< Back to previous page

Project

Multi-target methodologies for the improvement of agricultural systems research - Study cases at system and field level

Several approaches and methodologies have been proposed in the literature to address the technical aspects of resource optimization and the improvement of management practices in agricultural systems. Similarly, field experimentation has benefited from multispectral imagery, obtained by unmanned aerial vehicles (UAVs), which allow to obtain crop data with cost-efficient nondestructive measurements. The overall objective of this PhD thesis was to develop and extend generic and multi-target quantitative methods to study the technical sustainability of agricultural systems (at the system level), and the statistical modeling of UAV-multispectral imagery (at field level) leading to the optimization of open-field agricultural experiments. This PhD applies the developed methodologies to different study cases.

The system analysis sections examine the potential environmental impacts and the most relevant biophysical factors explaining the yield gap and yield variability for potato cropping systems in the Central Peruvian Andes. The methods used were life cycle assessment, multivariate statistics, data envelopment analysis and crop simulation modelling. Taking into account the variability in potato production strategies, important environmental impact values were found for acidification and eutrophication (per ton of potato fresh weight), caused by the inappropriate or sub-optimal use of fertilizer sources. The k-means clustering algorithm identified three groups mainly defined by the nature of the inputs used for fertilization: inorganic, organic and mixed oriented. Exploratory factor analysis demonstrated that the first and second latent variables were correlated with an inorganic- and organic-oriented agriculture respectively; the inorganic system was associated with high values of potential environmental impacts. Relative environmental efficiency was linked to the quantity and source of the inputs, showing that potential environmental savings can be reached if more balanced input sources (mix of organic and inorganic) are employed. Similarly, the average potato yield gap was 42.1%, showing there is an important difference that needs to be reduced. The heterogeneous crop management practices of smallholders resulted in high variability in the dry weight production (710 to 18885 kg ha-1). The classification tree identified that inorganic N is the main factor characterizing the yield gap. The methodology identified that large yield gaps (Fourth quantile) are described by low inorganic N and scarce human labour energy, while small yield gap (First quantile) were mainly described by high N-inputs (inorganic and organic). This classification will be helpful to target inputs and site-specific agronomic recommendations towards closing the potato yield gaps.

The field analysis sections examine the feasibility of nonlinear mixed models to analyze UAV-multispectral canopy vegetation index from cassava (without treatment effect) and tomato (with irrigation treatment effect) experiments. Diverse methods were adapted for this purpose: segmentation algorithms (simple linear iterative clustering), affinity propagation, mixed modelling and resampling techniques. Object-based image analysis based on oversegmented multispectral imagery represented a good approach to extract canopy information of individual plants (experimental unit). A three-parameter logistic growth curve (non-linear mixed model) was found to be well-suited to fit the cassava and tomato canopy curves of the normalized difference vegetation index (NDVI). Resampling analysis showed that a suitable accuracy in parameter estimation can be achieved with fewer experimental units which could result in smaller agricultural experimental designs (cassava experiment). Similarly, differences were observed from 100% of the actual evapotranspiration (ETc) for all the treatments (75, 125 and 150% ETc) at maturity stage, when the cumulative effect of the water doses was well-defined and  reached its asymptote for the tomato experiment. The diagnosis plots and the root mean squared error of the observed and fitted NDVI indicated the suitability of using the three-parameter logistic mixed model.

This PhD research contributed to enhancing and extending the agricultural systems approaches towards technical sustainability. Likewise, it was also demonstrated that UAV-multispectral imagery, analyzed by nonlinear mixed models, provided useful insights towards field experiment optimization and treatment effect studies at field level.

Date:25 Apr 2016 →  20 Dec 2019
Keywords:multi-target analysis, agriculture
Disciplines:Agriculture, land and farm management, Biotechnology for agriculture, forestry, fisheries and allied sciences, Fisheries sciences, Soil sciences, challenges and pollution
Project type:PhD project