Fitting of Segmented Gaussian Plume Model Predictions on Measured Data

An improvement of mathematical model predictions of environmental pollution can be achieved on basis of assimilation of model simulations with real observations incoming from terrain. In this article we pay attention to development and investigation of applicability of one simple empirical method of objective analysis based on least square approach. Output background fields of resulting potentially dangerous endpoints are modified by measurements in such a way, that resulting respond surface is fitted towards measurements through the iterative adjustment of a certain selected set of model input parameters. In spite of a certain limitations this approach has occurred to be applicable for the first preprocessing of the model predictions and simulated measurements. It can support robustness of decision making and can contribute to early detection of possible fatal decision maker errors due to misinterpretation of input parameters of an accidental release scenario. ASSESSMENT OF ACCIDENT CONSEQUENCES Potential failures occurred in man-made processes can cause dangerous phenomena resulted in accidental releases of harmful substances into the living environment. Hazard evaluation and decision-making focused on early warning and protection of population has the highest priority. Reliable and up to date information represents basic inevitable conditions for effective management of intervention operations targeted on consequence mitigation during emergency situations. This appeared to be a basic lesson for further progress of emergency preparedness procedures, which has arisen from Chernobyl accident where lack of reliable information has shown to be the main reason of poor effectiveness of countermeasures. Decision making has to be supported by proper user-friendly simulation software tool complied with advanced theoretical methodology with access to all necessary relevant latest data. Crisis management should come out from reliable simulation of space and time of accident evolution, which should take into account all available information including physical knowledge of problem, expert judgement of input data, online measurements from terrain and others. The subject of investigation concerns evaluation of consequences of radioactivity propagation after an accidental release from nuclear facility. Transport of radioactivity is simulated by mathematical models from initial atmospheric propagation, deposition of radionuclides on the ground and spreading through food chains towards human body. In the final step a hazard estimation based on doses of irradiation is integrated into the software system HARP. Our access is mentioned in (Pecha et al. 2007). FROM DETERMINISTIC TO PROBABILISTIC APPROACH AND DATA ASSIMILATON Recent trends in risk assessment methodology insist in transition from deterministic procedures to probabilistic approach which enables generate more informative probabilistic answers on assessment questions. Corresponding analysis should involve uncertainties due to stochastic character of input data, insufficient description of real physical processes by parametrisation, incomplete knowledge of submodel parameters, uncertain release scenario, simplifications in computational procedure etc. Simulation of uncertainty propagation through the model brings data not only for the probabilistic assessment mentioned above (Pecha et al. 2005) but also for another main task of analysis called assimilation of model predictions with real measurements incoming from terrain. Data assimilation represents the way from model to reality and can substantially improve the model predictions. There are several important sources of information that can enter the assimilation procedures. Basic physical knowledge is included in prior fields (resulted vectors) predicted by simulation model. Assumptions related to the random characteristics of model inputs are supported by some kind of expert judgements (Goosens 2001). Substantial benefit can result from accessibility of data incoming from terrain. Merging of all these contending resources is a principle of assimilation and had shown to be very promising in many branches of contemporary Earth sciences (e.g. Drécourt 2004). Each such resource can be known on a certain degree of details (e.g. dense or rare measurements in space and time, complete or only partial knowledge of model error covariance structure, cases with indirect observations etc). Available information determines the option of suitable assimilation technique. We are considering the assimilation techniques in broader sense (Hofman 2007) from simple interpolation (poor model predictions, but dense and precise observed data) up to advanced statistical methods when full description of error covariance structure is needed e.g. in (Kalnay 2003). DATA ASSIMILATION (DA) USING MINIMISATION TECHNIQUE (MT)