Integration and interpolation free air anomalies with basis in an ann and kriging
Abstract
In this paper the integration and interpolation of Free Air anomalies coming from land and ocean in a 3º x 3º area around the Brazilian Vertical Datum (BVD) are analysed. For this purpose new gravity surveys were carried out as well as data compilation and the use of existing models. They were collected information
(latitude, longitude, ortometric height, gravity and free air anomalies) for 916 stations in the BVD region, and used the SAM_30s model from SRTM with 30” resolution, and the KMS2002 global free air ocean anomalies. With the collected information corrections were computed and the integration resulted in a new set of free air anomalies. From this new data set was formed a regular data grid by interpolating free air anomalies with basis in an Artificial Neural Network (ANN). The more adequate architecture was determined based on 18 experiments. The initial evaluation of the grid generated with ANN was realized by comparing it with a grid determined by Kriging. They were also used an independent data set for evaluating both grids. Numerical tests and statistical analysis of both grid were carried out in order to evaluate the grid quality. A 5,59 mGal RMS error was found for the better ANN generated grid.
(latitude, longitude, ortometric height, gravity and free air anomalies) for 916 stations in the BVD region, and used the SAM_30s model from SRTM with 30” resolution, and the KMS2002 global free air ocean anomalies. With the collected information corrections were computed and the integration resulted in a new set of free air anomalies. From this new data set was formed a regular data grid by interpolating free air anomalies with basis in an Artificial Neural Network (ANN). The more adequate architecture was determined based on 18 experiments. The initial evaluation of the grid generated with ANN was realized by comparing it with a grid determined by Kriging. They were also used an independent data set for evaluating both grids. Numerical tests and statistical analysis of both grid were carried out in order to evaluate the grid quality. A 5,59 mGal RMS error was found for the better ANN generated grid.
Keywords
Anomalias ar livre; Redes neurais artificiais
