Contributed Session 1: Spatio-Temporal Methods in Environmental Analysis

Presence-only data represent “point locations where a species has been recorded as present”. Species distribution modelling (SDM) methods, including point processes, are used to analyse such data. We use point processes to classify taxonomically unknown events using local indicators of spatio-temporal association (LISTA) and to fit thinned inhomogeneous multitype Poisson process models before and after classification. The methods are applied to data on fish eggs belonging to Gadus morhua and Melanogrammus aeglefinus, collected near the Gulf of Maine between 1995 and 1999.

This work introduces a penalized spatio-temporal quantile regression model to analyze tail behaviors of data manifesting complex spatio-temporal patterns. This approach involves minimizing a suitable functional that combines a data misfit term with a separable space-time penalty defined through Partial Differential Equations. The proposed model is particularly well-suited for environmental studies, such as those on air pollution, where data often display pronounced skewness and spatio-temporal heterogeneity.

Accurately forecasting people’s mobility among different areas is crucial for urban planning, transportation optimization, and emergency management. Applying a trivariate VARX+DHR model to one year of hourly mobile phone traffic data in a high-flood-risk area of Brescia, we achieved reasonable forecasting accuracy but observed non-Normal residuals with heavy tails and time-varying heteroskedasticity. To address this issue, we applied the multiplicative component standard (mcs) GARCH model to decompose volatility into daily, hourly, and intradaily components.
These volatility series were used for time series clustering, employing the fuzzy approach to identify similar mobility patterns across the considered areas.