Assessment of land use dynamics of the N’djili catchment in DR Congo: implication for catchment planning

Abstract

The lack of the basic data in both ungauged and poorly gauged catchments make the tasks of water resources management and planning challenging in these areas. This leads to ineffectiveness and improper assessment of water resources in most of the poorly gauged catchments. Thus, this study reports on three approaches that assess temporally and spatially the land use of the N’djili Catchment in relation to water resources management. The N’djili Catchment whose principal tributary is the N’djili river falls in the lower drainage basin of the Congo River. With daily average of 22.3 m3/s, the N’djili river contributes to water supply of closely two-third of the population in the suburbs as well as private connections of Kinshasa with an estimated 330,000 m3 of drinking water per day. Using four variables from meteorological, water quality and hydrological parameters, as well as daily sediment loads and daily discharge respectively generated from turbidity and water level measurements were processed. Variables such as sediments, discharge, rainfall and potential evapotranspiration were tested successively for homogeneity (Pettit test), time series and double cumulative curves analysis after performing the normality test. These trends analyses revealed significantly poor long-term correlation between meteorological and other group of parameters. The land use classification was done through unsupervised classification followed by Sieve and Clump post-classification methods for the following years 1987, 1995, 2001 and 2012. The confusion matrix was used and the average overall accuracy obtained for all the land use classification were 79.03% and Kappa coefficient accuracy ranged from 0.56 to 0.79. The analysis of variance revealed a significant (p-value <0.05) change occurred within land use classes. Hence, SWAT model was then used to simulate sediment loads based on the land use scenarios and the outputs revealed a progressive increase in sediment pattern as the changes occurred in land use. The model performance was evaluated using the following indices: Nash-Sutcliffe Efficiency (NSE), Percent bias (PBIAS), Root square mean error to the standard deviation (RSR), Pearson’s correlation coefficient (r) and coefficient of determination (R2). The dimensionless indices indicated that satisfactory to very good results were achieved from the simulations of the sediment loads. Finally, two regression models were used to predict the extent to which sediment loads and turbidity affect the water treatment cost of the N’djili river. Non-linear and multi-linear regression analyses were performed. The coefficient of determination (R2 =0.82) from non-linear regression indicated that the amount of chemical used increases as the turbidity loads increase reaching a saturation level. On the other hand, the multi-linear regression model was built based on thirteen most sensitive parameter sets from the SWAT sensitivity analysis. The model showed that significant (p-value 0.04) and satisfactory prediction of sediment loads was achieved by considering only the Pearson’s correlation coefficient (r = 0.63) while the coefficient of determination (R2 ) only indicated a value of 0.40. This study reveals that the change in loads sediment identified through the analyses is a result of landuse practices rather than changes in meteorological variables and these results call for appropriate measures in terms of landuse activities and management as these lead to an increase in the costs of treatment of water supply for the city of Kinshasa