Design and performance of urban sponges in red soil: improvement of physical and chemical properties

This study investigated the physical and chemical properties of a single or combination of permeable materials which can be used as fillers in the Sponge City program in China. Four types of fillers, perlite, coral sand, vermiculite and ceramsite, were selected from six alternative fillers by an analytic hierarchy process. The optimal city sponge, which consists of vermiculite (10 cm), ceramsite (15 cm), perlite (15 cm), coral sand (20 cm) and Canna indica L, was found by the orthogonal experiment (L16(4)). The results of the simulated rainwater experiment of the optimal sponge showed that the permeability coefficient K10, NH3-N, total phosphorus (TP) and chemical oxygen demand (COD) removal rate were 1.20± 0.23 mm/s, 96.6± 0.2%, 36.8± 0.07% and 9.6± 0.07% respectively. The results suggested that the optimal sponge had an excellent treatment effect on NH3-N in rainwater while ensuring rapid infiltration. It provided a simple, economical and effective method for rainwater treatment and the Sponge City program in the future. This is an Open Access article distributed under the terms of the Creative Commons Attribution Licence (CC BY 4.0), which permits copying, adaptation and redistribution, provided the original work is properly cited (http://creativecommons.org/licenses/by/4.0/). doi: 10.2166/wcc.2020.020 ://iwaponline.com/jwcc/article-pdf/12/2/371/865790/jwc0120371.pdf Yunpeng Jing Jian Li (corresponding author) Yimin Mei Xiao Liu Xuelan Yu Xuezhou Hu Fangfang Song Key Laboratory of Jiangxi Province for Persistant Pollutants Control and Resources Recycle, School of Environmental and Chemical Engineering, Nanchang Hangkong University, Nanchang, 330063, China E-mail: lijian@nchu.edu.cn Mingming Lu Department of Chemical and Environmental Engineering, University of Cincinnati, Cincinnati, Ohio, USA


INTRODUCTION
In recent decades in China, about ten million people have entered cities (Qiu ) every year. As a result of urbanization, a large number of buildings were built, resulting in decreased green areas and an increase of impervious zones, which increased the risks of urban flooding. In China, the economic loss and casualties in cities affected by flooding have increased significantly (Liao et al. ). In 2011, the worst urban waterlogging occurred in China, which caused nearly 400 billion yuan of economic losses (Wu et al. ).
In 2016, 28 provinces in China experienced severe flooding during weeks of torrential rainfall (Jiang et al. ). Urban rain water management has become a serious challenge.
Meanwhile, many strategies promoting integrated management of urban water have been practiced among industrialized nations. Such efforts include the Low-Impact Development (LID) in the United States (Pyke et al. ), the British Sustainable Urban Drainage System (SUDS) (Mitchell ) and the Australia Water-Sensitive Urban Design (WSUD) (Morison & Brown ), as well as the LID in New Zealand, and the Low-Impact Urban Design and Active, Beautiful, Clean-water Program (ABC) in Singapore, to name just a few (Voyde et al. ). After studying these practices, the concept of 'sponge city' was formally introduced by the Chinese government in December 2013 (Wang et al. ). Since then, the concept of sponge cities has been actively promoted by the Ministry of Housing and

Urban-Rural Construction and the Ministry of Water
Resources have actively promoted sponge cities. The purposes of 'the Sponge City' are as follows: (1) Improving effective control of urban peak runoff, and improving the abilities of temporarily storing, recycling and purifying storm water; (2) There are a few types of LID, such as biological drainage systems, rain water gardens, permeable sidewalks and green roofs etc. (Everett et al. ). Among these LID designs, the rain water garden was considered as the main solution to urban flooding problems in China (Wang et al. ). A rain water garden is a biological detention facility, which can effectively achieve natural purification of rain water. Compared with a large-scale municipal transformation, utilizing rain water gardens to improve waterlogging and runoff pollution is more practical. Rain water gardens can effectively reduce peak runoff and relieve the pressure of municipal drainage, and can also purify water quality, improve regional climate and the livability of the city. Meanwhile, some studies on rain water gardens focus on the filler performance. Li et al. However, a rain garden on red soil has been rarely studied.
The red soil areas in China usually have the largest annual rainfall, and red soil has poor permeability, agglomeration difficulty, low base saturation and low nutrients levels (He et al. ). Furthermore, red soil is the most widely distributed soil type in China, with a total area of 56.9 million hm 2 , mostly in the vast low hills of the central Asian tropics between 25 and 31 north latitude. So it is important to study the performance of a rain garden constructed on red soil.
In this study we designed a series of experiments based on rainfall and soil conditions in Nanchang (Jiangxi Province), with an emphasis on red soil and local plants in rain garden designs. The study aims to obtain optimum rain garden design parameters, including a combination of filler ratio and plant type. Physical and chemical parameters including the permeability coefficient K 10 of the sponge and the pollutant removal rate were measured. These parameters were analyzed by using an analytic hierarchy process and comprehensive scoring method of orthogonal tests. Finally, an optimum sponge structure suitable for red soil was obtained, with pollutant removal rate quantified, which can provide guidance for the construction of a sponge city in red soil regions in China.

The study area
The city of Nanchang is the capital city of Jiangxi Province. It is located in the north of central Jiangxi, on the southwest bank of Poyang Lake between 115 89 0 and 116 35 0 east longitude and between 28 10 0 and 29 11 0 north latitude ( Figure 1).
The whole area is dominated by plains, accounting for 35.8%. The southeast is relatively flat, the northwest hills are undulating, the water network is dense, with many lakes. The water area is 2,204.37 km 2 , accounting for 29.78%. The annual rainfall is 1,600-1,700 mm with 147-157 precipitation days, an annual average of 5.6 rainstorm days and an annual average relative humidity of 78.5%. The city is dominated by red soil. Nanchang is a typical southern city that often experiences urban disasters, especially flood.

The experimental device and measurement method
The main experimental device is shown in Figure 2. This device consists of an aquifer (50 mm), cover layer (50 mm), planting layer (150 mm), filler layer (650 mm) and drainage layer (100 mm), and it is separated by geotextiles between each layer. Simulation rain water flows out of the tank, through each layer, and finally flows out of the drain layer.
Then the water sample is collected and determined by standard methods. In this study, permeability coefficient K 10 , chemical oxygen demand (COD), NH 3 -N concentration and total phosphorus (TP) concentration were measured as representative physical and chemical indicators respectively.
In order to improve the accuracy of the data, these indexes were analyzed using standard methods. GB7788-87 Determination of Forest Soil Permeability was used to measure the K 10 for every sample, and Nessler Reagent Spectrophotometry GB7479-87 was used to obtain the NH 3 -N concentration. TP concentration and COD were measured by Ammonium Molybdate Spectrophotometry GB11893-89 and Rapid Oven Method respectively.

Preparation of simulated rain water
The water quality of road runoff rain water in Qianhu campus of Nanchang University were reported as follows: 4.13-557.38 mg/L of COD, 0.01-2.41 mg/L of TP, and 0.04-5.29 mg/L of NH 3 -N (Jiang ). Since the selected area had less pollution than the area in the literature, lower concentrations were used for simulated rainwater with 160 mg/L of COD, 1.20 mg/L of TP and 1.40 mg/L of NH 3 -N by dissolving 1.2754 g potassium hydrogen phthalate, 0.5235 g ammonium chloride and 0.5308 g potassium dihydrogen phosphate were dissolved in 1 L water respectively.
In order to simulate true rainfall, we used Nanchang rainstorm intensity formula: where the value of P (recurrence interval) is 5. If A is defined as 9.57(1 þ 0:69logP), then the value of A is 14.19 mm/h. Rainwater runoff was estimated using the following (Li et al. ): where A represents the rain force value when P is five years in Nanchang, a represents the instrument floor area of 600 cm 2 , and k represents the confluence area ratio of 20 (Zhang & Xu ). Then, Q simulation of 4.73 mL/s is obtained and used in subsequent tests.

Preparation of filler materials and plants
Since filter materials are usually used in large quantities in the green space around roads, there should be sufficient supplies which are low cost and environmentally friendly.
The six filler materials selected include perlite, coral sand, maifan stone, vermiculite, ceramsite and bio-charcoal (Ying & Zhao ). They were all purchased from a garden product market in Nanchang, and the physical properties of these fillers are listed in Table 1.

In addition, Coreopsis drummondii Torr et Gray, Conyza
Canadensis(L.) Cronq, Canna indica L and Arachis duranensis were selected as tested plants. They were collected from Nanchang Hangkong University. Especially, in order to improve the reliability of this experiment, the plants were planted in prepared soil, which consisted of 65% sand, 30% red soil and 5% nutritious soil, a month in advance.

Analytic hierarchy process
Multiple factors need to be considered during the screening of single fillers, however, the quality and importance of each factor is different. Such a multi-objective decision-making case can be solved by the analytic hierarchy process (Gao ). The analytic hierarchy process is a combination of qualitative and quantitative, systematic and hierarchical analysis methods with the following six steps: (1) Divide the factors involved into several levels and establish a multi-level hierarchical model.
(2) According to the evaluation scale, the initial relative importance of each element of the same level is determined, and quantified, then the evaluation matrix A is established.
(3) Calculate the eigenvectors of the evaluation matrix and the corresponding maximum eigenvalues to determine the relative importance of each layer of elements.
(4) The consistency check of the matrix is performed, including the calculation of the consistency index (CI), the average random consistency index (RI), and the consistency ratio (CR). If CR <0.1, then the consistency of the evaluation matrix is considered acceptable.
(5) If the above steps are passed and the consistency test is passed, the obtained feature vector is the weight of each factor.
(6) Combine the weights of various factors, and finally determine the total weight of the layer elements of the scheme to the target layer, and select the optimal filler according to the weight.
In the single filler screening experiment in the first step, the analytic hierarchy process was used to analyze the results, and the four fillers used in the combined experiment were screened, and then the next experiment was carried out.
Comprehensive scoring method for multi-index orthogonal experiment The experiment of combined fillers was carried out by the Taguchi's orthogonal array experiment with a five-factor four-level orthogonal experimental design, and the experimental results include four indexes as follows: permeability coefficient K 10 , COD removal rate, TP removal rate and The key to this method is scoring. The scoring should not only take into account the importance of each indicator, but also consider the pros and cons of each indicator.
Generally speaking, the comprehensive scoring method is similar to the standardization process in the analytic hierarchy process, that is, the best value of each indicator is large deviation values in each set of data, the average value of the permeability coefficient K 10 of each filler was obtained and compared, as shown in Figure 4. It can be seen from Figure 4 that the permeability coefficient from large to small is as follows: perlite > coral sand > maifan stone > vermiculite > ceramsite > biocharcoal. The high permeability coefficient of perlite is mainly due its relatively uniform particle size (between 2 and 3 mm), porous surface lighter than water density. Coral sand has a large porosity and a porous surface, which may correlate to its penetration ability. Maifan stone is hard and does not swell after immersion, and its small porosity is the reason for the low permeability coefficient. Comparing vermiculite to ceramsite, vermiculite is smaller, its bulk density is larger but its permeability coefficient is greater. Mainly because the ceramsite texture is hard and light, a very strong extrusion phenomenon between the ceramsite was observed when it floats on the water surface after full immersion. It causes the porosity to decrease more sharply than in the partially immersed state, and results in a decrease in the permeability coefficient.
Bio-charcoal has the lowest permeability coefficient, in spite of a strong adsorption capacity and great filtration effect.

The analysis of pollutant removal effect
In order to analyze the removal effect of COD, NH 3 -N and TP by each filler, water samples were collected after 60 minutes and various chemical indicators were measured. The experimental data of each indicator is shown in Table 2.
All six fillers have certain purification abilities for simulating road rainwater. The removal rate of NH 3 -N by various fillers is generally higher than that of TP and COD. This is consistent with previous reports, mainly attributed to the good adsorption properties of these materials for NH 3 -N

Screening and evaluation of fillers
In this experiment, permeability coefficient, price and pollutant removal rate were selected for evaluation and analysis.
Since the units of each indicator were different, it was difficult to compare directly, so it was necessary to standardize before comparison. The actual data of each indicator and the standardized results are shown in Tables 3 and 4.
The target layer is the ideal filler (A). The factors of the criterion layer are the permeability coefficient K 10 (B 1 ), the price (B 2 ), and the pollutant removal efficiency (B 3 ). The three factors of the scheme layer are COD removal rate (C 1 ), NH 3 -N removal rate (C 2 ) and TP removal rate (C 3 ).
First, the hierarchical model was built, as shown in Figure 5.
Then the relative importance of permeability coefficient K 10 , price and pollutant removal efficiency were determined according to the 1-9 ratio scale method, as shown in Table 5.
So, a pairwise comparison matrix of three indexes (permeability coefficient K 10 , price and pollutant removal efficiency) was obtained according to Table 5, as shown in Table 6.   Table 6 can be converted into an evaluation matrix: A ¼     After λ max is calculated, the consistency check of evaluation matrix A can be performed.
First, the consistency indicator (CI) is calculated: Then, since the order of A was 3, the average random consistency index (RI) is obtained according to   As before, after the λ max is calculated, a consistency check of evaluation matrix B can be performed: Since CR is <0.1, the consistency of the evaluation matrix passed the test, so the eigenvector W 1 is weights of COD, NH 3 -N and TP.
Next, the total weight of the plan layer factor is calculated, as shown in Table 9.
Multiply the scores of the permeability coefficients K 10 , price, COD removal rate, NH 3 -N removal rate and TP removal rate of the six fillers by their respective total weights and obtain the total scores of six fillers, see Table 10.
The total score of the materials is ranked from high to low as perlite > ceramsite > vermiculite > coral sand > maifan stone > bio-charcoal. The top four fillers are used as fillers for combined experiments to find the combination of high permeability and pollutant removal capacity.

Screening experiment and data processing analysis of combined filler experiments
In order to obtain objective and accurate experimental results, we designed a five-factor four-level orthogonal experiment, L16(4 5 ), with factors and levels of the experiment shown in Table 11, and the experimental arrangement is shown in Table 12. The orthogonal experiment is an effective way to determine the optimum conditions among many factors (Shao et al. ).
Composite filler design and its permeability coefficient and removal effect When the permeability coefficient K 10 was measured, plant factors were not considered. The experimental procedure  was consistent with the determination method of the permeability coefficient K 10 of single filler material. Photographs of the experiment are shown in Figure 6.
The filler was cleaned and dried before each experiment, then all experiments were carried out according to Table 12, an orthogonal test table, which is based on the orthogonality to select some representative points from the comprehensive test. Then, according to Equations (1) and (2), rainwater runoff simulation flow q is 4.73 mL/s. All experiments were carried out in the device shown in Figure 2. Finally, in each experiment, a 1 L water sample was collected and measured after 60 minutes.

Analysis of results
The permeability coefficient K 10 of the combined filler is shown in Figure 7. In general, the composite filler had a lower permeability coefficient K 10 than the single filler, and this was due to different fillers stacked together which led to an increase in bulk density and a decrease in porosity.
In and 65 cm (thickest), respectively, and they all had poor permeability coefficients K 10 . When the thickness of the sponge was 45-55 cm, the permeability coefficient K 10 of the sponge was high.
In order to obtain the optimal sponge structure, the permeability coefficient K 10 and contaminant removal capacity of the sponge must be considered at the same time. In the study, a comprehensive scoring method for the multi-index orthogonal experiment was adopted.
First, according to the results of orthogonal experiments, as is shown in Table 13, we know the optimal and worst value for each indicator, and the optimal value is marked as 10 points, with the worst value marked as 1 point; other values are calculated according to their proportion of the difference between the optimal value and the worst value. Second, when calculating the composite score, stipulate the weight of the permeability coefficient K 10 as 2, and other indicators are 1. So, a comprehensive score is the sum of 2× permeability coefficient K 10 þ COD removal      Table 16.
For the optimal sponge structure, as much as 96.6% of NH 3 -N can be removed, and the permeability coefficient K 10 is up to 1.2. The results suggest that this analytical method is very reliable in selecting the factors.

CONCLUSIONS
In this paper, a high NH 3 -N removal urban sponge was designed for application in red soils, which consisted of 15 cm ceramsite þ10 cm vermiculite þ15 cm perlite þ20 cm coral sand with Canna indica L. Performance indicators of the optimal sponge are as follows: the permeation coefficient, K 10 , of 1.20 mm/s, with an NH 3 -N removal rate of 96.6%, a TP removal rate of 36.8%, and a chemical oxygen demand removal rate of 9.6%. The results showed that the urban sponge designed in this experiment not only has excellent permeability, but it can also effectively remove contaminants in runoff rain water. The whole design and selection process can be very helpful for the construction of an urban sponge city in red soil areas in southern China and other red soil areas in the world.  1.20 ± 0.23 9.6% ± 0.07 96.6% ± 0.2 36.8% ± 0.07