, and provided a hypothetical example for soil functions.
Modeling results instead of measuring results can reduce farmer’s risk of participating in a scheme as s/he knows in advance if s/he gets paid when implementing particular practices on a specific area. Moreover, it might help when restoration of agricultural systems where the results can take several years. However, modelling biodiversity outcomes is highly complex and requires high precision to be suitable for agricultural policies. Machine learning can potentially contribute to model complex system and to increase the precision of predictions (e.g., Kosicki 2020). Thus, machine learning might facilitate result-based schemes that use modelled results. One drawback of machine learning in such setting is that the complex links between practices and modelled results are not transparent to farmers, which might reduce the acceptance of such schemes.
Method
i) Conceptual work on design of result-based payments based on machine learning and ii) machine learning. The thesis will us panel data from 150 grasslands. For machine learning in applied economics see Storm et al. (2020).
References
Bartkowski, B., Droste, N., Ließ, M., Sidemo-Holm, W., Weller, U., & Brady, M. V. (2021). Payments by modelled results: A novel design for agri-environmental schemes. Land Use Policy, 102, 105230.
Basso, B. (2021). Precision conservation for a changing climate. Nature Food, 2(5), 322-323.
Kosicki, J. Z. (2020). Generalised Additive Models and Random Forest Approach as effective methods for predictive species density and functional species richness. Environmental and ecological statistics, 27(2), 273-292.
Mack, G., Ritzel, C., & Jan, P. (2020). Determinants for the implementation of action-, result-and multi-actor-oriented agri-environment schemes in Switzerland. Ecological Economics, 176, 106715.
Matzdorf, B., Kaiser, T., & Rohner, M. S. (2008). Developing biodiversity indicator to design efficient agri-environmental schemes for extensively used grassland. Ecological Indicators, 8(3), 256-269.
Storm, H., Baylis, K., & Heckelei, T. (2020). Machine learning in agricultural and applied economics. European Review of Agricultural Economics, 47(3), 849-892.
Tasser, E., Rüdisser, J., Plaikner, M., Wezel, A., Stöckli, S., Vincent, A., ... & Bogner, D. (2019). A simple biodiversity assessment scheme supporting nature-friendly farm management. Ecological Indicators, 107, 105649.
Background We experience currently severe loss of our biodiversity. This biodiversity loss is also documented in our agricultural landscapes. Agri-environmental schemes can support farmers to implement practices that promote biodiversity. Result-based schemes are one alternative for policy makers to implement agri-environmental schemes (Mack et al. 2020). Result-based schemes compensate farmers, when specific pre-defined results are reached. The assessment of the results needs monitoring and indices that measure biodiversity at the landscape, farm, or plot level (Matzdorf et al. 2008, Tasser et al 2019). Recently, Bartkowski et al (2021) proposed an agri-environmental scheme based on modeled results, and provided a hypothetical example for soil functions. Modeling results instead of measuring results can reduce farmer’s risk of participating in a scheme as s/he knows in advance if s/he gets paid when implementing particular practices on a specific area. Moreover, it might help when restoration of agricultural systems where the results can take several years. However, modelling biodiversity outcomes is highly complex and requires high precision to be suitable for agricultural policies. Machine learning can potentially contribute to model complex system and to increase the precision of predictions (e.g., Kosicki 2020). Thus, machine learning might facilitate result-based schemes that use modelled results. One drawback of machine learning in such setting is that the complex links between practices and modelled results are not transparent to farmers, which might reduce the acceptance of such schemes.
Method i) Conceptual work on design of result-based payments based on machine learning and ii) machine learning. The thesis will us panel data from 150 grasslands. For machine learning in applied economics see Storm et al. (2020).
References Bartkowski, B., Droste, N., Ließ, M., Sidemo-Holm, W., Weller, U., & Brady, M. V. (2021). Payments by modelled results: A novel design for agri-environmental schemes. Land Use Policy, 102, 105230. Basso, B. (2021). Precision conservation for a changing climate. Nature Food, 2(5), 322-323. Kosicki, J. Z. (2020). Generalised Additive Models and Random Forest Approach as effective methods for predictive species density and functional species richness. Environmental and ecological statistics, 27(2), 273-292. Mack, G., Ritzel, C., & Jan, P. (2020). Determinants for the implementation of action-, result-and multi-actor-oriented agri-environment schemes in Switzerland. Ecological Economics, 176, 106715. Matzdorf, B., Kaiser, T., & Rohner, M. S. (2008). Developing biodiversity indicator to design efficient agri-environmental schemes for extensively used grassland. Ecological Indicators, 8(3), 256-269. Storm, H., Baylis, K., & Heckelei, T. (2020). Machine learning in agricultural and applied economics. European Review of Agricultural Economics, 47(3), 849-892. Tasser, E., Rüdisser, J., Plaikner, M., Wezel, A., Stöckli, S., Vincent, A., ... & Bogner, D. (2019). A simple biodiversity assessment scheme supporting nature-friendly farm management. Ecological Indicators, 107, 105649.
The goal of the thesis is twofold. First, providing a (short) theoretical framework of how modeled results based on machine learning could be integrated in result-based schemes (e.g., identification of advantages and disadvantages). Here connections to Bartkowski et al. (2021) and precision conservation (e.g., Basso 2021) can be made.
Second (main focus of the thesis), understanding if we can model biodiversity with high enough precision to be suitable for schemes, what model complexity is needed to achieve high enough precision, and quantifying what are risk implication for policymakers and farmers when we using modeled results (e.g., due to wrong prediction, no participation of farmers because of mis-specified model). To this end, an empirical case study of schemes for grassland conservation should be used.
The goal of the thesis is twofold. First, providing a (short) theoretical framework of how modeled results based on machine learning could be integrated in result-based schemes (e.g., identification of advantages and disadvantages). Here connections to Bartkowski et al. (2021) and precision conservation (e.g., Basso 2021) can be made. Second (main focus of the thesis), understanding if we can model biodiversity with high enough precision to be suitable for schemes, what model complexity is needed to achieve high enough precision, and quantifying what are risk implication for policymakers and farmers when we using modeled results (e.g., due to wrong prediction, no participation of farmers because of mis-specified model). To this end, an empirical case study of schemes for grassland conservation should be used.
Robert Finger (ETHZ: rofinger@ethz.ch), Sergei Schaub (sergei.schaub@agroscope.admin.ch). This thesis is jointly supervised by the AECP Group and Agroscope (Sergei Schaub)
Robert Finger (ETHZ: rofinger@ethz.ch), Sergei Schaub (sergei.schaub@agroscope.admin.ch). This thesis is jointly supervised by the AECP Group and Agroscope (Sergei Schaub)