Kartering av markanvändning med meteorologisk satellitdata för förbättring av en atmosfärisk spridningsmodell

Sammanfattning: Sulphur emissions during the last century has led to major acidification of the environment. The acidity of soil and water is a threat to health and economic interests of man, e.g. the food and paper/wood production in the agricultural and silvicultural industries. The Black Triangle region suffers from severe damage caused by acidification and actions to decrease the acid rain over fields and forests must continue. A useful tool when making forecasts, discussing sensitive areas and sources of pollution is an atmospheric dispersal model. One model has been developed in department of Meteorology and Environmental Protection at Karl’s University, Prague. The model did not use roughness parameters for different land use. Deposition of airborne pollutants occur at different rates over different land use. The aim was therefore set to create land use data and to enhance the model precision by use roughness parameters for each land use. Since the model covers a large part of Europe, a land use data base was established by analyzing data from the NOAA AVHRR sensor. Eros data center, USA, supplied the 10-day channel and vegetation index composites. To reduce the amount of data a standardized principal component analysis (sPCA) was performed. The standardization of the PCA makes the channels equal and minimizes signal to noise ratio. The PCA also presents variations in the data, e.g. seasons. During the sPCA the number of occasions that could be used were reduced. Cloud contaminated 30 of the 36 occasions that were available. The 10-day composites are put together of the pixels that have the highest NDVI-value (Normalized Differentiated Vegetation Index). This does not guarantee that the pixels are free from clouds but they have the lowest amount of cloudiness during the 10-day period. A maximum likelihood classification of the chosen data resulted in a land use data base with three major classes; agricultural land, coniferous forest and deciduous forest. Evaluations showed that the total certainty were up to 90% in some classifications. The classes Cities and Water were added from a digital map. Training areas for regions as large as this cannot be handled by field studies. Instead, data registered with Landsat TM as well as maps over the region were used to locate training areas. The dispersal model was run with and without the land use data and the results were compared. Results showed a decrease in deposition specially in the mountainous areas. One of the explanations is the high general roughness the model uses when run without land use data. The model will be calibrated with deposition data from gauges in the Czech republic.