An assessment of the BEST-GLUE procedure to estimate soil hydraulic properties and their uncertainty

Abstract

Understanding the soil's hydraulic characteristics is crucial in effectively modelling the hydrological response of the catchment to the rainfall and contamination transition. The soil hydraulic properties not only are critical in designing stormwater devices in the urban areas for stormwater management but also play an important role in understanding the wider catchment response to extreme weather conditions such as severe storms or prolonged droughts. A full understanding of the latter is becoming more crucial under the light of climate change, where flush flooding or extended droughts are becoming more frequent phenomena worldwide. In this way, developing a practical method to make it possible to carry out soil infiltration experiments on a wide scale is paramount. The soil infiltration tests can be categorised into two large groups, including the laboratory and in-situ methods. The in-situ method is preferred over the laboratory method since it can better assess catchment conditions. The Beerkan method, along with its estimation algorithm, Beerkan Estimation of Soil Transfer Parameters (BEST), is an appealing in-situ approach. It can characterise all soil hydraulic parameters, including water retention and soil hydraulic conductivity curves, using a simple ring infiltration test with null head and a soil sampling to determine the soil moisture and particle size distribution. Although over the last decade this method has attracted much attention, the robustness of the BEST original algorithm, BEST-slope, and its two variations, BEST-Intercept and BEST-steady, are questionable since they may either fail the estimation of soil parameters or produce some inaccurate estimates under certain conditions, including a poor description of transient or steady-state phases, and pronounced concavity or convexity of the cumulative infiltration curve. BEST-generalised likelihood uncertainty estimation (GLUE) is an alternative algorithm that employs the Bayesian Monte Carlo parameter estimation technique as a robust statistical approach, exploring the entire spectrum of each Beerkan parameter to quantify the uncertainty of the model predictions. The algorithm combines three different likelihood measurements based on observed transient flow, steady-state flow and experimental steady-state infiltration rate to provide the best fit between modelled and observed data. Although this method provides promising results by acquiring the data from published literature and demonstrating its capability to estimate the model parameters where the original method failed, it has not been thoroughly evaluated in a real case study. Thus, this study aims to fully examine this algorithm in challenging environments, including volcanic and heterogeneous native forest soils in Auckland, New Zealand, since the Beerkan original algorithm may struggle to estimate soil characteristics in some of these problematic soils. The Beerkan experiments are carried out in 16 different sites all across the Auckland region, including three hilly forest sites with the endangered native trees of Kauri, which struggle not only with the kauri dieback disease but also with the prolonged drought in Northland of New Zealand, one native forest site within the city, ten experiments in volcanic cones all across the Auckland, one experiment in the sandy soil near the beach and one test in the residential area with typical low permeability soil. For each site, at least two Beerkan tests are carried out. One test was on the soil's surface, and the other tests were in different depths up to 27cm. The test was carried out in different depths to achieve a correct underrating of the soil infiltration capacity since the heterogeneity of soil type and structure were prevalent in many sites. Guelph Permeameter (GP) tests were also carried out in each site alongside the Beerkan to validate the BEST-GLUE and BEST results by estimating the saturated hydraulic conductivity. The results of the experiments showed that the BEST-slope failed in almost 54% of cases. The BEST-slope mainly failed to produce results for highly porous organic soils or heterogeneous soils with a high concavity in the transient phase. However, for less problematic soils, where the infiltration curve was close to ideal shape (i.e. a gentle concavity at the start and then a linear shape toward the end of infiltration procedure), the BEST-slope algorithm produced reliable results, where its outcomes were very close to the BEST-GLUE and Guelph Permeameter. In contrast, the BEST-intercept and BEST-steady generally overestimated the soil hydraulic parameters, notably for problematic soil where the BEST-slope failed. However, in a few cases where the BEST-slope managed to produce results, the BEST-intercept and BEST-steady also produced reliable results close to the BEST-GLUE and Guelph Permeameter. The BEST-GLUE managed to produce results in all cases. In general, the success of the BEST-GLUE algorithm can be mainly attributed to the use of a multi-objective optimisation approach in which the algorithm combines the likelihoods associated with transient flow, steady-state flow and steady-state infiltration rate using Bayesian theory. The algorithm is also flexible in terms of modifying the model criteria, including the cutting threshold of behavioural solution and scaling factor, while encountering a difficult situation in problematic soils. Moreover, it allows the model parameters to vary in their feasible ranges to reach the best fittings in the modelling of all parts of infiltration data (i.e. transient flow and steady-state flow and steady-state infiltration rate). The results of this study showed that the BEST-GLUE is a robust alternative algorithm that is capable of utilising the Beerkan data at its highest potential. It also showed that the BEST alternative algorithms (i.e. BEST-intercept and BEST-steady) may produce less accurate results, particularly when the BEST-slope failed. Comparing the saturated hydraulic conductivity (Ks) estimated by BEST-GLUE and BEST with the Guelph Permeameter results showed that both algorithms can reliably estimate the soils hydraulic properties, where the results of BEST-GLUE and BEST algorithms were almost in one order of magnitude to GP results. This study showed that the Beerkan method, and its algorithms, particularly BEST-GLUE, is a reliable, cost-effective approach that can be extensively used in urbanised and rural areas to comprehensively understand the soil hydraulic characteristics, even under challenging field conditions.