Detectie van milieuveranderingen. Een toepassing van Structurele Tijdreeksmodellen en het Kalmanfilter
Visser H
Visser H
Citations
Altmetric:
Series / Report no.
Open Access
Type
Report
Language
nl
Date
2003-12-12
Research Projects
Organizational Units
Journal Issue
Title
Detectie van milieuveranderingen. Een toepassing van
Structurele Tijdreeksmodellen en het Kalmanfilter
Translated Title
Detection of environmental changes. An application
of Structural Time series models and the Kalman
filter
Published in
Abstract
Onderzoek aan milieuproblemen levert een grote
varieteit aan meetreeksen. Bij analyse van deze gegevens komen onder andere
de volgende vragen naar voren: 1) is er sprake van een trend in de data en
is de stijging of daling statistisch significant, (2) zijn er cyclische
signalen aanwezig en hoe zien die er uit, (3) wat is de invloed van externe
variabelen op de metingen, en (4) hoe kunnen we voorspellingen genereren met
onzekerheden? In dit rapport wordt de beschrijving gegeven van een
generieke statistische techniek waarmee antwoord kan worden gegeven op deze
vragen. De methode is gebaseerd op Structurele Tijdreeksmodellen en het
Kalmanfilter. Het bijbehorende softwarepakket heet TrendSpotter, en is
recentelijk beschikbaar gekomen voor toepassing op PC's. TrendSpotter werd
aanvankelijk ontwikkeld bij KEMA onder de naam KALFIMAC voor de analyse van
milieu-meetreeksen. Het pakket is door RIVM aangekocht in 1996, en
momenteel met toestemming van KEMA verbeterd en aangepast aan de
RIVM-praktijk (implementatie op PC en gebruik van S-PLUS). De methode heeft
een aantal unieke kenmerken met hoge relevantie voor milieu-onderzoek. We
noemen drie van zulke kenmerken:het schatten van trends met 5en of meer
buigpunten in de tijd, waarbij steeds alle onzekerheidsinformatie
beschikbaar is;het schatten van een cyclus waarbij de vorm van deze cylus
in de tijd mag evolueren;het schatten van weegfactoren voor verklarende
variabelen waarbij deze weegfactoren in de tijd mogen veranderen.Een
tweetal toepassingen uit de praktijk van het MilieuNatuurPlanbureau (MNP)
van het RIVM worden gegeven. De toepassingen zijn (A) het detecteren van
klimaatveranderin-gen als gevolg van het broeikaseffect, en (B) het schatten
van meteo-correcties voor luchtverontreinigende
componenten.
Are there significant trends in temperatures and precipitation over the past hundred years? And show these series some cyclic behaviour corresponding to sun spot numbers? Or, can we detect significant downward trends in concentrations of Particulate Matter? And what is the role of meteorological conditions? Are observed trends due to reduced emissions?In this report we describe a generic statistical tool dealing with these type of questions. The technique for analysing environmental time series is based on Structural Time Series Analysis and the Kalman filter. These techniques are well-known in fields as Econometrics and Signal processing and Control, but are relatively unknown in Environmental research. Structural Time-Series models can be seen as a modular 'toolkit': we can estimate trends, cycles and the influence of explanatory variables (also called 'regressors' or 'predictors'). Also combinations of these components can be chosen. Moreover, confidence limits are given for all estimation results.The associated software is called TrendSpotter and has been made available for both UNIX and PC. Early versions of TrendSpotter were developed at KEMA, under the name KALFIMAC. This report gives elaborate simulated examples illustrating the unique features of the methodology. These features are (among others) (i) estimation of flexible trends with elaborate uncertainty estimates, (ii) estimation of cycles where the form of the cycle may evolve over time, and (iii) estimation of time-varying weighing factors for explanatory variables.The modelling approach is applied to two environmental issues: (a) the estimation of trends and cycles in climatological time series, and (b) the influence of meteorological conditions to concentrations of Particulate Matter (PM10). The former issue has great relevance in the light of greenhouse-gas-induced climate change. The latter issue has great policy relevance due to the hypothesized link between policy-driven emission reductions and corresponding trends in concentrations.
Are there significant trends in temperatures and precipitation over the past hundred years? And show these series some cyclic behaviour corresponding to sun spot numbers? Or, can we detect significant downward trends in concentrations of Particulate Matter? And what is the role of meteorological conditions? Are observed trends due to reduced emissions?In this report we describe a generic statistical tool dealing with these type of questions. The technique for analysing environmental time series is based on Structural Time Series Analysis and the Kalman filter. These techniques are well-known in fields as Econometrics and Signal processing and Control, but are relatively unknown in Environmental research. Structural Time-Series models can be seen as a modular 'toolkit': we can estimate trends, cycles and the influence of explanatory variables (also called 'regressors' or 'predictors'). Also combinations of these components can be chosen. Moreover, confidence limits are given for all estimation results.The associated software is called TrendSpotter and has been made available for both UNIX and PC. Early versions of TrendSpotter were developed at KEMA, under the name KALFIMAC. This report gives elaborate simulated examples illustrating the unique features of the methodology. These features are (among others) (i) estimation of flexible trends with elaborate uncertainty estimates, (ii) estimation of cycles where the form of the cycle may evolve over time, and (iii) estimation of time-varying weighing factors for explanatory variables.The modelling approach is applied to two environmental issues: (a) the estimation of trends and cycles in climatological time series, and (b) the influence of meteorological conditions to concentrations of Particulate Matter (PM10). The former issue has great relevance in the light of greenhouse-gas-induced climate change. The latter issue has great policy relevance due to the hypothesized link between policy-driven emission reductions and corresponding trends in concentrations.
Description
Publisher
Sponsors
RIVM