A framework for evaluating air quality models

This paper describes a framework to evaluate air quality model predictions against observations. We propose the following relationship between observations and predictions from an adequate model% MathType!MTEF!2!1!+-% feaafeart1ev1aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn% hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr% 4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq-Jc9% vqaqpepm0xbba9pwe9Q8fs0-yqaqpepae9pg0FirpepeKkFr0xfr-x% fr-xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGabm4qayaaja% WaaSbaaSqaaiaaicdaaeqaamXvP5wqonvsaeHbfv3ySLgzaGqbaOGa% e8hkaGIaamiEamaaBaaaleaacaaIXaaabeaakiaacYcacaWG4bWaaS% baaSqaaiaaikdaaeqaaOGae8xkaKIaeyypa0Jabm4qayaajaWaaSba% aSqaaiaadchaaeqaaOGae8hkaGIaamiEamaaBaaaleaacaaIXaaabe% aakiab-LcaPiab-TcaRiabew7aLjab-HcaOiaadIhadaWgaaWcbaGa% aGOmaaqabaGccqWFPaqkaaa!4F93!\\hat C_0 (x_1 ,x_2 ) = \hat C_p (x_1 ) + \varepsilon (x_2 )\],where x ₁ refers to the inputs used in the model prediction C _p(x ₁), and x ₂denotes unknown variables which affect the observed concentration C ₀. The hats associated with C _pand C ₀denote transformations to convert the residual to a white noise sequence which is normally distributed. In this paper we assume % MathType!MTEF!2!1!+-% feaafeart1ev1aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn% hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr% 4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq-Jc9% vqaqpepm0xbba9pwe9Q8fs0-yqaqpepae9pg0FirpepeKkFr0xfr-x% fr-xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGabm4qayaaja% GaeyyyIORaciiBaiaac6gacaWGdbaaaa!3B39!\\hat C \equiv \ln C\].The standard deviation of determines the expected deviation between model prediction and observation. The purpose of model improvement is to make this deviation as small as possible.The formalism we have proposed is applied to the evaluation of two models developed by this author. We show how careful analysis of residuals can lead to improvements in the model. We have also estimated for each of the models.In the last part of the part of the paper we show how the statistics of can be used to interpret model predictions.