Nonparametric first-order analysis of spatial and spatio-temporal point processes: application to wildfire patterns


Wenceslao González Manteiga


Praia de Botafogo, 190 - sala 537


3 de Maio de 2018 às 16h

Spatial point patterns arise in a wide variety of scientific contexts, including seismology, forestry, geography and epidemiology.Wildfire is the most ubiquitous natural disturbance in the world and represents a problem of considerable social and environmental importance; particularly, in Galicia (NW Spain) arson fires are the main cause of forest destruction. Knowing the spatial distribution of forest fires would be a key factor for future development of fire prevention and fire fighting plans. Nonparametric estimation and bootstrap techniques play an important role in many areas of Statistics. In the point process framework, kernel intensity estimation has been limited to exploratory analysis due to its lack of consistency. This work addresses different procedures to obtain a consistent estimator of the first order intensity such as kernel estimation of the density of event locations and kernel intensity estimation based on covariates. We propose a smooth bootstrap procedure for inhomogeneous point processes in order to develop effective bandwidth selectors for kernel intensity estimation. The consistent estimators introduced above, are used to estimate the first order intensity of the wildfires registered in Galicia during the period 1999-2008. Finally this kind of estimators is used for two problems of interest:

a) The nonparametric comparison of first-order intensity functions and
b) One separability test for spatio-temporal point process. 

*Texto informado pelo autor. 


Wenceslao González Manteiga
Short-CV 1979. Degree in Mathematics. Santiago de Compostela. Emphasis: Statistics and Operation Research with “Special Award”. 1982. Ph. D. in Mathematics “Special Award”. University of Santiago de Compostela, Spain. Doctoral Thesis: “Consistency and asymptotic properties of non parametric density and regression estimators”. Supervisor: J.A. CristóbalCristóbal. Present: Full Professor of Statistics at the University of Santiago de Compostela.
Research Interests: Curve estimation; Time series; Neural Networks; Bootstrap; Spatial data analysis; Survival data analysis; Finite populations; Small areas; Biostatistics; Financial Econometrics, Environmental and Industrial Applications, Functional data, Goodnessof-fit tests for regression models.