Assessment of spatio-temporal variability of streambed hydraulic conductivity: A case study in the Pavanje river, India

Abstract

The hydro-geological properties of streambed together with the hydraulic gradients determine the fluxes of water, energy and solutes between the stream and underlying aquifer system. Uncertainty in stream-aquifer interactions arises from the inherent complex-nested flow paths and spatio-temporal variability of streambed hydraulic properties. The estimation and modeling of streambed hydraulic conductivity (Ks) is an emerging interest due to its connection to water quality, aquatic habitat, and groundwater recharge. Fragmenting streams with dams, diversions, and less frequently road culverts disrupt the longitudinal connectivity and capacity of a stream. Dam induced sedimentation affects hyporheic processes and alters substrate pore space geometries in the course of progressive stabilization of the sediment layers. The present study reports the spatial and temporal variability of streambed hydraulic conductivity along the stream reach obstructed by two Vented Dams in sequence. A detailed field investigation of streambed hydraulic conductivity using Guelph Permeameter was carried out in an intermittent stream reach of the Pavanje river basin located in the mountainous, forested tract of Western Ghats of India. Arriving at realistic statistical and spatial inference based on in-situ data collected is challenging, considering the possible sediment sources, processes, and complexity. Statistical tests such as Levene’s and Welch’s ttests were employed to check for various variability measures. The strength of spatial dependence and the presence of spatial autocorrelation among the streambed Ks samples were tested by using Moran’s I statistic. The measures of central tendency and dispersion pointed out reasonable spatial variability in streambed Ks distribution throughout the study reach during two consecutive years 2016 and 2017. The streambed was heterogeneous with regard to hydraulic conductivity distribution with high-Ks zones near the backwater areas of the vented dam and low-Ks zones particularly at the tail water section of vented dams. Dam operational strategies were responsible for seasonal fluctuations in sedimentation and modifications to streambed substrate characteristics (such as porosity, grain size, packing etc.), resulting in heterogeneous streambed Ks profiles. The channel downstream of vented dams contained significantly more cohesive deposits of fine sediment due to the overflow of surplus suspendedii sediment-laden water at low velocity and pressure head. The statistical test results accept the hypothesis of significant spatial variability of streambed Ks but refuse to accept the temporal variations. Advanced geo-statistical techniques offer a wide range of univariate or multi-variate interpolation procedures such as kriging and variogram analysis that could be applied to these complex systems. The deterministic and geostatistical approaches of spatial interpolation provided virtuous surface maps of streambed Ks distribution. The Moran’s I index approved the presence of spatial dependence in the heterogeneous streambed Ks samples. Interpolation maps of Inverse Distance Weighting (IDW) and Radial Basis Functions (RBF) were more accurate than the krigged surface maps; however, the prediction uncertainty was lower around the sampled values in ordinary kriging estimates compared to deterministic methods. In-situ measurement of streambed hydraulic conductivity all along the length of the stream may not be an ideal and cost-effective way. Hence, the soft computing approaches could be applied to induce a rule based relationship for estimating the values of streambed hydraulic conductivity at unmeasured locations using representative georeferenced neighborhood data. The artificial intelligence (AI) based spatial modeling schemes were tested to predict the spatial patterns of streambed hydraulic conductivity. The geographical coordinates (i.e., latitude and longitude) of the sampled locations from where the in-situ hydraulic conductivity measurements were made were used as model inputs to predict streambed Ks over spatial scale using artificial neural network (ANN), adaptive neuro fuzzy inference system (ANFIS) and support vector machine (SVM) paradigms. The statistical measures computed by using the actual versus predicted streambed Ks values of individual models were comparatively evaluated. The AI based spatial models provided superior spatial Ks prediction efficiencies with respect to both the strategies/schemes considered. The SVM model was found to predict reasonably accurate streambed Ks patterns.