Improved spatial outlier detection method within a river network

Abstract

A spatial outlier refers to the observation whose non-spatial attribute values are significantly different from those of its neighbors. Such observations can also be found in water quality data at monitoring stations within a river network. However, existing spatial outlier detection procedures based on distance measures such as the Euclidean distance between monitoring stations do not take into account the river network topology. In general, water quality levels in lower streams will be affected by the flow from the upper streams. Similarly, the water quality at some tributaries may have little influence on the other tributaries. Hence, a method for identifying spatial outliers in a river network, taking into account the effect of river flow connectivity on the determination of the neighbors of the monitoring stations, is proposed. While the robust Mahalalobis distance is used in both methods, the proposed method uses river distance instead of the Euclidean distance. The performance of the proposed method is shown to be superior using a synthetic river dataset through simulation. For illustration, we apply the proposed method on the water quality data from Sg. Klang Basin in 2016 provided by the Department of Environment, Malaysia. The finding provides a better identification of the water quality in some stations that significantly differ from their neighbouring stations. Such information is useful for the authorities in their planning of the environmental monitoring of water quality in the areas.