6 December 2015
By: David N. Barton, Tamara Benjamin, Carlos R. Cerdan, Fabrice
DeClerk, Anders L Madsen, Graciela M. Rusch, Alvaro Salazar, Dalia
Sanchez, Christobal Villanueva
The paper can be accessed here.
An expert system for the design of ideal livestock pastures using combinations of trees in live fences and on pastures was developed using Bayesian belief networks. The tool was created with information from Rivas, Nicaragua, on local farmer knowledge on tree species, trees costs and benefits, farmers. expressed needs and aspirations, scientific knowledge about functional traits in the design of silvopastoral systems. The tool identifies combinations of multifunctional trees which improve ecosystem services provision from pastures.
We demonstrate how the Bayesian inference and updating features of BBNs are attractive features of an agroecological expert system. Using Bayesian inference, field staff can use the expert system for real-time identification of portfolios of multifunctional trees that satisfy a profile of desired ecosystem services by the farmer.
The network can be used to reason deductively from farmer characteristics through adoption of trees species, to probable ecosystem services. The network can be used inductively from desired ecosystem services, through likely morphological traits and trees species, to likely farmer characteristics (Figure 2).
The online demo network is shown for the nodes related to trees in pastures (nodes for trees in live fences are not available to keep the demo simple).
Figure 2 shows how Bayesian Belief Networks (BBN) integrate different knowledge domains in the study of agrosilvopastoral systems.
The network can be used to reason deductively from farmer characteristics through adoption of trees species, to probable ecosystem services.
The network may be used to reason inductively from desired ecosystem services, through likely morphological traits and trees species, to likely farmer characteristics.
The online demo network is shown for the nodes related to trees in pastures (nodes for trees in live fences are not availablbe to keep the demo simple). The structure of the entire model can be found here.
The example is based on [Barton et al, 2015]. Field data is based on [Marie 2010, Mosquera 2010, Salazar 2012].
Below are some HUGIN widgets for interacting with the model shown on the right (click on the probability bar to instantiate a node or remove evidence).
Another dimension of the BBN is the display of farmer knowledge on ecosystem disservices of trees in pastures.
Another dimension of the BBN is the use of a network to display terminology translation between local knowledge and expert knowledge on functional traits and ecosystem services
Useful references include:
[Kjærulff, U. B. and Madsen, A. L. (2013)] Bayesian Networks and Influence Diagrams: A Guide to Construction and Analysis. Springer, Second Edition.
[David N. Barton, Tamara Benjamin, Carlos R. Cerdan, Fabrice DeClerk, Anders L Madsen, Graciela M. Rusch, Alvaro Salazar, Dalia Sanchez, Christobal Villanueva (2016)]. Assessing ecosystem services from multifunctional trees in pastures using Bayesian belief networks. Ecosystem Services, Vol 18, pages 165-174. Online.
[Marie, C. (2010)]. Of trees and pastures - management of woody resources and the social dynamics in La Chocolata, Rivas, Nicaragua. CIRAD, Montpellier, France.
[Mosquera, D. (2010)]. Conocimiento local sobre bienes y servicios de especies arboreas y arbustivas en sistemas de produccion ganadera de Rivas, Nicaragua. CATIE, Turrialba, Costa Rica.
[Salazar, A. (2012)]. Modelo experto para el analisis de la adopcion de arboles en pasturas del tropico seco de Nicaragua. CATIE, Turrialba, Costa Rica.
For further details contact: David Barton (David(dot)Barton(at)nina(dot)no)