Temporal Modeling of Geographically Weighted Regression for Extraction of Relationships between Land Use/Land Cover and Water Hardness

The effects of human activities on the surface water quality have been exhaustively investigated. The results have shown that the impact of land use/land cover on the surface water quality varies from a location to another. However, the temporal characteristics of the problem are poorly understood. This paper examines the hypothesis states that the impact of land use /land cover on the surface water quality varies with time i.e., it is a spatio-temporal linkage. Spatio-temporal data mining analyses were deployed, as they are capable of extracting novel patterns and correlations hidden in the data. Water quality parameter including As (Arsenic) was considered for 12 water quality monitoring stations across Seattle, Washington. These parameters were examined against five land use/land cover types (urban, cultivation, hay, forest, and wetland) for 9 years from 1998 to 2006 to study how land use/land cover influence the parameters. Due to non-stationary intrinsic of the problem as well as spatial auto-correlation exist between the values observed at the monitoring stations; the ordinary least square regression produces unwanted bios in the results. Hence, the geographical weighted regression (GWR) was used in order to model the spatially varying characteristics of the problem. Furthermore, to incorporate the time-varying effects, the model was calibrated for values collected at the stations separately for wet and dry seasons.
The linkage between LULC and the amount of Arsenic (as the case water quality parameter) at each station was extracted using the temporal geographically weighted regression (Equation 6 at paper) separately for the years 1998 to 2006. Figures 6 and 7 (at paper), respectively, show the results for the wet and dry seasons of the year 2006, classified based on the residual square ( ), which is a measure of goodness of fitness. Larger values for indicate more linkage between the LULC class and the water quality parameter. (Note: because of the limitation in space, the results for other years are not shown).
We also applied a significance test on the extracted linkage at the confidence level of 95%. The results illustrate that except for cultivate lands, other classes could reasonably show the spatial variety of the linkage. On the other hand, comparing the results of the wet and dry seasons shows that the model could extract the linkage in wet seasons more efficiently than dry seasons. For example, while there is a negative relation between the urban land use and the amount of Arsenic for wet seasons at most of the stations, there is no significant relation between them at dry seasons. The reason could be the effect of seasonal changes on the water quality parameters due to seasonal rainfalls. On the other hand, there is a positive linkage between forests and the amount of Arsenic for both of wet and dry seasons. However, this linkage completely follows different patterns for wet and dry seasons, which could possibly happen because of the land cover change in the wet and dry seasons. These two examples certify that seasonal changes have considerable effects of the linkage between LULC and water quality parameters, and so must be treated separately.