Spatio-temporal incremental data modelling for multidimensional environmental analysis
View fulltext online
Citation:Song, L. (2014). Spatio-temporal incremental data modelling for multidimensional environmental analysis. Unpublished doc thesis. Submitted in partial fufillment of the requirements of the D. Comp. Program. Department of Computing and Information Technology. UNITEC Institute of Technology
Permanent link to Research Bank record:https://hdl.handle.net/10652/2478
A variety of environmental problems increasingly attract academic research in order to protect ecosystems and minimise negative effects on human health. Advanced computational environmental analysis technologies have the potential to detect, monitor and perhaps effectively control these problems. However, computational environmental analysis is a complex and difficult problem to solve. The problems associated with environmental analysis are the underlying data. These data are collected from sub-optimal positions in urban and rural areas. The limited number of monitoring stations in the network means that we are collecting enough samples over time, however, insufficient data samples are collected across the monitored area. We determined the difficulty associated with computational analysis of environmental data via a critical review of existing approaches in the literature. Our review confirmed the biggest problem was associated with data collection, including noise introduced into the data stream, the big data problem in the form of an endless data stream, and missing data samples caused by ineffective equipment or poor placement of the monitoring equipment. In this thesis, we document our research into computational environmental methods for addressing land usage and air quality problems. The detection of land use change is a process of identifying differences in the state of a phenomenon by observing time-lapsed landscape imagery. Motivated by a simple neural pattern recognition mechanism, we propose a novel “one-step-more” incremental learning change detection method. In this method, an agent discovers knowledge from the first image using pixel-level incremental learning. When we detect changes in the subsequent image, the discovered knowledge model is updated and ready for the next change detection iteration. This is what we have called the “one-step more” incremental learning method. Powered by incremental data modelling techniques, the system demonstrates the capability of continuously detecting time sequenced imagery. Additionally, the method is shown to be computationally inexpensive when initializing and updating the change detection model. Land encroachment monitoring is essential to assist the economic growth, sustainable resource use and environmental protection of a city. We investigate land encroachment on public parks in the area of Auckland New Zealand, in which the proposed “one-step-more” method employed to analyse 26 Auckland parks. The obtained average region of interest (ROI) detection accuracy is 99.91% on five popular park related objects i.e. fences, houses, parks, trees and roads. The effectiveness of the proposed method is demonstrated on four categories of encroachment: periimanent land cover & use, and temporary & physical boundary encroachment. We document two detailed and comprehensive investigations of computational methods for air quality analysis. The first investigation is indoor emission source detection and the second is outdoor air quality prediction. Emission source detection indoors is important when locating the possible origin of pollution that could have a negative effect on the health of occupants. Outdoor air quality prediction is essential when attempting to minimise the negative effect on the health of citizens and the ecosystem. Addressing indoor emission source detection, we propose a novel inter-pollutant correlation analysis technique. Unlike other documented solutions that analyze merely primary pollution, our method is further enhanced by calculating intrapollutant correlation coefficients for characterizing and distinguishing emission events. Extensive experiments show that seven major indoor emission sources are identified by the proposed method, including (1) frying canola oil on electric hob, (2) frying olive oil on an electric hob, (3) frying olive oil on a gas hob, (4) spray of household pesticide, (5) lighting a cigarette and allowing it to smoulder, (6) no activities, and (7) using an exhaust or ventilation system. Furthermore, our method improves the detection accuracy by a support vector machine, compared to the classification without data filtering, and with feature extraction of PCA and LDA. Addressing outdoor air quality predication, we propose a novel spatial dataaided incremental support-vector regression (SaIncSVR). We overcome some of the problems associated with other prediction models. Existing models often demand data to be presented in a more convenient form than real data facilitates, for instance complete data, time-series complete, or a specified data capture process making them inadequate when used for environmental prediction. We conduct extensive experiments of PM2:5 prediction over 13 monitoring stations in Auckland, New Zealand. The experimental results indicate that our SaIncSVR performs improved PM2:5 prediction. Our method shows the capability to overcome missing data problems, an improvement over many other documented techniques. We compare our method with a local incremental support-vector regression. Further for PM2:5 spatial prediction, we conduct experiments on outdoor air quality prediction using a data-driven Gaussian geometry model. This model shows promise on a limited number monitoring stations and warrants further investigations. We promote incremental data modelling, given its benefits in processing large amounts of data with parallelization capability in many dimensions, as a constructive solution to spatio-temporal computational environmental analysis. This research documents systematic and comprehensive research in the area of computational environmental analysis. For this study we investigated a variety of environmental problems, environmental data, change & emission source detection, pollution variation & distribution, and data modeling & prediction. The proposed computational methods could be useful for other environmental detection and prediction problems, especially those using large, asynchronous and spatial-temporal data.