Application of some statistical methods for evaluation of groundwater observations

• Main Author: Ackerberg, Björn
• Format: Others
• Language: English
• Published: KTH, Mark- och vattenteknik 2002
• Online Access: http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-1531; http://nbn-resolving.de/urn:isbn:91-7283-327-0

With the objective of reviewing different statisticalmethods for evaluation of groundwater data and the design of agroundwater observation network, a comprehensive literaturesurvey was performed. The literature survey focuses on spatialstatistics (geostatistics) but also includes methods toevaluate time-series of groundwater data and the determinationof the sampling frequency. A method is developed which providesa means of quantifying the accuracy of an existing groundwatermonitoring network with regards to spatial interpolation andthe locations of the corresponding observation points. Thespatial interpolation method of ordinary kriging was used. Aresult from ordinary kriging was estimated (interpolated)levels at unmeasured points, but also a kriging variance. Thekriging variance can be interpreted as a measure of theestimation accuracy and used as a criterion for network design.Design of a monitoring network for groundwater levels in anarea includes the selection of: - the number of observationpoints and - the spatial locations of observation points.The method was applied to design a monitoring network inan area in a glaciofluvial deposit, the Nybro esker, which isthe main aquifer for the water supply of the Kalmar-Nybroregion in the southeast of Sweden. This thesis shows that it ispossible to quantify the accuracy of an existing observationnetwork using the average kriging variance as a measure ofaccuracy. It is also possible to describe how this krigingvariance changes (increases) when the observation network isreduced. By using this variance is it possible to rank thedifferent points in the network as to their relativeimportance. It is thus possible to identify the points, whichare to be removed when the observation network is reduced, onepoint at a time. This study shows that a monitoring network inthe study area could be reduced by 35% while the increase inaverage estimation (kriging) variance is only about 10%.Although the method is applied to groundwater levels in aglaciofluvial deposit, it is applicable also to other variablesthat can be considered regionalized and to other geologicalenvironments.