Keywords
Spatial clustering; machine learning; sustainable agriculture; decision-making; lattice data.
Location
Session B4: Environmental and Agricultural Modelling for Ecosystem Services
Start Date
13-7-2016 2:50 PM
End Date
13-7-2016 3:10 PM
Abstract
Accounting for the direct and indirect role of ecosystem services (ES) in supporting human activities is a priority among governmental targets to develop strategies of impact mitigation with the mutual effect of increasing human well-being. Although a great deal of research is currently being done on ES assessment, the volume of available data and the complexity of mechanisms make the decision-making process difficult. We highlight the strengths of unsupervised clustering for handling geographic multidimensional datasets. We explain why the outputs, both in qualitative and quantitative form, can help stakeholders to conceive sustainable policies. At the same time, we point out the fact that hierarchical clustering algorithms are currently still the most used in decision-making, although they rarely provide optimal results. Moreover, the spatial connectivity between the studied administrative areas is never taken into account, even though many phenomena occurring in a spatial unit influence the neighbouring areas. To illustrate these statements we present a case study in Luxembourg, which focuses on the chemical emissions of agricultural fertilisers and the socio- demographic outcomes. We present an ongoing R package combining hierarchical clustering and dynamic reallocation algorithms to improve the significance of results. It also integrates spatial statistics and advanced clustering methods to take into account the interconnectivity of spatial units. In addition, the algorithm computes the usual statistical criteria to assess the clustering quality. Unfortunately, statistical criteria never guarantee the true representation of the real world; and basing decisions on unrepresentative outputs might induce unfavourable consequences. We present a function using a variable selection strategy so to identify the best clustering through a trade-off between statistical criteria and decisional composite indicators - measuring the socio-ecological disfavour and the overall fertiliser impacts on ecosystems.
Included in
Civil Engineering Commons, Data Storage Systems Commons, Environmental Engineering Commons, Hydraulic Engineering Commons, Other Civil and Environmental Engineering Commons
Advanced clustering to integrate spatial socio-ecological components in decision-making: a case study on sustainable agriculture
Session B4: Environmental and Agricultural Modelling for Ecosystem Services
Accounting for the direct and indirect role of ecosystem services (ES) in supporting human activities is a priority among governmental targets to develop strategies of impact mitigation with the mutual effect of increasing human well-being. Although a great deal of research is currently being done on ES assessment, the volume of available data and the complexity of mechanisms make the decision-making process difficult. We highlight the strengths of unsupervised clustering for handling geographic multidimensional datasets. We explain why the outputs, both in qualitative and quantitative form, can help stakeholders to conceive sustainable policies. At the same time, we point out the fact that hierarchical clustering algorithms are currently still the most used in decision-making, although they rarely provide optimal results. Moreover, the spatial connectivity between the studied administrative areas is never taken into account, even though many phenomena occurring in a spatial unit influence the neighbouring areas. To illustrate these statements we present a case study in Luxembourg, which focuses on the chemical emissions of agricultural fertilisers and the socio- demographic outcomes. We present an ongoing R package combining hierarchical clustering and dynamic reallocation algorithms to improve the significance of results. It also integrates spatial statistics and advanced clustering methods to take into account the interconnectivity of spatial units. In addition, the algorithm computes the usual statistical criteria to assess the clustering quality. Unfortunately, statistical criteria never guarantee the true representation of the real world; and basing decisions on unrepresentative outputs might induce unfavourable consequences. We present a function using a variable selection strategy so to identify the best clustering through a trade-off between statistical criteria and decisional composite indicators - measuring the socio-ecological disfavour and the overall fertiliser impacts on ecosystems.
