Modelo de apoio a decisão ao planejamento territorial de silvicultura baseado em análise multicritério de redes neurais artificiais

Abstract

As landscape became humanized biodiversity declines due to habitat depletion and the conservational biodiversity-oriented landscape planning can no longer be ignored. Nonetheless, economic activities are equal important and can’t be put apart. Over the last decade of the XXI century there is a growing need for educated decision making regarding planning policies that combine social, economic, cultural and environmental features. Nowadays, spatial planning policies have been enhanced by embedded modern technologies In this study we propose a bottom-up approach methodology for forestry planning development. To pursue this, several concepts, methods and technologies are assembled together, such as such as artificial neural networks (ANN), namely Multilayer perceptron (MLP), who shows to be a promising alternative to regression techniques, Geographic Information Systems (GIS), Multicriteria analysis and Scenario building. Finally, a decision support model for forestry planning through multicriteria artificial neural network is set up. The watershed of Ijuí is located between the geographical coordinates 27 º 45 - 26 º 15' N and 53 º 15'-56 º45' E, and have 10.731,86 km2 in area. The methodology focuses on the analysis of predictive scenarios ("what if") aimed at assessing silviculture activity against other stocks of interest over the territory. Based on the21 variables selected four actions has been take in account as the most representative of the different agents interest involved namely: no action taking place on the territory (Nothing), increased forestry activity (ExFlo), activity growth agricultural (ExAgri) and the action that advocates the conservation of native forests (Cons). For these four actions, five scenarios was setting up: Current Condition (CA), economic growth (+ Econ), economic degrowth (-Econ), positive social impact (+Idese) and negative social impact (-Idese). The training of ANN was performed with the software Statistica ® 12.0 to perform multilayer perceptron network (MLP) to run different parameters for the different scenarios. Therefore, for the Nothing scenarios MLP parameters were two hidden layers and 10 hidden neurons, and for the scenarioswithExAgri, Cons and ExFlo actions, only one hidden layer and ten hidden neurons were used. Performed in a GIS environment, actions spatial simulations occurring under different scenarios resulted in twenty predictive output’s. Results analysis show that in not predicting any action, areas with native forest tend to decrease over the territory in whatever scenario, so as forestry use, these latter analysis may not occur if a scenario is -Idese. For ExFlo action scenarios that were presented less favorable -Idese e-Econ. However, for the action Cons, the only favorable to increased forestry activity scenarios were CA and + Idese and the latter case also presented favorable for ExAgri action. In synthesis, this study showed that a bottom-up approach model combining statistical methods associated with GIS, in particular ANN, is able to capture and model the complexity of forestry planning trough predictive modeling, presenting future alternative scenarios based on local agents interests or actions, and provide with proven accuracy results to feed educated policies guidelines Although, the method discussed is presented as an alternative tool of territorial planning, however, it requires consideration of its limitations as a mechanism of abstractions from the reality of events over the territory. Therefore, it is an obvious certainty, the need for research institutions in geo and the numerous educational institutions to invest in research and generation of knowledge-based models for territorial analysis with the use of ANN in order to make all these more streamlined procedures for predicting, reliable and performed with smaller costs.