Assessment of a non-parametric method for hydrologic data disaggregation in space-time based on K-Nearest Neighbors (Case Study: Three sub-basin in West of Urmia Lake)

Stochastic parametric disaggregation models that maintain spatial and temporal correlation are widely used in hydrology. To avoid high complexity and large number of parameters, which imposes a significant amount of uncertainty to the results, use of non-parametric disaggregation methods has been widely suggested as an alternative by researchers. Among the non-parametric modeling methods, the K-nearest neighbors method proposed by Prairie et al. gains strong mathematical basis and inherent simplicity. In our work, the modified disaggregation approach of the K-nearest neighbors method is used for temporal and spatial disaggregation of rainfall and river flow values and the performance is evaluated. The exploited flow and rainfall data correspond to three stations in three sub-basins located at the west of Lake Urmia are used. The total amount of annual rainfall and flow of the three stations are generated using stochastic lag -1 autoregressive model (AR (1)). Using the non-parametric disaggregation model, the generated annual values are disaggregated into three sub-basins. The annual values for each sub-basin are then disaggregated into different months. Comparing statistics of disaggregated data with those for historical data, shows the good performance of the proposed disagreation model and its ability to disaggregate streamflow and rainfall data.