Desarrollo de un modelo de evaluación de tierras en red neuronal (Sierra2) para la selección de especies arbustivas en la reforestación de zonas mediterráneas, un nuevo componente del sistema MicroLEIS
2007
Heredia Guerrero, N.
In order to obtain landscapes with greater diversity, attending so much to the type of native species as to the vertical strata of vegetation in a Mediterranean environment, a data processing tool (model Sierra2) was developed. As a part of the MicroLEIS DSS, it facilitates the selection and distribution of the matorral species, for punctual studies like major scale areas. The artificial neuronal network (ANN) procedure was selected for modelling, for its great capacity to predict results and its immunity to the noise. 39 representative species of the Mediterranean matorral were selected attending to their bioclimatic distribution and then, the geo-edapho-climatic requirements of each species were established (least common), so that all the species were characterized by the same number of 14 variables. The bibliographical information was complemented with expert information, through surveys to specialists in plant and soil sciences, and specific soil-plant relations. Finally, Sierra2 was calibrated and validated with ANN quickpropagation training algorithm, according to the previously formulated shrub system. This ANN was subdivided in two subnets: a) formed by the most determinant variables in the shrub species distribution and b) considering the remainder eleven variables. This last subnet contributes to the greater precision in the final percentage of suitable species for the specific zone that was evaluated in each application. From the results, the high reliability of the model is to stand out in the neuronal network developed for the selection of the most suitable shrub species in new application scenarios. Using the first subnet, a spatial application was realized in the Cordoba region. The inputs are maps and the outputs 39 forest aptitude and forest sustainability maps.
显示更多 [+] 显示较少 [-]