Review of Subsurface Dam Technology Based on Japan’s Experience in the Ryukyu Arc

Abstract

Based on the success of an irrigation project that utilized two subsurface dams as water sources on Miyako Island, ten additional subsurface dams have now been completed. The technologies that have made the giant subterranean dam possible are the integrated storage model for creating water utilization plans and the Soil Mixed Wall method for constructing cut-off walls. Although it might be tempting to assume that all subsurface dams in the Ryukyu limestone region were built under identical topographical and geological conditions, the reality is quite different. Each dam faced unique geological and construction challenges that engineers skillfully overcame during the building process. The purpose of this paper is to introduce information on the planning and construction technology of agricultural subsurface dams in the Ryukyu Arc, which has not been reported in English so far, and to clarify the characteristics of agricultural subsurface dams in the Ryukyu Arc. There is a strong correlation between the gross reservoir capacity and the active capacity of large-scale subsurface dams. Eleven percent of the construction cost was the cost of design and investigation. The water price is the same as or slightly higher than that of surface dams.

1. Introduction

A subsurface dam is a facility that enhances groundwater capacity by constructing a cut-off wall within an aquifer. It effectively raises the groundwater level and prevents seawater intrusion in coastal areas. Subsurface dams had already been built in Sardinia and North Africa during the Roman period [1]. Its history goes back about 2000 years. However, it was not until the early 1980s that the first subsurface dams using modern civil engineering techniques were constructed [2]. In this paper, current technologies for subsurface dams are reviewed. “Subsurface dam” and “Underground dam” are synonyms. In this paper, the terms are unified to the subsurface dam.

The Okinawa Islands, along with the Miyako and Yaeyama Islands, were returned to Japan from US occupation in May 1972, after marking 27 years since the end of the war. Even before Okinawa’s return to Japan, the Ministry of Agriculture, Forestry, and Fisheries (MAFF) had dispatched a research team to develop agricultural water resources. This team, in collaboration with the Ryukyu government, conducted a preliminary study using electrical exploration and borehole surveys to assess the potential for groundwater development.

Despite the fact that the Miyakojima area, where nearly the entire island is covered with Ryukyu limestone, constitutes a significant agricultural region accounting for approximately 20% of the farmland area in Okinawa Prefecture, it frequently faces droughts due to the lack of a river system. Securing agricultural water on this island was a top priority in Okinawa Prefecture’s agricultural policy.

Sugawara [2,3,4], an expert of the Ministry of Agriculture, Forestry, and Fisheries (MAFF), had once proposed a water source development plan involving subsurface dams. However, his idea was rejected at the time due to its innovative nature. Sugawara studied the hydrogeological map of Miyako Island, which was completed at the end of 1973, and came up with the concept of developing water sources using subsurface dams. He also played a key role in securing a budget for the development of subsurface dam technology. This budget request marked the beginning of the “Investigation for Subsurface Dam Development” in 1974. As part of this investigation, the Minafuku Dam was constructed. Its success has confirmed the feasibility of subsurface dams [5].

Numerical analysis technology is crucial for various aspects of subsurface dam implementation, including water-usage planning, water intake facility design, and drainage facility design. Technology that visualizes long-term invisible groundwater is important, especially when creating budget proposals. Essentially, this technology is necessary to communicate the benefits of subsurface dams to individuals who may not be familiar with groundwater, much like visible surface dams.

Yoshikawa [5,6] developed an integrated storage model as a simulation tool for the Minafuku Dam. In this model, each sub-basin is represented by a groundwater tank model, and multiple such models are connected in series to represent the entire basin. This model [5] is an extension of the original surface water Sugawara Tank model [7] to a groundwater model. It has a remarkable capability of calculating daily groundwater level fluctuations over a 30-year period. As a result, engineers can determine the optimal size of the dam following a procedure similar to that used for surface dams.

A review of the Minafuku Dam embankment assessed that it is quite difficult to construct the embankment to a significant depth (more than 50 m deep) with the applied grouting method [8]. Consequently, the exploration of alternative cut-off wall construction technologies became imperative. Under these circumstances, in July 1984, Seiko Kogyo Co., Ltd., Osaka, Japan (at the time) [9] gave a presentation on the possibility of applying the Soil Mixed Wall (SMW) method to the subsurface dam cut-off wall construction at the Miyako office of the Okinawa General Bureau. The Seiko Kogyo had succeeded the SMW method used commercially for the first time in Japan in 1976 [9]. Therefore, this construction method was not widely known at that time (the early 1980s). Therefore, to confirm the workability of the SMW construction method on Ryukyu limestone, The Miyako office conducted a test construction of a wall box (3 m × 3 m × 60 m) impounding water using the SMW method at the planned construction site of the Sunagawa subsurface dam [8]. As a result, it was confirmed that construction was possible. Details of the construction test were described by Shimoji [8], the director of the Miyako office at the time.

Based on the aforementioned core technologies, an irrigation project utilizing two underground dams as water sources was completed in 1993. Over three decades later, a total of twelve underground dams have been constructed. Each of these subsurface dams was built under distinct geological and construction conditions, with engineers overcoming various challenges during the construction process.

The locations and limited information on agricultural subsurface dams in Japan can be found in Nishigaki et al. (2004) [10], Japan Green Resources Agency (2004) [11], Ishida et al. (2011) [12] and MAR Portal (https://www.un-igrac.org/ggis/mar-portal) (accessed on 31 July 2024) (Stefan and Ansems, 2018) [13]. However, these sources do not include the aforementioned experiential and knowledge-based information. Most of this information is primarily introduced on Japanese websites and magazines, making it difficult for overseas researchers to access. The purpose of this paper is to share some of this information with international researchers and engineers. It will introduce information on the planning and construction technology of subsurface dams, which has not been reported in English so far, as well as the environmental impact of subsurface dams. It should be noted that the information presented here is only a portion of the extensive information available in Japanese.

2. Subsurface Dam Background

2.1. Ryukyu Arc and Subsurface Dam

Japan’s agricultural subsurface dams utilize Ryukyu limestone, which is distributed across the Ryukyu Arc, an island arc extending approximately 1200 km from Kyushu to Taiwan (see Figure 1). To the east of the Ryukyu Arc lies the Ryukyu Trench, which reaches a maximum depth of over 7000 m. This trench serves as an extension of the Nankai Trough. Specifically, the Ryukyu Trench represents a subduction zone where the Philippine Sea Plate subducts beneath the Okinawa Plate [14]. Within the Ryukyu Arc, the Okinawa Trough lies on the northwest side and has a water depth ranging from 1000 to 2000 m. This trough is an integral part of the island arc-trench system, with the Ryukyu Islands and Ryukyu Trench forming key components. Notably, the Okinawa Trough functions as a back-arc basin [15].

The expansion of the Okinawa Trough occurs at a rate of 1–2 cm per year in the northwest-southeast direction within the central and northern parts, while in the southern part, it expands at a faster rate of 3–5 cm per year in the north-south direction. This dynamic geological activity has resulted in the formation of two deep submarine canyons within the Ryukyu Arc. These canyons appear to have been shaped by left-lateral strike-slip faults with varying displacements [15]. Specifically, the northern canyon corresponds to the Tokara Strait. The southern canyon is known as the Kerama Gap. Both canyons reach depths exceeding 1000 m. Consequently, the Ryukyu Arc is divided into three distinct regions: Northern Ryukyu, Central Ryukyu, and Southern Ryukyu [15].

Figure 1. Relationship between geology and subsurface dams from southwestern Japan to the Ryukyu Arc (Geological map is modified from [16]. The background submarine topographic map is from the Geospatial Information Authority of Japan [17]). The black square indicates the area of Figure 2.

Figure 1. Relationship between geology and subsurface dams from southwestern Japan to the Ryukyu Arc (Geological map is modified from [16]. The background submarine topographic map is from the Geospatial Information Authority of Japan [17]). The black square indicates the area of Figure 2.

The geological composition beneath the Ryukyu limestone, especially the bedrock beneath the aquifers of the subsurface dams, varies from place to place. This is because the pre-Neogene bedrock of the Ryukyu Arc has a zonal structure [18]. The pre-Neogene bedrock of the Ryukyu Arc has long been recognized as an extension of the zonal structure of Southwest Japan [18]. The zonal structure of Southwest Japan can be divided into the outer belt (Pacific Ocean side) and the inner belt (Japan Sea side) by the east-west Median Tectonic Line [19]: The inside of the outer belt is divided into the Sanbagawa belt (high-pressure metamorphic belt), the Chichibu Belt (Permian-Triassic accretionary complex), and the Shimanto Belt (Cretaceous-Neogene accretionary complex) by tectonic lines (faults). The tectonic line that forms the boundary between the Chichibu Belt and the Shimanto Belt is called the Butuzou tectonic line. The internal structure of the inner belt can be divided into seven zones, as shown in Figure 1 [16].

The Ryukyu Arc zonal structure of pre-Neogene bedrock is divided into the following five belts: (1) Iheya Belt (Permian to Jurassic accretionary complex), (2) Nakijin Belt (Triassic accretionary com-plex), (3) Motobu Belt (Early Cretaceous accretionary complex), (4) Nago Belt (Late Cretaceous accretionary complex), and (5) Kiyou Belt (Eocene accretionary complex) (Figure 2) [20]. These belts are separated by northwest-dipping reverse faults, and the ages of the belts gradually decrease toward the structurally lower belts [20]. Belts (1)–(3) and (4)–(5) are considered to be derived structural zones from the Chichibu Belt and the Shimanto Belt in southwestern Japan, respectively [20].

Figure 2. Relationship between geology and subsurface dams around Okinawa Island (modified from [20]).

Figure 2. Relationship between geology and subsurface dams around Okinawa Island (modified from [20]).

Conversely, the southern area of the main island of Okinawa, situated south of the Shimajiri line (referred to as the Tengan tectonic line according to Konishi [18]), belongs to the Shimajiri Belt of Tertiary bedrocks. This region is characterized by lowlands with an altitude of less than 200 m (Figure 2). The Shimajiri Belt comprises the Neogene Shimajiri Group and the Quaternary Ryukyu Group (see

Table 1. Comparison map of the Neozoic stratigraphy of the Ryukyu Islands, where subsurface dams were constructed (modified from [15]).

Geological Period			Ryukyu arc Islands
			Ishgaki etc.	Hateruma	Miyako	Kume	Okinawa	Ie	Izena	Okino erabu	Kikai
Cenozoic	Quaternary	Holocene		Coral reef	Coral reef	Coral reef	Coral reef	Coral reef	Alluvium S.	Coral reef	Coral reef
									Alluvium C.
		Pleistocene	Ryukyu G.	Ryukyu G.	Ryukyu G.	Ryukyu G.	Ryukyu G.	Ryukyu G.	Sandy Limestone	Ryukyu G.	Ryukyu G.
									Utibana F.
					Shimajiri G.		Shimajiri G.				Shimajiri G.
	Neogene	Pliocene		Shimajiri G.		Shimajiri G
		Miocene
			Yaeyama G.			Aratake F.
	Paleogene	Oligocene
		Eocene	Miyara G.
		Paleocene
Pre-Tertiary			Tomuru F. Tomisaki F.				Nagi F. etc.	Ie F.	Moromi F.	Neore F.
									Iheya F.	Granites
	Coral reef: Raised coral reef				Alluvium S.: Alluvium Sand Layer				Alluvium C.: Alluvium Clay Layer
		lack of sediment			Island with subsurface dam				Subsurface dam aquifer
					Impermeable layer for subsurface dams

). While sandstone is found in the lower part of the Shimajiri Group in certain areas, the upper mudstone layer is widespread. The total thickness reaches 2000 m. The Shimajiri Group is distributed across Kikai Island, Kume Island, the central and southern parts of Okinawa Island, Miyako Island, and its surrounding islands. However, it is not distributed in areas such as Ishigaki Island in the Southern Ryukyu Arc [15]. The Shimajiri Formation represents a layer where significant amounts of sand and mud from the Chinese continent were deposited on the semi-deep-sea floor during the Late Miocene to Pliocene period. Additionally, the lower layer of the Shimajiri Group on Miyako Island exhibits sedimentary facies from extremely shallow seas, inner bays, and near river mouths. Furthermore, the Shimajiri Group on Kume Island is composed of the Masha Formation, which is composed of shallow-marine sandy to muddy rocks, and the Ueshirodake Formation, which is composed of volcanic rocks [15]. Therefore, it is thought that the environment at the beginning of the deposition of the Shimajiri Group was a shallow marine environment, which changed to a deep marine environment in the latter half [14].

The Ryukyu Group is distributed as terrace deposits, reaching an elevation of approximately 200 m above sea level on most of the islands of the Central and Southern Ryukyus [15]. The average layer thickness is approximately 60 m. This group comprises porous limestone layers, which are characteristic of coral reef complexes, and layers containing terrigenous sediments. The main Ryukyu limestone was formed approximately 1 million to 0.5 million years ago [15].

The Shimajiri Group was deposited in a deep-sea environment, often referred to as a “sea of mud”. In contrast, the Ryukyu limestone was deposited in a “sea of coral reefs”. This transition occurred during the period between the late Pliocene (approximately 2.58 million years ago) and 1.7 million years ago [14]. The shift from a “mud ocean” to a “coral reef ocean” was dramatic and is attributed to significant tectonic movements. Specifically, this transformation is likely linked to the expansion of the Okinawa Trough [14]. The Okinawa Trough trapped abundant clay and sand from mainland China, potentially rendering the sea to the east of the Trough clear enough for coral growth. Before the formation of the Ryukyu limestone, the pre-tertiary rocks experienced uplift and erosion. This geological process affected not only the Shimajiri Belt but also the Iheya Belt. Notably, geological surveys of the Ie subsurface dam revealed a valley structure where rivers eroded accretionary zone rocks (see Section 4.3). This significant geological event is referred to as the “Shimajiri” tectonic movement [15].

During the final stage of deposition for the Ryukyu Group, the main part of this group was segmented into blocks due to tectonic movements. This specific tectonic event is referred to as the “Uruma” tectonic movement [15]. It involves the uplift or subsidence of the Earth’s crust, accompanied by fault activity. Notably, several fault structures extending in the northwest-southeast direction are recognized on Miyako Island (see Section 3.1) [14]. Conversely, in the southern part of Okinawa Island, the Ryukyu limestone forms a block-shaped plateau that gradually tilts southward due to numerous conjugate faults. These faults exhibit fault cliffs ranging from 10 to 20 m in the ENE-WSW and ESW-WNW directions (see Section 4.1) [15]. This block-like plateau topography contributed to the formation of cuesta topography and groundwater basins, which were utilized for constructing subsurface dams.

2.2. Geological Characteristics of Islands with Subsurface Dams

Table 1 presents a comparative overview of the geological stratigraphy of the Ryukyu Arc, specifically focusing on the locations where subsurface dams were constructed. These dams were built from north to south, encompassing Kikai, Okinoerabu, Izena, Ie, the southern part of Okinawa Island, Kume (referred to as the Central Ryukyu), and Miyako Islands (referred to as the Southern Ryukyu). Notably, all subsurface dams, except for the Senbaru Dam on Izena Island, were constructed using the Ryukyu limestone aquifer. Pre-Neogene tectonic belts classification further categorizes these islands as follows: Ie and Okinoerabu Islands belong to the Iheya Belt, and Kikai, the southern part of Okinawa Island, Kume, and Miyako Islands belong to the Shimajiri Belt.

The main aquifer on Izena Island is the Uchibana Formation of diluvium in the diluvial layer. It has also been deposited in underground valleys of the Senbaru lowlands. Geologically, it is feasible to construct a large-scale subsurface dam for saltwater intrusion prevention using the Uchibana Formation as an aquifer. However, in practice, small-scale subsurface dams were built, utilizing the alluvial clay layer as the base layer and the alluvial sand layer as the aquifer. The decision to adopt this approach was driven by the small scale (water storage capacity) required for the dam on Izena Island. For further details on the method of planning agricultural subsurface dams, please refer to Section 2.4.

2.3. Hydraulics of Subsurface Dam

2.3.1. Karstification and Groundwater Flow

When carbonate deposits resulting from the accumulation of corals and foraminifera are uplifted by tectonic movements, groundwater circulation contributes to the creation of karst landforms. These landforms include closed depressions (such as sinkholes, uvalas, and poljes), limestone caves, and subterranean rivers. The continuous development of these karst features is primarily due to carbonate dissolution [21]. Since karst is a specific medium with two simultaneous types of flow, laminar and turbulent, the management and development of subsurface dams in karst areas differ significantly from other more homogeneous environments [22]. Karst aquifers, which develop in fault-fractured limestone and soluble limestone, exhibit distinct types of porosity and aquifer properties. The evolution of groundwater flow influenced by karstification can be categorized into three stages: diffusion, mixing, and conduit [21]:

(1)

Diffuse Flow-type aquifer (Initial Phase): During this stage, groundwater flow occurs as a matrix flow through primary pores. Darcy’s law can be applied to groundwater flow in this state.

(2)

Mixed Flow-type aquifer (Transition Phase): Over time, conduits (such as caves) begin developing due to the dissolution of carbonate rocks. Sinkholes typically form during the initial phase of karst topography evolution. As sinkholes enlarge, surface water directly enters the conduit network, further promoting its development. Part of the groundwater flow passing through the conduit network becomes larger, leading to a shift from diffusion to mixed flow. Darcy’s law remains applicable to groundwater flow in this state, except in the conduit area.

(3)

Conduit Flow-type aquifer (Mature Stage): When the karst aquifer matures into the conduit type, groundwater flow concentrates in the major channels of the network, significantly reducing the dependence on matrix flow. Darcy’s law cannot be applied to groundwater flow in this state.

In conduit networks, such as those found in Croatian karsts, the groundwater level (GWL) changes extremely rapidly (up to 100 m in 24 h). The total difference between the maximum and minimum GWL in one piezometer can reach up to 200 m [22]. When modeling groundwater flow through an aquifer in a conduit network, an analytical model that assumes Darcy’s law cannot be applied. The biggest obstacle lies in the uncertainty regarding the dimensions and location of major conduits. It is unpredictable how large the hydraulic conductivity of a conduit will be and how widely it will vary. Consequently, reliably modeling the functionality of this system becomes challenging [22].

Karst topography is also commonly observed in areas where Ryukyu limestone is distributed. However, when examining the specific extent of the Ryukyu limestone region, the type of groundwater flow is not solely the conduit type. Instead, it is considered to be a mixed type, wherein a certain proportion of matrix flow coexists. This inference is supported by the following observation: two piezometers located close to each other within a limestone aquifer in the southern part of the main island of Okinawa exhibit completely different responses to the same amount of precipitation. Type 1: Fluctuates rapidly in response to rainfall. Type 2: Fluctuates slowly. Type 1 is likely related to caves, as limestone caves are distributed near the observation point. Conversely, Type 2 is attributed to matrix flow, as no caves are present [23]. Direct evidence supporting the notion of mixed flow in the Ryukyu limestone aquifer is that groundwater flow in the Minafuku and Sunagawa basins can be modeled using Darcy’s law, as demonstrated below.

2.3.2. Hydraulics of Ryukyu Limestone

The porosity and hydraulic conductivity of hard limestone, which is neither fractured nor dissolved, are less than 5% and 10⁻⁸ m/s, respectively, rendering hard limestone poorly permeable. However, the fractured or dissolved portions of limestone become highly permeable. For instance, when limestone with a porosity ranging from 2 to 27% and a permeability between 10⁻⁷ and 5 × 10⁻⁵ m/s undergoes fracturing, and the porosity increases by 0.1 to 0.4%, its water permeability can surge by a factor of 10 to 1000 [24]. Moreover, the permeability of the conduit is even greater. Figure 3 illustrates the frequency distribution of permeability for Ryukyu limestone in the Minafuku Dam basin of Miyako Island based on pumping tests [25]. This distribution exhibits an almost lognormal pattern, with an average value of 3.54 × 10⁻³ m/s. However, there are also small peaks at 10⁻⁴ m/s and 10⁻² m/s, indicating that the permeability of the Ryukyu limestone is not uniform.

The effective porosity is a critical parameter in determining the storage capacity of a subsurface dam. To estimate the effective porosity, several methods are employed, including (1) Pumping tests (to determine the storage coefficient), (2) Rainfall coefficient (which relates the water level rise to precipitation), (3) Neutron moisture logging (measuring the difference in volumetric water content between unsaturated and saturated conditions), and (4) Laboratory testing of the boring core samples. Figure 4 illustrates the effective porosity of the aquifer estimated using various methods [25]. However, it is important to recognize that these measurement methods have advantages and disadvantages, and they may not always be the optimal approach for determining the storage volume of subsurface dams. In the Miyako Island area, during a subsurface dam project, a large-scale water intake test was conducted at the Minafuku Dam [26]. The goal was to pump out nearly all of the stored water to accurately determine the effective porosity. Over a period of 145 days, three water intake tests were performed, resulting in a total pumping amount of 700 ×10³ m³. The drawdown during the pumping period was measured at 70 observation wells. Assuming a constant recharge amount, the effective porosity was calculated based on the pumping amount, drawdown volume of limestone, and water balance of the cut-off wall leakage. The resulting effective porosity ranged from 8.2% to 10.9%, with an average value of 10% adopted for the project [27].

The challenge with large-scale pumping tests lies in accurately estimating the recharge during the pumping period. In the case of the water pumping test at Minafuku Dam, the analysis was conducted under no-precipitation conditions and with constant recharge from the upstream basin. To address this, an integrated storage model was employed, which factors in both precipitation and recharge from upstream. Figure 5 illustrates the fluctuations in water levels over a five-year period. It spans from before the closure of the underground valley with the cut-off wall to the dam’s full reservoir level (EL.31 m) in Sunagawa Dam sub-basin 4 [28]. Additionally, the figure displays the simulated water level fluctuations for various effective porosity values: 5%, 8%, 10%, 12%, and 15%. Observation well data with water levels ranging from 14 m to 24 m (from May to December 1994) closely align with the 15% water level fluctuations. Conversely, data with water levels between 25 m and 30 m (from May to October 1995) exhibit a water level fluctuation close to 12% [28]. It is important to note that these values pertain to the effective porosity of sub-basin 4, while the design effective porosity of 10% for the entire basin remained unchanged.

2.3.3. Hydraulics of Basement Rocks of Subsurface Dame

The permeability of the Shimajiri Group mudstone, which serves as the hydrogeological foundation for a subsurface dam, is less than 1.0 × 10⁻⁷ m/s, rendering it an ideal base for such a dam [26]. On Kume Island, the Shimajiri Group tuff breccia exhibits a permeability of less than five lugeons [29]. Additionally, the permeability of sedimentary rocks, such as slate (which forms the base of the subsurface dam on Ie no Shima), is less than 1.0 × 10⁻⁶ m/s. Notably, tuffaceous slate has a small permeability (of the order of k = n × 10⁻⁸ m/s), making it function effectively as a hydrogeological base. Conversely, green rocks and chert exhibit k values greater than 1.0 × 10⁻⁶ m/s, indicating that they are leaky bedrock [30].

2.4. Agricultural Irrigation Plan Using Subsurface Dam as Water Source

First, let us explain the terminology [11]. The catchment area is the area that supplies water to the subsurface dam reservoir under natural conditions. The catchment area is determined by a low-permeability bedrock topography map but is generally estimated from the surface topography. The gross reservoir capacity is the amount of groundwater in the reservoir at the full reservoir level. The full reservoir level is the water level in the reservoir when the overflow begins from the crest of the cut-off wall. As mentioned in Section 3, the full reservoir level is not horizontal in the case of a subsurface dam. Water is usually taken from a subsurface dam by intake facilities, such as wells and collection wells. A significant head difference is required to collect groundwater. This is the same even when gravity intake is possible. Therefore, when designing intake facilities, it is necessary to ensure a minimum water depth that can extract the maximum planned intake amount. This water depth is determined by the hydraulic properties of the aquifer, layout density of the facilities, amount of base recharge in the planning reference year, etc. This water level is the minimum reservoir level, and the amount of water stored at that time is the dead capacity. Note that when a pump is installed in a well to pump water, the minimum reservoir level will drop in a funnel shape centered on the well. For surface dams, this corresponds to the storage capacity for sedimentation, but for subsurface dams, it is exactly the dead capacity [5]. Active capacity is the amount of groundwater obtained by subtracting the dead capacity from the gross reservoir capacity.

The storage capacity of agricultural subsurface dams in Japan varies, ranging from 240 × 10³ m³ at the Senbaru Dam on Izena Island to 10,500 × 10³ m³ at the Fukusato Dam on Miyako Island (see

Table 2. Agricultural Subsurface dam specifications in the Ryukyu Arc.

Prefecture	Island	Dam Name	Dam Type	Construction Period	Irriagtion Area ha	Dam Height (m)	Dam Length (m)	Gross Reservoir Capacity (1000 m3)	Construction Method for Cut-Off Wall
Okinawa	Miyako	Minafuku	subsurface	1977–1979	9156	17	500	700	Grouting
		Sunagawa	subsurface	1988–1993		49	1677	9500	SMW
		Fukusato	subsurface	1994–1998		27	1790	10500	SMW
		Nakahara	subsurface	2009–		55	2350	10500	SMW
		Bora	subsurface	2009–		26	2600	2200	SMW
	Okinawa	Komesu	subsurface	1993–2003	1352	69	2320	3460	SMW
		Giiza	subsurface	1999–2001		53	969	390	SMW
		Yokatsu	subsurface	1997–2007	225	68	705	3963	SMW
	Izena	Senbaru		1999–2008	520	14	497	790	Steel sheet pile
	Ie	Ie	subsurface	2004–	668	56	2612	1408	SMW
	Kume	Kaniin	Combined *1	1995–2005	338	52	1088	1580	SMW
Kagoshima	Kikai	Kikai	subsurface	1993–1999	1677	35	2281	1800	SMW
		Kikai 2
	Okinoerabu	Okinoerabu	subsurface	2007–	1497	48	2414	1085	SMW

*1 a surface water-groundwater storage type subsurface dam.

). Additionally, the construction of the fourth Bora subsurface dam is currently in progress as part of the Miyako Island project. The variation in dam size across different projects is primarily due to the differences in the irrigated area covered by each project.

The Senbaru subsurface dam serves as a water source for irrigating 520 ha of farmland in Izena Village [31]. This small-scale subsurface dam effectively prevents saltwater intrusion and has a gross reservoir capacity of 240 ×10³ m³. Notably, the geology of the Senbaru lowland features an underground valley, which could potentially accommodate the construction of a large subsurface dam (as discussed in Section 4.4). Despite the feasibility of constructing a large subsurface dam, the decision was made to construct a small subsurface dam based on the amount of water required to irrigate 520 ha of agricultural land, equivalent to 240 ×10³ m³ [31].

The capacity of agricultural irrigation dams in Japan refers to the dam capacity (also known as the dam dependence capacity) required to secure the necessary water supply during a drought year (which serves as the reference year for design). These drought years typically occur approximately once every 10 years. For the Izena Island Project, the planning reference year is 1991 [31]. Consumptive water use represents the annual amount of soil water absorbed by crops grown in an irrigated area. For instance, consider the daily consumptive water use of sugarcane from June to August, which amounts to 5 mm per day. It is important to note that this value can vary based on the crop species and season. According to the Izena Island Project [31], this consumptive water uses amounts to 3400 × 10³ m³. The net water requirement is calculated by subtracting the effective rainfall for the reference year (1991) from the consumptive water use.

When irrigating a field, there is a loss of irrigation water, so the gross water requirement is the net water requirement plus the loss. For instance, sprinkler irrigation losses account for 15% of the net water requirement. In the Izena project plan, the gross water consumption, which includes irrigation losses, amounts to 2.1 × 10⁶ m³ [31]. This gross water amount also encompasses any irrigation facilities constructed in the past, if applicable. The dependent water requirement of the water source is calculated by subtracting the current available water amount from the gross water requirement. Essentially, this value represents the amount of water that can be generated by constructing dams, among other measures. According to this plan, the dependent water requirement for the Izena project is 830 × 10³ m³. Specifically, 790 × 10³ m³ came from the Senbaru subsurface dam and surface dam, along with a small pond reservoir of 40 × 10³ m³ that was newly constructed [31].

3. Lessons from Constructions of the Minafuku Dam and the Sunagawa Dam

3.1. Subsurface Dams in Miyakojima Island

Miyako Island, situated in the subtropical zone, experiences an annual average temperature of 23 °C and a precipitation of 2200 mm. The Miyako Islands comprise Miyako Island itself and several surrounding remote islands, including Irabu Island, Ikema Island, and Kurima Island (see Figure 6a). The topography of Miyako Island is characterized by a relatively flat plateau, with its highest point reaching 113 m above the mean sea level (EL). In this study, we present a 3D terrain model and a 3D basement rock model for an island featuring a subsurface dam. The raster data for the 3D terrain model were obtained from the Geospatial Information Authority of Japan’s basic map information download service. We visualized these raster data using the free software SAGA-GIS v.9 [32].

To create the 3D foundation model, we followed these steps: (1) Convert the literature contour map to a digital contour map DXF file using CAD. (2) Convert the DXF file into a shapefile using the free software QGIS 3.34.9 [33]. (3) Input the shapefile into SAGA-GIS and convert it to a raster file using multilevel B-spline interpolation within the SAGA-GIS functions. (4) Display the raster files in 3D using the same SAGA-GIS functions employed for topographic maps. The 3D diagrams depicting other subsurface dams in this paper were generated using a similar approach.

Figure 6. Map of Miyako Island and surrounding remote islands (a), Miyako Island 3D topographic map (b), 3D bedrock (Shimajiri Group mudstone) topographic map (c) [34], A−A’ geological cross-section (d) [34]. Ryukyu limestone is distributed between maps (b,c). The original data of (c,d) are sited from [34]. The Minafuku Dam site is located in the midstream part of the Bora basin. For instructions on how to create a 3D map, see the main text.

Figure 6. Map of Miyako Island and surrounding remote islands (a), Miyako Island 3D topographic map (b), 3D bedrock (Shimajiri Group mudstone) topographic map (c) [34], A−A’ geological cross-section (d) [34]. Ryukyu limestone is distributed between maps (b,c). The original data of (c,d) are sited from [34]. The Minafuku Dam site is located in the midstream part of the Bora basin. For instructions on how to create a 3D map, see the main text.

The geology of Miyako Island consists of highly permeable Ryukyu limestone overlaying the Shimajiri Group mudstone. Figure 6b,c depict the 3D topography and 3D basement topography of the island [34]. Additionally, Figure 6d presents a geological cross-section along the A-A’ line [34]. Between Figure 6b,c, Ryukyu limestone is intercalated with a thickness ranging from 20 to 50 m. The ground surface is covered by lateritic clay soil called Shimajiri Marji soil, which is derived from Ryukyu limestone [35] and has a thickness of 0.5 to 1 m. Furthermore, Ohnokoshi clay (depicted in Figure 6d) is a clay deposit found in depressions of the Ryukyu limestone terrain with a thickness varying from 0 to 10 m. This clay originated from the reworking of lateritic clay soils derived from mudstone. It has also been suggested that the origin of the Shimajiri Marji soil and the Ohonokoshi clay is eolian dust and loess from the continent [35,36].

Miyako Island is divided into 23 catchments by numerous faults that orient from NW to SW and are distributed throughout the island [25]. Dams have been constructed in the Sunagawa and Fukusato basins, as shown in Figure 6c. Currently, the Miyako Islands project has expanded to include farmland on Irabu Island, which is connected by a 3540-m-long bridge (known as the Miyako-Irabu Islands project). As part of this project, the construction of the Nakahara subsurface dam (with a gross reservoir capacity of 10.5 × 10⁶ m³) and the Bora subsurface dam (with a gross reservoir capacity of 2.2 × 10⁶ m³) is underway. These dams will serve as additional water sources for new irrigation areas.

3.2. Core Technology for Subsurface Dam Construction

3.2.1. Integrated Reservoir Modeling

In the design of water intake and drainage facilities, numerical analysis models based on finite element and finite difference methods are employed to forecast short-term local fluctuations in the groundwater table. However, applying these models to long-term water use planning, specifically for daily water level calculations over a 30-year period, remains challenging. The integrated reservoir modeling described in the MAFF’s “Technical Manual for Planning and Design of Subsurface Dam” [37] has been successfully utilized in water use planning for numerous agricultural subsurface dam projects. Unfortunately, the intricate details of this technology are seldom documented in scientific papers.

The simulation model anticipated for the Minafuku Dam not only simulated groundwater flow but also necessitated a comparable level of analysis as surface dam planning. The specific analyses are as follows:(1) Calculating daily water levels over a 30-year period. (2) Determining the dam’s dependency on water volume and its capacity for the reference year of design based on the results of water level calculations.

Yoshikawa [5,6] initially attempted a two-dimensional, planar, unsteady infiltration model based on the finite difference method and the finite element method during the early stages of model development. However, several factors prompted him to abandon the use of these methods:

(1)

The finite difference method requires short calculation time intervals to stabilize the solution, making it challenging to perform simulations over a 30-year period.

(2)

These models calculate the hydraulic head potential, which renders them unsuitable for addressing groundwater depletion conditions in certain parts of the basin, particularly those occurring within the Miyakojima aquifer.

(3)

The process of feedback between the unsaturated infiltration recharge model and the saturated zone groundwater flow model, along with adjusting the calculated groundwater head based on observed data, demands significant effort and time.

In response to these challenges, Yoshikawa developed a novel analysis method known as the integrated storage model [5,38,39]. This approach is outlined in the literature, specifically in reference [5] and in subsequent works. However, it is important to note that this model cannot be universally applied to groundwater flow in all karst regions. Applicable karst areas are the diffuse and mixed stages of karstification.

This method involves dividing the target basin into several sub-basins. Boring surveys have revealed that the bedrock topography beneath the aquifer in the Minafuku Dam basin resembles a boat-bottom shape. In Figure 7a, the Minafuku Dam basin (with an area of 3.93 km²) is further partitioned into seven sub-basins (ranging from 0.3 km² to 1.1 km² each). An observation well (indicated by the red triangle) is strategically placed at the center of one of these sub-basins for model verification. Water movement within each sub-basin is described using a tank model, which is a simple conceptual model consisting of a series of three interconnected tanks.

Each sub-basin is represented by a tank model that integrates both the surface and subsurface systems. Within each sub-basin tank model: The uppermost tank and middle tank correspond to vertical infiltration in the unsaturated aquifer. The lowermost tank simulates saturated groundwater and has a boat-shaped design. The entire target watershed is depicted by connecting these tanks (as shown in Figure 7b). Here are the specific details for each tank: the Upper Tank with Height hi (mm), which contains two outflow holes for downward infiltration with heights h_b1 and h_b2. Each outflow hole has outflow coefficients of α₁ and α₂ (Figure 7c) [5]. Middle Tank: Also equipped with two outflow holes for downward infiltration.

Calculations for the uppermost and middle tanks are performed using height units (mm) for input precipitation (mm). If precipitation (−evapotranspiration) exceeds the height of the uppermost tank (mm), surface runoff from sub-basin 1 flows downstream into sub-basin 2. Additionally, the precipitation in the first-stage tank falls into the second-stage tank in proportion to the height of the outflow hole. The same mechanism applies to the amount of water falling from the second-stage tank to the third-stage tank.

When precipitation enters the lowermost groundwater system tank, it is converted into water volume Q (m³) by multiplying the height (h) with the catchment area. Subsequently, Q is further converted to the groundwater level (H) using the H-Q curve equation described below. The subsurface system interacts with adjacent sub-basins 1 and 2, and this interaction is proportional to the hydraulic head difference (governed by Darcy’s law, as expressed in Equation (1)):

$$\mathrm{Q}=KeS∆h/L,$$

(1)

where K is the permeability coefficient [m/day], e is the effective porosity [-], S is the water flow cross-sectional area between the upper and lower sub-basins [m²], Δh is the groundwater level difference between the lowermost tanks of the upper and lower sub-basins [m], and L is the distance between the tank centers [m]. S is calculated from the H-S curve equation.

The contour lines in Figure 7a represent the base rock topography based on the boring data. From this information, we can calculate the relationship between the volume of limestone, V, and the height, H (groundwater level), of Ryukyu limestone in each sub-basin, resulting in the H-V curve (as depicted in Figure 8a).

Assuming an effective porosity of λ, we can derive the H-Q curve, which relates the groundwater level (H) to the water storage volume (Q) (as shown in Figure 7a). Additionally, Figure 8b illustrates the H-Q curve for each sub-basin. Furthermore, based on Figure 7a, we can determine the relationship (H-S curve) between the boundary cross-sectional area (S) through which groundwater flows between sub-regions and the groundwater level (H).

The groundwater flow between sub-basins can be calculated using Darcy’s law equation by considering the water level difference between the two lowermost tanks and the cross-sectional area for water flow (as shown in Figure 9a) [23].

The permeability coefficient K represents the composite permeability between the two tanks, calculated using the following formula: In this context, the permeability for the upstream tank is denoted as k₁. The permeability of the downstream tank is denoted as k₂. Additionally, the corresponding lengths are d₁ and d₂.

$$K={\displaystyle \frac{d}{\frac{d_1}{k_1}+\frac{d_2}{k_2}}},$$

(2)

If a cut-off wall is inserted between the upstream tank and downstream tank, the permeability above the crest of the cut-off wall remains unchanged. However, below the crest, the composite hydraulic permeability K_wall, which includes k₃ and d₃ of the dam body, is expressed by the following equation (as shown in Figure 9b):

$$K_{ wall}={\displaystyle \frac{d}{\frac{d_1}{k_1}+\frac{d_2}{k_2}+\frac{d_3}{k_3}}},$$

(3)

The integrated storage model was also applied to the analysis of a confined aquifer (Senbaru subsurface dam [40]) and a coastal two-layer density flow (Nakajima subsurface dam [41]). Yoshimoto et al. [42] and Yoshimoto [23] developed a model that combined an integrated storage sub-model and a nitrogen balance sub-model to predict NO₃-N concentration fluctuations in the Komesu underground dam basin in the southern part of Okinawa Island and by incorporating a cut-off wall into this model (Equation (3)), they were able to predict NO₃-N concentrations after the construction of the Komesu subsurface dam (see Section 5.1). Notable features of their model include that ten sub-basins (see Section 4.1) are modeled on ten Excel sheets, and the relationships between each sheet are expressed by Darcy’s law (Equation (1)).

This model was able to calculate the hydraulic gradient of the groundwater table by dividing the basin into sub-basins. These gradients consistently exhibit values of 2/1000 or higher in each basin of Miyako Island [5]. However, subsurface dams have reduced this hydraulic gradient by 50% to 80% (e.g., Sunagawa Dam: 7/1000 to 0.1/1000, Fukusato Dam: 6/1000 to 3/1000) [43]. The excess water stored above the dam crest can be anticipated in advance during water-usage planning. For instance, if the basin area of the Minafuku Dam resembles a triangular basin, even with a 50% decrease in the hydraulic gradient, the surplus water storage volume would be 120,000 m³. This accounts for 30% of the effective water storage capacity of 410,000 m³. This surplus water represents a critical feature of the water storage function provided by subsurface dams.

3.2.2. Construction Method of Cut-Off Walls for Subsurface Dams

The Okinawa General Bureau summarized the issues with grouting methods as follows [26]: (1) The strength of Ryukyu limestone varies widely, ranging from 1.0 to more than 400 kgf/cm², and its distribution is heterogeneous [25]. (2) Many caves have developed in the region. Milanović [44] highlighted that the high heterogeneity of karst permeability makes it challenging to apply grouting. Specifically, the large hydraulic gradients imposed by the reservoir accelerate the leakage. These high gradients also intensify the dissolution expansion of cracks and bedding planes by several orders of magnitude [22].

The forward-stage injection method used for the Minafuku Dam is one of the standard grouting techniques. Figure 10 illustrates the permeability distribution map after the completion of the Minafuku Dam (Figure 10a) and provides details of the grouting specifications (Figure 10b). The target permeability improvement was set at 5 × 10⁻⁷ m/s. The experiment was conducted using two patterns:(1) a 7-line zigzag cross pattern and (2) a 4-line cross zigzag pattern. The spacing between lines in both patterns was 0.5 m (Figure 10b). In the case of the 7-line pattern, the total width of the cut-off wall was expected to be 5 m due to the spreading effect of injection from the outer rows. The red circles in Figure 10b indicate the locations of the check boreholes used to verify the permeability improvement of the cut-off wall. In the inspection borehole, packers were placed at every 5-m depth, and an injection permeability test was conducted. The improvement in the 4-row pattern area was greater than 1 × 10⁻⁶ m/s. In the 7-row pattern area, approximately 60% achieved a target water permeability coefficient of 5 × 10⁻⁷ m/s. The improvement up to 1 × 10⁻⁶ m/s was about 20% [25].

Several locations were observed where grout milk was leaking over the dam crest at an elevation of 33 m. Due to the risk that this leakage could obstruct the dam’s overflow, a 3.5 m diameter intake groundwater shaft was installed downstream of the cut-off wall near the center of the underground valley (Figure 10a). This vertical shaft includes a horizontal drainage borehole at the 33-m level of the dam crest, which penetrates the cut-off wall and serves for drainage from the reservoir side. Additionally, the shaft features a horizontal water intake borehole at the bottom [26].

The challenges preventing the improvement of permeability to the target level include the following: (1) Disturbance of the grout hole arrangement pattern due to hole bending, and (2) the inability to maintain consistent grout stage length every 5 m due to the heterogeneous hardness distribution of limestone [26]. To address these issues, the Double-Packer Grouting technique was employed. This low-pressure injection method allows for the repeated injection of different grout materials. To prevent the collapse of the excavation hole walls, an outer tube is used during injection. The Okinawa General Bureau conducted a test of this method, but unfortunately, satisfactory results were not achieved [26]. As a result, the evaluation of the cut-off wall construction in the Minafuku experiment indicates that achieving the target permeability of 5 × 10⁻⁷ m/s using the grouting method is challenging. Consequently, it was concluded that a new construction method is necessary for the Sunagawa and Fukusato subsurface dams [26].

In the early 1980s, there were examples of drilling into rocks using single-shaft augers, but only a few examples of construction using three-shaft augers [45]. The SMW method developed by Seiko Kogyo Co., Ltd. effectively constructed cut-off walls in tuff with uniaxial compressive strengths of 10 to 400 kgf/cm², similar to Ryukyu limestone, in subway construction in Fukuoka City from 1981 to 1984 [45]. The drilling speed of the single-shaft auger was 15 cm/min for rock with a uniaxial compressive strength of about 400 kgf/cm². The drilling speed of the three-shaft auger in the SMW method was 3.7 cm/min [45]. The Okinawa General Bureau focused on the SMW method as an alternative to the grouting method for cut-off walls.

The SMW method is a prominent technique within the deep mixing method (DMM). DMM encompasses various in situ soil treatment techniques that involve mixing soil with cementitious and/or other materials [9]. These materials, known as “binders”, can be introduced into soil treatment either in dry or slurry form. The injection process occurs through a hollow rotating mixing shaft equipped with a cutting tool at the tip. Additionally, the shaft above the tool may feature discontinuous auger flights and/or mixing blades or paddles (as depicted in Figure 11a,c).

Column diameters typically range from 0.6 to 1.5 m, with a typical maximum depth of 40 m [9]. The SMW method had a diameter of 0.5–0.7 m and could be constructed to a depth of 70 m [46,47]. The resulting cement-bonded soil materials generally exhibit higher strength, lower permeability (10⁻⁸ m/s or less), and lower compressibility than natural soils. Deep mixing method (DMM) technology primarily reflects the efforts of Japanese and Scandinavian construction companies over the past 50 years [9]. According to the developmental history of DMM technology, Seiko Kogyo played a pivotal role in advancing this technology in Japan [9].

The Seiko Kogyo (currently Seiko Tone Co., Ltd.) was founded in Osaka in 1965, inheriting technology from a company with experience in pile driving. In 1972, Seiko Kogyo embarked on the development of the Soil Mixed Wall (SMW) construction method. This method involves using multiple augers stacked on top of each other to enhance the continuity of the lateral treatment and the homogeneity of the treated soil. In 1976, the SMW method was commercially applied for the first time in Japan [9] and was used in the construction of the Fukuoka subway from 1981 to 1984 [45]. At that time, the SMW construction method was still relatively new and had not yet garnered widespread attention. Therefore, the Okinawa General Bureau carried out test construction on Miyako Island and obtained promising results regarding its applicability to Ryukyu limestone [8].

Based on the results of the trial construction, the Okinawa General Bureau decided to construct the cut-off wall using the SMW method in four phases (see Figure 11a):

(1)

Casing drilling with a single auger.

(2)

Advance drilling stage: During this stage, a single auger drills a guide hole while discharging the cement liquid to the lower end of the water-stop wall.

(3)

Three-axis drilling stage: In this stage, two left and right three-axis augers (each with a diameter of φ550 mm × mm³) are aligned with the guide hole and drilled while discharging cement liquid. The result is a 0.9-m-long and 0.5-m-wide cut-off wall. This wall is constructed by discharging solidified liquid (cement slurry) when the auger is pulled up and mixing it with the in-situ soil (as depicted in Figure 11d). If there is no auger wrap (approximately 50 mm), it will be rebuilt (referred to as an adjustment pile, as shown in Figure 12a).

(4)

Finally, in the last stage, the clamshell drains the earth and excess injected fluid, allowing the groundwater to overflow over the dam crest.

During the construction of the Ie subsurface dam, a cut-off wall with a maximum depth of 70 m was built. The auger shape for the deep section deviated from the standard φ550 mm × 90 cm pitch to a larger diameter of φ700 mm × 120 cm pitch. This modification resulted in a reduction in the number of adjustment piles (as shown in Figure 12b) [47].

Finally, the jet grouting method [48,49] should be mentioned, although there are no examples of its application to subsurface dams in Japan. The jet grouting method is of significant importance in subsurface dam technology since it was applied to the Parisha Riverbed subsurface dam (dam length: 756 m, dam height: 8.5 m, gross reservoir capacity: 429,000 m³), the first subsurface dam in China, in 1988 [50]. Furthermore, this method has been successfully used in the construction of five large-scale subsurface dams with a reservoir capacity of over 10 million m³ and has established its position as a core technology for subsurface dams in China [51]. The cut-off wall constructed using this method has a thickness of more than 0.18 m and a permeability of less than 1 × 10⁻⁸ m/s [50]. For details on this method, see [48].

4. Experience of Four Subsurface Dams in the Ryukyu Arc

4.1. Komesu Subsurface Dam

The Irrigation Project for the Southern District of Okinawa Main Island, implemented from 1992 to 2005, involved the construction of two subsurface dams, Komesu and Giiza, in Itoman City, located in the southern part of Okinawa Main Island (Figure 1 and Figure 2). Groundwater from these two dams irrigates fields covering an area of 1352 ha. In this section, we focus on the Komesu Dam, which represents Japan’s first full-scale subsurface dam designed to prevent saltwater intrusion.

The geology of southern Okinawa Island comprises the Shimajiri Group mudstone as the base layer and the Ryukyu limestone as the aquifer. This region can be divided into three basins: the Komesu, Makabe, and Ueshiro basins. Faults separate each basin (Figure 13). Specifically, The Komesu Basin terminates at the Komesu Coast. To the north of the Komesu basin lie the Makabe and Ueshiro basins. Based on the hydrogeological structure and water level distribution, these three basins (covering approximately 8.5 km²) are further subdivided into ten sub-basins ranging from 0.26 to 1.4 km² each [23,42]. Groundwater flows into the sea through three main routes (as depicted in Figure 13b) [23,42]:

Route 1: Komesu (4) -> Komesu (2) -> Komesu (1) -> Sea

Route 2: Makabe (2) -> Makabe (1) -> Komesu (3) -> Sea

Route 3: Ueshiro (2) -> Ueshiro (1) -> Komesu (6) -> Komesu (3) -> Sea.

The length of the cut-off wall of the Komesu Dam is parallel to the coastline and spans 2320 m. The deepest point of the wall reaches −69.4 m (Figure 14). Constructed using the SMW method, the permeability of this cut-off wall was verified through a boring survey after its completion, confirming a permeability of less than 1 × 10⁻⁸ m/s. The gross reservoir capacity of the dam is approximately 3.5 × 10⁶ m³, with an active capacity of 1.8 × 10⁶ m³. Additionally, eighteen pumping wells were installed in the pump site area, each with a maximum daily pumping capacity of 2000 cubic meters. In the design reference year (1971, with an annual precipitation of 1442 mm), the water level is predicted to decrease to a minimum level of −11.6 m above sea level (Figure 15) [52].

Figure 15a illustrates the electrical prospecting resistivity profile along the coastline and the vertical profile of electrical conductivity (EC) values from the observation wells [53]. The typical EC profile reveals a clear freshwater–saltwater interface in the low-resistivity section (<100 Ωm), which corresponds to soft limestone. However, in the high-resistivity section (>1000 Ωm, associated with hard limestone, the freshwater–saltwater interface is less distinct [53].

The project has defined a permissible salt concentration (chloride ion concentration) in irrigation water of 200 mg/L based on the salinity tolerance of green beans. Groundwater with a chloride ion concentration of 200 mg/L is similar in density to fresh water, behaving similarly. However, saltwater, which has a high chloride ion concentration, has a greater density than freshwater, making it a viable target for removal. For salinity control, the project considers water with an electrical conductivity of 5000 μS/cm or more (chloride ion concentration of approximately 1500 mg/L) as saltwater. The water level corresponding to this value serves as the “salt-freshwater boundary”, representing the top surface of the residual saltwater mass [54].

Before the dam closure in 1994, the depth of this interface was −20.0 m at the western end of the dam and −30.0 m at the eastern end (as shown in Figure 15b) [53]. However, after the dam’s completion in 2004, this depth further decreased to −30.0 m at the west end and −40.0 m at the east end (as depicted in Figure 16) [52].

The evolution of the residual saltwater mass during subsurface dam operation was investigated using indoor tank experiments and numerical simulations (e.g., [52]). According to these results, when groundwater from the upstream reaches the cut-off wall during rainfall, the saltwater mass is pushed upward. Therefore, it is estimated that the residual saltwater in the embankment body gradually decreases due to overflow when the subsurface dam is full (Figure 17a) [52,54]. However, with the same level of precipitation as the reference year of design, the groundwater level is predicted to decrease to a lower water level of EL −11.6 m. This has the potential to transport saline water into the catchment area, bypassing the embedding foundation of the cut-off wall and even penetrating through the cut-off wall (Figure 18b). Consequently, saltwater removal wells have been installed behind the cut-off wall for desalination (Figure 16) [52,54].

Yoshimoto et al. [55] investigated the distribution of residual saltwater 10 years after the cut-off wall sealed off the underground valley. Compared to the depth of the residual saltwater distribution in 2004, immediately after the completion of the cut-off wall, the boundary was approximately 5 m deeper. Furthermore, no expansion of the saltwater distribution due to seepage from the cut-off wall or foundation was observed. Sirahata et al. [56] developed a method for evaluating the function of cut-off walls in subsurface dams, which prevent saltwater intrusion by analyzing the tidal response of groundwater levels.

4.2. Kanjin Subsurface Dam

Kanjin Subsurface Dam is a surface water-groundwater storage-type subsurface dam located in the northwestern part of Kume Island. This type of dam, constructed by sealing a large cavern with concrete plugs, has been built in the karst areas of southern China [57], where there are more than 20 karst-type subsurface dams with storage capacities ranging from 10⁵ to 10⁷ m³ [57]. Examples include the Maguan Dam in Guizhou and the Shilin and Wulichong Dams in Yunnan [50]. These dams were successfully constructed by plugging large karst caves and grouting limestone cracks to create reservoirs. The Wulichong Dam stands out as one of the largest subsurface dams in China (and the world) [57]. The maximum depth of the cut-off wall is 260 m, with a length of 1333 m. The largest limestone cave within the dam area measures 33.46 m high, 13.9 m wide, and 2 to 10 m long. These caves were treated with concrete plugs [57]. The underground reservoir covers an area of 25.4 km² and reaches a depth of over 107 m, with a water storage capacity of 80 × 10⁶ m³.

Kume Island is approximately 100 km west of Naha City on Okinawa Island and is the westernmost of the Okinawa Islands (Figure 2). Kumejima has a trapezoid shape, with a base measuring approximately 13 km from northwest to southeast, covering an area of 59.53 km². The entire island is part of Kumejima Town, and the population is around 7000. Additionally, there are remote islands, such as Oujima Island, on the east side of Kume Island (Figure 18a).

The island is divided into four topographical and geological areas (Figure 18b) [29]: (1) The Tertiary Miocene Aratake Formation area, (2) the Early Pliocene Shimajiri Group area, (3) the Quaternary Pleistocene Ryukyu limestone area, and (4) the Holocene coastal area.

(1)

The first area is a mountainous region that includes Mount Arla (EL.287 m) in the southern part of the island, where the Aradake Formation, is composed of andesitic tuff breccia.

(2)

The second area is another mountainous region, including the highest point, Mount Ueshirodake (EL.326 m), on the northern part of the island. In this region, the Shimajiri Group consists of the Maja Formation, which is composed of sandy silt, sandstone, conglomerate, and tuff, and the Ueshirodake Formation, which is composed of basaltic lava and tuff breccia. A fault is estimated to exist between the first and second areas.

(3)

The Ryukyu Group is distributed in the northwestern part of the island, forming two level terraces: low terraces around EL.10 m and high terraces between EL.20 and EL.50 m.

(4)

The fourth area is located in the lowlands between the Aradake area and the Shimajiri area.

The original depression that became the surface water storage area, dammed by the Kanjin Dam, is an uvala of karst landform located on the border between the Ueshirodake area and the Ryukyu limestone area (Figure 19c). The uvala was formed by the erosion of surface water at the edge of a terraced plateau of Ryukyu limestone. The size of the uvala is 0.18 km², measuring 640 m NW-SE × 450 m NE-SW and approximately 20 m deep. The river water that entered the depression flowed subsurface into the Ryukyu limestone through five ponors (blue dots in Figure 19c). The groundwater flowed through the Ryukyu limestone and finally flowed out into the sea [29].

Since the permeability of the Ueshirodake Formation is generally less than 5 Lugeon units, this Formation functions as the base of a subsurface dam (Figure 19d). The Kanjin Subsurface Dam, which has a cut-off wall (length: 1088 m and height: 52.1 m) constructed in the Ryukyu limestone, stores groundwater in the Ryukyu limestone and also stores surface water in the uvala (Figure 20). The SMW construction method was applied to the cut-off wall construction, except for the cave treatment (described later). The grouting method has also been applied only in some areas [29].

At Kanjin Subsurface Dam, at its full water level of 25.7 m, the groundwater storage capacity of the limestone aquifer is 140 × 10³ m³ (with an effective porosity of 7%). This amount corresponds to approximately 9% of the irrigation requirement. The active capacity totals 1510 × 10³ m³, which is the sum of the water volume and surface water storage volume of 1370 × 10³ m³. The gross reservoir capacity, including the dead capacity (70 × 10³ m³) of the surface reservoirs, is 1580 × 10³ m³ (as shown in Figure 20a).

Spillways are required for surface water storage areas. As a result of the flood analysis [29], the flood volume, until reaching the critical high-water level EL 31.0 m was considered to be the temporary storage volume. When the flood volume exceeds the critical high-water level, the water discharged is planned to be sent from a spillway with an orifice to a doline seepage pond located downstream of the dam, where it will be infiltrated (Figure 20a).

In the cut-off wall construction of this type, large cave treatments can make construction progress difficult. No caves were recognized during the pre-construction boring surveys of the Kanjin Dam. During the construction of the cut-off wall, the existence of a cave was inferred from the overflow of the SMW drilling fluid. In order to understand the shape and properties of the cavity, a boring survey was arranged in a grid pattern and a microgravity survey as cave surveys were carried out. As a result, the size and shape of the cave and the condition of the inclusions inside were clarified [58].

Table 3. Properties of cavities distributed in the cut-off wall section of the Kanjin Subsurface Dam. The cave number corresponds to the cave number in the cross-sectional view of the cut-off wall in Figure 20b [58]).

No.	1	2	3	4	5
Shape	Single hole			Aggregation of multiple holes
Height	Ca. 3 m	Ca. 7 m	Ca.9 m	Ca. 2 m~10 m	Ca. 2 m~10 m
Altitude	EL. 12~15 m	EL. 15~22 m	EL. −3~6 m	EL. −10~17 m	EL. −15~25 m
Relationship with G.T.	Above G.T.		Below G.T.	Mostly below G.T.
Groundwater flow velocity	-	-	18~47 m/hr	Slight flow	19~26 m/hr
Properties of deposits in the cave	-	-	Compact clay with high moisture content	Sandy clay with gravel	Clay, silt, sandy clay with gravel, gravel
Filling rate of cavities with sediment	-	-	70%	78%	90%

G.T.: groundwater table.

shows the characteristics of the five caves [58]. The cave numbers in Table 3 correspond to the cavity numbers in the cross-sectional view of the cut-off wall, as shown in Figure 20b. The cavity was approximately 3 m to 10 m in width and height. Since the size of cavities that can be treated using only the SMW construction method is approximately 2 m [26], a special construction method was required to treat the cavities in the Kanjin Dam [58].

The caves were located both above and below the groundwater table. However, a large cave, specifically No. 3, which was approximately 9 m wide and high, was located below the groundwater table. These caves were perpendicular to the dam axis and extended both upstream and downstream. The internal materials consisted of relatively compact clayey soil and mud with a high-water content. Groundwater flow ranged from low to 47 m/h [58]. Large cave treatments for cavities without filling were plugged with concrete. In particular, deep caves No. 4 and No. 5 were sealed off by installing steel pipe sheet piles upstream and downstream of the caves as a pretreatment. After excavating the cofferdam using a full-turn auger machine, the earth and sand were removed using a hammer-grab. Finally, the caves were filled and closed with soil cement [58].

4.3. Ie Subsurface Dam

The Ie Subsurface Dam was constructed on Ie Island, 9 km northwest of the Motobu Peninsula in northern Okinawa Island (Figure 2). The cut-off wall length extends for 2612 m, with an average dam height of 55.9 m, and the gross reservoir capacity is about 1.4 × 10⁶ m³. The Ie Subsurface Dam has three notable features: (1) The basement of the subsurface dam is not the impermeable Shimajiri Group, but rather leaky Mesozoic accretionary mélange sedimentary rocks (see Table 1). (2) Unlike the subsurface dams on Miyako Island and Okinawa Island, which were constructed across underground valleys caused by faults, the Ie Dam is built across an underground valley resulting from old river erosion. (3) The maximum depth of the cut-off wall is 70 m or more [46].

The Ie Irrigation Project, which was implemented from 2004 to 2017, uses a subsurface dam as a water source to irrigate 668 ha of farmland. Ie Island is an oval plateau island with a long axis of 8 km. The total area is 22.77 km². The north coast is lined with cliffs approximately 60 m high, and the terrain slopes gently toward the south. The south coast consists mostly of sandy beaches. Most of the island area is composed of Ryukyu limestone. However, Mt. Shiro (EL.172 m), which has an isolated and sharp topography located slightly east of the center of the island, is composed of chert basement rock (Figure 21) [46].

The geology of Ie Island can be divided into three layers from the bottom layer to the top layer: the Cretaceous Ie layer (basement rock), basal gravel layer (basal conglomerate of unconformity), and Ryukyu limestone. Although the Ie Formation had been only confirmed at four sites, including Mt. Shiro [59], the subsurface dam boring survey revealed the detailed distribution of the Ie Formation around the cut-off wall (Figure 21c,d). The following rocks are distributed as a sequence of geological units from west to east around the dam site: greenstone, slate, and chert (Figure 21d). The strike direction of the strata is northeast-southwest, and the dip is northwest [46].

Before the deposition of the Ryukyu limestone, the Ie Formation was elevated by tectonic movements and eroded by rivers. The erosion valleys consist of a west underground valley on the right bank of the dam and an east underground valley on the left bank. Cut-off wall lines were planned across these valleys (Figure 21c). Figure 22 shows the geological cross-section along the cut-off wall. The western underground valley is 450 m wide and over 30 m deep, while the eastern underground valley is 540 m wide and over 20 m deep. The total length of the cut-off wall is 2576 m. This long wall is not just a straight line across the underground valley; rather, it forms a U-shaped zigzag shape (see Figure 21). The reason for this wall shape is the lack of a suitable pocket in the underground valley to store the irrigation water requirement. The following design philosophy was adopted: The chert layer with cracks and the greenstone layer containing limestone gravel with holes were defined as the leaky hydraulic basement layer. The line of the cut-off wall was planned to enclose only the slate layer distribution area as a reservoir. This design philosophy is based on the fact that the permeabilities of greenstone and chert are large (approximately k = 10⁻⁶ m/s or more), while the permeability of slate is small (approximately k = 10⁻⁸ m/s order) [30,46].

The Ie Dam cut-off wall construction was a challenging project because a deep construction section of 50 m or more accounted for 37% of the total construction length, and there was a section of rock where the maximum construction depth was approximately 75 m. Additionally, a layer of hard boulders with a maximum diameter of approximately 1 m was included at the bottom of the 20 m thick base layer that filled the underground valley, making construction difficult. However, these difficulties were overcome through improvements to the SMW method [46] and the development of a special tricone bit [47].

4.4. Senbaru Subsurface Dam

The Senbaru Subsurface Dam is a subsurface dam built on Izena Island to prevent saltwater intrusion. The purpose of the Izena Project (1999–2008) was to construct a facility to stably irrigate 520 ha of farmland on Izena Island. The Senbaru dam was constructed as part of a 790 × 10³ m³ water source for irrigation [31]. The topic of this section is the design of water intake facilities to prevent land subsidence and groundwater salinization.

Izena Island (14.12 km²) is an island located approximately 95 km northeast of Naha City (Figure 2). This island, along with the surrounding small islands of Gushikawa Island and Yanaha Island, makes up Izena Village. The outline of Izena Island shows a hexagonal shape, with each side approximately 2–3 km. The mountain range runs from Mt. Ufu (EL. 119.9 m) in the northwestern part of the island to Mt. Chijin (EL. 119.6 m) in the southeastern part of the island (Figure 23a). Mt. Menner (EL. 84.9 m) exists isolated in the northwestern area. Pre-Tertiary cherts are distributed in the area between Mt. Ufu and Mt. Chijin. The Senbaru lowland, an alluvial lowland, is distributed between Mt. Ufu and Mt. Menna (Figure 23b) [59,60].

The Pre-tertiary system of Izena Island is composed of layered chert (Iheya Formation: Permian to the Middle Triassic) and the Moromi Formation (Early Jurassic), which is mainly composed of alternating layers of turbidite sandstone and shale (Figure 23b). The Iheya F. is distributed near Mt. Ufu. The Moromi F. is distributed around Mt. Mennar [59]. Three levels of terrace are recognized around the Senbaru lowland: a high-level terrace of EL.45–50 m, a middle-level terrace of EL.30–45 m, and a low-level terrace of EL.8–17 m (Figure 23b). The thickness of the high- and middle-level terrace deposits is less than 5 m. They are stratified gravel beds containing chert breccia and cobblestones. The low-level terrace deposit is called the Uchibana Formation [59].

Figure 24 is a schematic geological cross-section across the Senbaru lowland along line A-A’ in Figure 23. The geology of the Senbaru lowland consists of a pre-tertiary basement, the Uchibana Formation, and the alluvial layers [59]. The depth of the Uchibana Formation reaches approximately 36 m. Ryukyu limestone is only distributed on a small scale in the western part of Izena Island [59]. Therefore, it is not suitable for target aquifers for groundwater development. The old dune deposits are well-sorted reddish-brown medium- to fine-grained sand layers. They are found above the Uchibana Formation in the innermost part of the Senbaru lowland. New sand dunes are developing almost parallel to the coastline near the Senbaru lowlands and village flatlands (Figure 23b). Alluvial lowlands, including the Senbaru lowland, are distributed to fill the mountains of Izena Island. The aquifers on Izena Island are the gravel layer of the Uchibana Formation, alluvial sand layers, and dune deposits [59,60].

The Senbaru Subsurface Dam, in a broad sense, built in the Senbaru lowland, consists of three facilities: a surface reservoir, a subsurface reservoir, and two PC tanks (Figure 23b). The subsurface reservoir is the Senbaru dam in the narrow sense with steel sheet pile cut-off walls (water storage capacity: 240 × 10³ m³ (porosity: 0.15), cut-off wall length: 479 m, height: 15.5 m). The stored groundwater is taken in by an intake trench buried in a comb-like shape within the storage area (Figure 23b and Figure 25a). The surface water reservoir (water storage capacity: 513 × 10³ m³) is a reservoir dug into the pre-tertiary Izena Formation. The water storage capacity of the first PC tank is 25 × 10³ m^3, and that of the second tank is 12 × 10³ m³. The total water storage capacity of Senbaru dam is 790 × 10³ m³, considering these water storage facilities as an integrated dam [31].

The alluvial layer in the Senbaru lowland is distributed from an altitude of 3 m, and the maximum thickness is about 35 m. The alluvium consists of a lower clay layer and an upper layer of fine to coarse sand mixed with coral (Figure 24a) [60]. The clay layer that forms the base of the Senbaru Dam has an average thickness of 5–7 m. Figure 24c shows a 3D model of the clay superstructure created from the boring data. The reservoir layers are the upper alluvial layer and the new dune layer. The thickness of these layers around the water intake facility is 3 m. [31].

The crest height of the cut-off wall was set to EL.+1.0 m since the average high tide level is EL.+0.919 m. This water level is the full water level (FWL). Aquifers (sandy layers) are distributed above EL.−2.0 m. The alluvial clay layer of the dam foundation is distributed below EL.−2.0 m. Therefore, the depth of the intake water pipe embedded in the intake trench was set to EL.−2.0 m. This depth is the low water level (LWL). The groundwater is taken from the aquifer between EL.+1.0 m (FWL) and EL.−2.0 m (LWL) (Figure 25b) [31].

The water intake pipe is a high-density polyethylene perforated pipe with a diameter of 450 mm (Figure 26a). The total length of intake trenches No. 1 to No. 4 is 1126 m (Figure 25a). Groundwater that flows into the water intake trench is collected in a relay tank. The collected groundwater is sent from the pump installed at the bottom of the relay tank to the PC tank via a water pump (Figure 25b). The amount of water intake is controlled by the water level in the relay tank. Water stored in subsurface dams is preferentially used for irrigation. If there is a shortage of groundwater storage, the water is switched to the surface reservoir by controlling the gate of the pipe that connects the relay layer and the surface reservoir (Figure 25b) [31].

The intake water amount can be calculated using a modified Equation (4) from the popular culvert formula. Equation (4) replaces the aquifer thickness term in the original equation with the average value of the upstream and downstream water heads [61]. Figure 26b shows the parameters of this equation.

$$Q={\displaystyle \frac{2\pi k\left(H_0-H_A\right)L}{In\frac{{sinh}^2\left\{\pi R/\left(H_A+H_0\right)\right\}+{sin}^2\left\{\pi a/\left(H_A+H_0\right)\right\}-1}{{sin}^2\left\{\pi \left(y_0+r\right)/\left(H_A+H_0\right)\right\}-{sin}^2\left\{\pi a/\left(H_A+H_0\right)\right\}}}},$$

(4)

where Q is the total amount of inflow into the culvert (in this analysis, pumped water amount), k is the permeability of the aquifer, H₀ is the initial water depth (water depth outside the influence area), H_A is the water head potential inside the culvert, L is the length of the culvert (unit length L = 1.0 m in this analysis), R is the sphere of influence, y₀ is the elevation difference from the center of the culvert to the impermeable layer, a is the height from the impermeable layer at the inflow point, and r is the radius of the culvert.

A continuous pumping test to verify Equation (4) was conducted for 6 days at a pumping rate of Q = 3825.5 m³/day. Figure 25a shows the water level drop contour map for the maximum water level drop. The area of influence from the intake water trench was 67 m–86 m. Figure 25a shows that the groundwater was pumped uniformly rather than in a funnel-like manner, as occurs with tube well pumping. The permeability of the aquifer calculated back from Equation (4) was 5.7 × 10⁻⁵ m/s [61].

The cut-off wall of the subsurface dam is only about 400 m from the coast. According to the simulation of the integrated storage model, the minimum reservoir water level in the planning base year is EL.−0.7 m. This situation could cause infiltration of saline water bypassing the cut-off wall embedded in the base clay.

The Izena project evaluated the flow rate and salinity concentration of saltwater flowing behind the cut-off wall through steady seepage flow analysis [61]. The analytical model was able to reproduce the actual water intake amount and the salinity concentration of the intake water with an error of less than 3%. The project used this model to perform a number of simulations by varying the actual water intake and storage parameters. By regression analysis of the results, the project derived the relationship between the salinity TE (mg/L) of groundwater taken from the trench and the groundwater level (Equation (5)) [61].

$$TE=a\times {\left(EL\right)}^2+b\times EL+c,$$

(5)

where, EL = groundwater level of the reservoir, a = 0.0092X − 11.743, b = −0.0308X + 47.085, c = 0.0275X + 271.61. X = Electrical conductivity of the diluvial layer (μS/cm).

Equation (5) calculates the conditions under which seawater moves solely by advection, driven by the pressure difference between the sea surface and the reservoir level. When the water level in an underground reservoir drops below sea level, seawater infiltrates into the reservoir. By utilizing Equation (5), the project can predict changes in the electrical conductivity of the seawater that bypasses the bottom of the cut-off wall, as well as the salinity of the trench intake water, based on fluctuations in the reservoir groundwater level. The predictions of the project indicated that changes in the salinity of intake water when assessed against the water level in the planning reference year, would remain within the allowable limit of 500 μS/cm [61].

5. Discussion

5.1. Environmental Impacts of Subsurface Dams

In the early 1990s, when construction of the Sunagawa Dam began, the concentration of NO₃-N in groundwater steadily increased in some areas of Miyako Island and the southern part of the main island of Okinawa [62,63,64,65]. These areas were predicted to exceed the environmental standard of 10 mg/L in the near future [65]. This situation also occurred in some agricultural areas of Japan. In 1993, the Ministry of the Environment (ME) of Japan designated the concentration of NO₃-N in surface water and groundwater as a monitoring item and started annual monitoring [66]. In 1999, the ME updated the environmental law to set 10 mg/L of NO₃-N in groundwater as the environmental standard [66]. NO₃-N pollution can be caused by anthropogenic sources such as fertilizers, livestock manure, and domestic wastewater. In areas where NO₃-N pollution is occurring or likely to occur, the goal was set to identify the source of NO₃-N pollution and prevent the progression of the pollution. This goal was also set in Miyakojima and the southern part of the main island of Okinawa; however, the fact that NO₃-N concentrations could be related to subsurface dams in these areas made the solution even more complicated.

When a subsurface dam is constructed, the natural flow of groundwater is artificially altered, resulting in congestion of the groundwater flow below the full water level in the basin behind the dam. Furthermore, if the groundwater pumped from the subsurface dam basin is repeatedly used to irrigate farmland in the same basin, part of the groundwater will be recycled through the same upland soil. This recycling process repeatedly loads the groundwater with nitrogen leached from farmland soil, leading to an increase in the NO₃-N concentration in stagnant groundwater, which has been a concern (e.g., [65,67]).

The increase in NO₃-N concentration in groundwater in the subsurface dam basin consists of two components: (1) The NO₃-N concentration component due to increases in load sources (chemical fertilizers, livestock waste treatment, and domestic water) that occur commonly throughout Miyako Island, including the subsurface dam basin, and (2) the NO₃-N concentration component caused by changes in the natural groundwater flow due to the subsurface dam and repeated irrigation of groundwater from the reservoir area. The increase in NO₃-N concentration caused by component (1) occurs even in basins without subsurface dams. This increase in NO₃-N concentration can be reduced by reducing the load source. The increase in NO₃-N concentration caused by component (2) is a phenomenon that occurs only in subsurface dam basins. To stop the increase in NO₃-N concentration, it is necessary to adjust the height of the cut-off wall crest and land use above the dam reservoir. When assessing the impact of a subsurface dam on NO₃-N concentration, it is necessary to clearly distinguish between the two components. Some literature does not mention this prerequisite and attributes the increase in NO₃-N concentrations solely to dam construction (e.g., [68]).

According to nitrate–nitrogen monitoring by the Ministry of the Environment of Japan (ME), the rate of exceeding environmental standards in groundwater across Japan gradually increased from 3% in 1994. Since 1996, this rate has remained at a high level of 5–6%. However, after the Livestock Excretion Act was fully implemented in 2004, the rate of exceeding environmental standards began to decrease and has recently remained at around 3% [66]. The Livestock Manure Act (1) regulates the piling up and digging up of livestock manure in order to ensure its proper management (treatment and storage), and (2) promotes the use of livestock manure [69].

On Miyako Island, the NO₃-N concentration in groundwater began to decrease around 1989, before the Livestock Excretion Law came into force. There are several possible reasons for the decrease in the NO₃-N concentration. One important reason for this is that the standard amount of fertilizer for farmland was lowered in Okinawa Prefecture in 1986 (e.g., the standard amount of nitrogen fertilizer for summer-planted sugarcane was lowered from 310 kgN/ha to 240 kgN/ha) [70]. Another possible cause is that since about 1980, part of the sugarcane fields has been converted to pasture as a result of the expansion of livestock farming [67], based on the fact that Okinawa’s pasture productivity is three times the Japanese average [71]. Since 1989, the NO₃-N concentration in groundwater has decreased and has recently remained at around 5 mg/L (Figure 27). Therefore, when assessing the impact of subsurface dams on NO₃-N concentrations on Miyako Island, the impact of the underground dam will be evaluated based on the decreasing trend in NO₃-N concentrations.

Recent studies on the impact of subsurface dams on NO₃-N pollution have led to two different conclusions. One opinion, represented by Ishida et al. [43], is that subsurface dams do not deteriorate groundwater quality. The other opinion, represented by Yoshimoto et al. [42] and Su et al. [50], is that subsurface dams do deteriorate groundwater quality.

Ishida et al. [43] analyzed monthly NO₃-N concentration trends and changes in NO₃-N concentration distribution patterns in the backwater basin of the cut-off wall over a 14-year period, including before and after the construction of the Sunagawa subsurface dam in Miyakojima. As a result, they concluded that the construction of the subsurface dam would not have a negative impact on groundwater quality based on two points: the NO₃-N concentration trend in the subsurface dam basin follows the same decreasing trend as the overall trend in Miyakojima, and the construction of the subsurface dam has had no effect on this trend.

On the other hand, Yoshimoto et al. [42] developed a model combining a water balance tank sub-model and a nitrogen balance sub-model to predict NO₃-N concentration fluctuations in the Komesu subsurface dam basin in the southern part of Okinawa Island. In their study, they predicted future NO₃-N concentrations under the simple assumption that annual precipitation would decrease at a rate of 5.5% per 100 years. As a result, they estimated that the minimum groundwater level would decrease in the subsurface dam basin and that the NO₃-N concentration would increase by 0.79 to 1.46 mg/L over 100 years. Sun et al. [50] investigated NO₃-N concentrations before and after the completion of the cut-off wall (2006) in the saltwater intrusion prevention Wanghe subsurface dam reservoir in Shandong Province, China. Their findings showed that a subsurface dam hindered groundwater circulation and caused nitrate to accumulate in the reservoir. In addition, Ke et al. [71] and Gao et al. [72] emphasized that the construction of a subsurface dam affects nitrate pollution upstream based on laboratory experiments and numerical simulations. However, it should be noted that their study did not mention the NO₃-N concentration components in the basin without a subsurface dam, as mentioned above. For example, if the assumption made by Yoshimoto et al. [42] that a decrease in recharge would occur, the NO₃-N concentration would increase even in basins without subsurface dams.

Whether the construction of the underground dam affected the time series of NO³-N concentrations could be analyzed using causal inference. Before examining the causal inference, let us consider the impact of pollution due to groundwater recycling. This is because irrigation engineering states that irrigation water does not affect groundwater when irrigated below the TRAM. If groundwater recycling does not impact pollution, then the examination of the causal inference becomes clear. TRAM (Total readily available moisture) is the maximum amount of water that can be used in one irrigation when the field is sufficiently dry. TRAM is calculated according to the following formula [73]:

$$TRAM=\left(FC-Q_w\right){\displaystyle \frac{D}{C_p}}$$

(6)

where, FC (vol%) = field capacity, Qw (vol%) = moisture content of irrigation starting point, and D (mm) and C_p (%) = thickness and soil moisture extraction pattern of important soil layer for growth of an effective soil layer, respectively. Yoshinaga [74] showed that the representative TRAM of the Shimajiri Marji soil on Miyako Island was 30 mm, so here we assume TRAM = 30 mm.

Once the TRAM is determined, the irrigation interval can be calculated using the following formula [73]:

$$Ii={\displaystyle \frac{TRAM}{daily water saving irrigation by crops}}$$

(7)

For example, if the consumptive use of sugarcane is 5 mm per day, even if 30 mm of irrigation is applied every six days, this irrigation water will simply be taken up by the sugarcane or evaporate and will not affect the groundwater.

Data from lysimeter experiments conducted at the Okinawa Prefectural Agricultural Research Center Miyako Branch from July 2005 to October 2006 [75] validated the concept of TRAM. The size of the lysimeter unit was 3 × 3 m^2, and the depth was 135 cm. Summer-planted sugarcane was planted in lysimeter units filled from bottom to top with Shimajiri Marji soil, a typical soil of Miyako Island. Chemical fertilizer was applied at 240 kgN/ha according to the fertilization standard [75]. In addition to precipitation, irrigation water was applied as needed, and the movement of water and leached fertilizer through the soil was monitored [75]. The cumulative amount of precipitation during the experiment was about 3014 mm (the Kagamihara Meteorological Observatory). Although Nakagawa et al. [75] did not indicate the amount of irrigation water, the irrigation water was estimated based on the amount of water consumed by the sugarcane, and a TRAM of 30 mm is about 1155 mm. Thus, a total of 4169 mm of water was applied. The monitored infiltrated water volume was 2570 mm. The total amount of leached nitrate was 63.6 kgN/ha [75] (Figure 28).

On Miyako Island, groundwater runoff from precipitation is assumed to have a runoff rate of 0.4 [64,76]. If this assumption is applied, the infiltration from precipitation is 1206 mm. The infiltration from irrigation water is 462 mm. However, the total of 1668 mm of water cannot explain the measured total infiltration water of 2570 mm. If we assume that only precipitation exceeding 30 mm runs off into groundwater, the measured values can be explained as follows. First, precipitation over 30 mm was extracted from the precipitation data of the meteorological observation station every 5 days. The cumulative precipitation was calculated. The cumulative precipitation over 30 mm was 2519 mm. This is almost equal to the total infiltration water of 2570 mm. Figure 28 also shows the variation in cumulative precipitation. Note that the accumulation process of precipitation almost perfectly traces the accumulation process of the monitored infiltration water. This indicates that most of the irrigation water is not added to groundwater through crop consumption and evapotranspiration, and therefore, irrigation water below 30 mm cannot add leached chemical fertilizer to groundwater.

Causal inference is carried out through intervention analysis of time-series data. Generally, an intervention in intervention analysis means a procedure, law, policy, etc., that aims to change the trend of time-series data. Intervention analysis has also been used in the impact assessment of surface dams. Hipel et al. [77] first applied intervention analysis to the impact assessment of surface dams to analyze the impact of the old Aswan Low Dam (which was built in 1902). The analysis showed that a significant decrease in flow occurred in 1903 when the dam was put into operation. They suggested that there is a great possibility of applying intervention analysis to the impact analysis of water resources from dam construction.

Sakizadeh and Chua [78] conducted an intervention analysis of the impact of the Karkheh Dam (which was built in 2001) in Iran on groundwater quality using a Bayesian structural time-series model. Bayesian statistics is a mathematical procedure that applies probabilities to statistical problems based on Bayes’ theorem. Bayes’ theorem is a mathematical model for estimating how the probability of an unknown state (prior probability) changes when a certain factor occurs (posterior probability). They analyzed the time-series data of several water quality types, including total dissolved solids (TDS), from 1996 to 2012. The results demonstrated that the construction of the dam reduced TDS concentrations by approximately 43%.

Imaizumi evaluated the impact of a subsurface dam using intervention analysis with a Bayesian structural time-series model (unpublished data). The CauseImpact package in R was used for the intervention analysis. The response time series (the time series of NO₃-N concentrations at Kajidou in the Fukusato basin where the subsurface dam was constructed) and the control time series (the time series of NO₃-N concentrations at Muiga spring in the Nakahara basin where the subsurface dam was not constructed) were input to the model (Figure 27). Figure 29 shows the results of the analysis. The solid line in the “original” panel in the top figure represents the observed data of NO₃-N concentrations in the Fukusato subsurface dam basin. The dotted line represents the estimated data (i.e., concentration fluctuations in the absence of the subsurface dam). The blue band area represents the 95% confidence interval of the estimated data.

Notably, around 2004, the observed data began to fall outside the confidence interval; after 2009, they consistently fell outside the confidence interval. Subsurface dam intervention resulted in a decrease in NO₃-N concentration compared to the scenario without intervention. The middle “pointwise” panel illustrates the difference between the observed data and counterfactual predictions. Finally, the bottom “cumulative” panel depicts the cumulative effect of subsurface dam intervention.

According to the Causal Impact summary report, the average value after the construction of the subsurface dam is 5.7 mg/L. This is 17% lower than the average concentration of 6.9 mg/L if the subsurface dam had not been constructed. The probability that this effect is obtained by chance is extremely small (Bayesian one-tailed area probability, p = 0.001), which means that the causal relationship with the subsurface dam can be considered statistically significant.

The decrease in NO₃-N concentration is unlikely to be caused by the stagnation of groundwater flow due to the construction of the subsurface dam. This is more likely to have been caused by the dilution of groundwater with high NO₃-N concentrations due to the inflow of groundwater with low NO₃-N concentrations into the space created by pumping out a large amount of stored water in the Fukusato Dam reservoir (Imaizumi unpublished data). This result supports the conclusions of Ishida et al. [43]. Therefore, it can be concluded that the Sunagawa and Fukusato subsurface dams on Miyako Island do not contribute to an increase in NO₃-N concentrations in groundwater but rather contribute to a decrease in NO₃-N concentrations.

The anisotropic permeability structures of alluvial gravel and Ryukyu limestone are completely different. The results of the causal inference shown here are for groundwater in the Ryukyu limestone aquifer. It should be noted that the applicability of these results to subsurface dams in alluvial gravel aquifers in coastal plains has not been proven. Based on numerical simulations, Goa et al. [79] pointed out that subsurface dams in alluvial gravel aquifers can unintentionally expand the nitrate accumulation zone, especially when a low-permeability layer is present.

5.2. Relationships between Catchment Area, Gross Reservoir Capacity, Active Capacity of Subsurface Dams in the Ryukyu Arc

Here, the characteristics of agricultural subsurface dams in the Ryukyu Arc are discussed based on published data on the catchment area, total storage capacity, and effective capacity of subsurface dams. The data and sources used in the discussion are summarized in

Table A1. Specifications of water source dams for national irrigation and drainage projects.

	Subsurface Dam															Dam Construction Cost Rate ****		0.570
No.	Prefecture	Island	Project District	Dam Name	Location	Construction Period	Gross Reservoir Capacity (×103 m3)	Active Capacity (×103 m3)	Active Capacity Rate *	Dead Capacity (×103 m3)	Dead Capacity Rate **	Dam Height (m)	Dam Length (m)	Catchment Area (km2)	Irrigation Area (ha)	Total Project Cost ($)	Subsurface Dam Construction Costs ($)	Water Price *** ($/m3)	References
1	Okinawa	Miyako	Miyako	Minafuku	Miyako-jima City	1977–1979	700	400	57.1%	300	42.9%	16.5	500	1.2	8160	8,022,640			[80,89,90]
2				Sunagawa		1988–1993	9500	6800	71.6%	2700	28.4%	50	1677	7.2		640,000,000	365,000,000	25
3				Fukuzato		1994–1998	10500	7600	72.4%	2900	27.6%	27	1790	12.3				Sunagawa (24), Fukuzato (26)
4			Miyako·Irabu	Nakahara		2009–	10500	9200	87.6%	1300	12.4%	55	2350	9.5	996	663,150,000	378,202,734	35	[89]
5				Bora		2009–	2200	1600	72.7%	600	27.3%	26	2600	5.2
6		Okinawa	Southern Okinawa Island	Giza	Yaebise Town	1999–2001	390	210	53.8%	180	46.2%	53	969	1.2	1352	373,000,000	212,726,563	105	[91,92]
7				Komesu	Itoman City	1993–2003	3460	1810	52.3%	1650	47.7%	69.4	2320	3.9
			Yokatsu	Yokatus	Uruma City	1997–2007	3963	1382	34.9%	2581	65.1%	67.6	722	2.9	225	72,770,000	41,501,641	30	[93,94]
8		Ie	Ie	Ie	Ie Village	2004–2017	1408	754	53.6%	654	46.4%	55.9	2612	1.4	668	16,770,000	9,564,141	13	[95,96]
9		Izena	Izena	Senbaru	Izena Village	1999–2008	790	240	30.4%	550		13	550	2.4	520	143,000,000	81,554,688	340	[97]
12		Kume	Kanjin	Kanjin	Kume City	1995–2005	1580	1510	95.6%	70	4.4%	57.6	1070	0.8	338	135,000,000	76,992,188	51	[97]
10	Kago-shima	Kikai	Kikai	Kikai	Kikai Town	1993–1999	1800	1313	72.9%	487	27.1%	35	2280	3.9	1677	251,240,000	143,285,313	109	[98,99]
11		Okinoerabu	Okinoerabu	Okinoerabu	Tina Town	2007–2018	1085	596	54.9%	489	45.1%	48.2	2414	13.26	1497	350,150,000	199,694,922	335	[100]
	Surface dam
1	Okinawa	Ishigaki	Miyaragawa	Masesato	Ishigaki City	1977–1982	2300	1300	56.5%	1000	43.5%	27	367	4.82	3460	389,220,000	18,010,000	14	[82,101,102]
2				Sokobaru		1982–1992	13000	12850	98.8%	150	1.2%	29.5	1331	5.04			179,080,000	14
3				Ishigaki		1979–1992	420	400	95.2%	20	4.8%	18.5	65	1.5			9,290,000	23
4			Nagura	Nagura		1980–1998	3970	3820	96.2%	150	3.8%	38.7	400	3.45	760	259,600,000	147,090,000	39
5		Okinawa	Haneji-ohokawa	Makiya	Nago City	1985–2006	1470	1260	85.7%	210	14.3%	33.6	171	4.2	1326	393,250,000	142,420,000	113

* Active capacity rate = Active capacity/Gross reservoir capacity. ** Dead capacity rate = Dead capacity/Gross reservoir capacity. *** Water price = Active capacity (m³)/Subsurface dam construction cost ($). **** Dam construction cost rate = Subsurface dam construction costs/Total project cost of the Miyako project.

in the Appendix A.

Figure 30a shows the relationship between the gross reservoir capacity (x) and catchment area (y). The figure shows that Okinoerabu Dam is plotted completely apart from the other subsurface dams. If Okinoerabu Dam is excluded, there is a high correlation between the gross reservoir capacity and catchment area, shown as y = 0.0009x + 1.0026 (R² = 0.85). Since the underground geological structure of Okinoerabu Dam is more complex than expected, the catchment area determined from the surface topography may not accurately represent the groundwater catchment area.

Figure 30b shows the relationship between the gross reservoir capacity (x) and active capacity (y). There is a high correlation between these items, as shown by y = 0.7916x − 345.02 (R² = 0.96). The ratio of active capacity to gross reservoir capacity ranges from 30.4% (Senbaru Dam) to 95.6% (Kanjin Dam), with an average of 62.3%. The high ratio of the Kanjin Dam is because it is a surface water-groundwater storage-type subsurface dam.

Figure 30c shows the relationship between the gross reservoir capacity (x) and dead capacity (y). The published active capacity should be the value obtained by subtracting the dead capacity determined by the simulation from the gross reservoir capacity. Because only gross reservoir capacity and active capacity have been published, here we use the formula: dead capacity = gross reservoir capacity − active capacity. The dead capacity rate is the dead capacity/gross reservoir capacity. As mentioned in Section 4, the dead capacity of the saltwater intrusion prevention-type Komesu Dam and the surface water-groundwater storage-type Kanjin Dam are different from those of other subsurface dams.

5.3. Characteristics of Construction Costs of Agricultural Subsurface Dams in the Ryukyu Arc

The cost of subsurface dams is discussed in this section based on the total publicly disclosed costs of irrigation projects. The yen-dollar exchange rate from 1990 to 2010, when many agricultural subsurface dams were constructed, ranged from 125 to 80 yen per dollar; therefore, the yen-dollar exchange rate was standardized at 100 yen per dollar. Project costs and sources are shown in Table A1 in Appendix A.

The total project cost of the upland irrigation (8160 ha) project on Miyako Island, which uses the Sunagawa and Fukusato subsurface dams as water sources, is $640 million [80]. The subsurface dam construction cost is 57% of the project cost ($365 million). The remaining project cost is for six farm ponds (total water storage capacity of about 140,000 m³) and the water transmission pipeline system (134 km) [80]. Details of subsurface dam construction costs have not been published except for the Sunagawa and Fukusato dams [26]. Therefore, the project costs of subsurface dams other than the Sunagawa and Fukusato dams were roughly estimated by multiplying the total project cost by 0.57. Therefore, it should be noted that the discussion here is only an estimate.

Of the construction costs for the Sunagawa and Fukusato dams, 73% (32% for the Sunagawa Dam and 41% for the Fukusato Dam) were for the cut-off walls and intake facilities (well construction and pump costs). 11% was spent on design and investigation costs. Of the construction costs for the Sunagawa and Fukusato dams, the costs of intake facilities were 37.5% and 39%, respectively (Figure 31).

The design and investigation costs for the Sunagawa and Fukusato dams were 11% (about $40 million) of the total construction cost of the subsurface dam [26]. This cost includes the design cost of the facilities, but most of the cost is for investigations, including drilling investigations. It is impossible to create a bedrock contour map with the accuracy of the surface topography map. In addition, since the water intake facilities are installed underground, information on the subsurface geology is required. Therefore, the proportion of investigation costs to the construction costs of a subsurface dam tends to be higher than that of a surface dam. For example, in the geological investigation of the Sunagawa Dam, geological investigation drilling was carried out at about 500 points, covering a total of more than 20,000 m. In addition, more than 200 pumping tests were also carried out [36]. The density of the drilling investigation was 69 holes/km² (2780 m/km²).

Vertical electrical surveys were also carried out at about 500 points to investigate the bedrock depth [81]. Most of the electrical surveys were carried out in the pre-project planning stage to investigate the feasibility of building a subsurface dam, and their costs are largely not included in the project cost. It should be noted that these data from the electrical surveys were not used to draw bedrock topography maps for the projects. This is related to the fact that the correlation between the bedrock depth estimated by electrical surveys and the actual bedrock depth by drilling surveys was R² = 0.79 [81]. The difference between the estimated bedrock depth and the actual measured depth was 10–30 m. The difference tended to increase with bedrock depth [81]. Since subsurface dams store groundwater over a wide area, an error of several meters can result in an error of several tens of percent of the gross reservoir capacity. Therefore, only data from drilling surveys are used to draw bedrock topography maps for agricultural subsurface dams in Japan. Geophysical surveys are only applied to special cases, such as cavity surveys for subsurface dams (Section 4, Kanjin Dam).

5.4. Characteristics of Water Prices of Agricultural Subsurface Dams in the Ryukyu Arc

Finally, let us consider water prices (construction costs/active capacity). It should be noted that the water prices of other subsurface dams, other than the water prices of the Sunagawa and Fukusato Dams, have rough assumptions in the estimation of construction costs, so the discussion here is only an estimate. The water prices of Sunagawa Dam and Fukusato Dam are about $24/m³ and $26/m³, respectively. The water prices, including other subsurface dams, range from $13/m³ (Ie Dam) to $340/m³ (Okinoerabu Dam). In Okinawa Prefecture, four agricultural surface dams were constructed on Ishigaki Island and the northern part of Okinawa Island between 1982 and 2006. Their construction costs have been published [82]. These water prices ranged from $14 to $113/m³ (Maezato Dam $14/m³, Sokobaru Dam $14/m³, Nagura Dam $39/m³, Ishigaki Dam $23/m³, and Makiya Dam $113/m³). Although it is difficult to compare with Japanese surface dams because calculation conditions are different, the water prices of 32 irrigation surface dams in Australia range from $0.5/m³ to $37.5/m³ [83]. The water prices of subsurface dams in the Ryukyu Arc have been shown to range from the same to slightly higher than those of surface dams.

6. Conclusions

The core technologies of the large-scale subsurface dams constructed in the Ryukyu Arc are the integrated water storage model, which calculates the groundwater flow and water balance, and the Soil Mixed Wall method (SMW method) used for embankment construction. The integrated water storage model contributed to obtaining the budget for the construction of the subsurface dam by visualizing the daily fluctuations in the groundwater level over a period of 30 years. The SMW method was applied to the Sunagawa Dam, the first large-scale subsurface dam in Japan, and has since been applied to saltwater intrusion prevention-type subsurface dams, surface water-groundwater storage-type subsurface dams, and deep subsurface dams of 70 m or more, and has evolved.

One of the most concerning issues regarding the environmental impact of subsurface dams is their effect on the NO₃-N concentration in groundwater. The impact of concern is the nitrogen concentration enrichment in groundwater due to two factors: stagnation of groundwater flow below the full reservoir level of the subsurface dam and repeated groundwater irrigation of fields in the subsurface dam basin. The results of an investigation using the TRAM in irrigation engineering confirmed that irrigation water below the TRAM does not affect groundwater quality. The results of an investigation using causal inference have statistically demonstrated that subsurface dams do not affect the increase in NO₃-N concentration. Subsurface dams further reduced the NO₃-N concentration trend because large amounts of groundwater pumped from the reservoir refreshed groundwater below the full water level.

There is a correlation of R² = 0.96 between the gross reservoir capacity and active capacity of large-scale subsurface dams constructed in the Ryukyu Arc. Therefore, once the gross reservoir capacity is determined, the active capacity and dead capacity can be estimated. In an irrigation project using a subsurface dam as the water source on Miyako Island, the construction cost of the subsurface dam was 57% of the total project cost. 11% of the construction cost was the cost for design and investigation. This high cost was due to drilling surveys used to draw the bedrock topography map. If highly accurate geophysical exploration is developed, the cost will decrease. The water price of subsurface dams in the Ryukyu Arc is the same as or slightly higher than that of surface dams.

Currently, the construction of subsurface dams is desired as a water source dam in arid regions or as a dam to prevent saltwater intrusion near the coast. These subsurface dams are gravel aquifer-type dams. Therefore, the technology for Ryukyu limestone described here may not be directly applicable to these. It should also be noted that the integrated water storage model cannot be applied to areas of conduit flow-type aquifers.

In the Ryukyu limestone region, if the impermeable ground in a valley-like topography is located at a limited depth for the application of the SMW method (about 100 m depth) [26], it seems possible to construct a subsurface dam. However, if the impermeable ground is deeper than this, the groundwater floats on top of the seawater as a freshwater lens due to density differences. No technology has been developed to thicken freshwater lenses in these areas (Tsuken Island [84], Tarama Island [85], Minamidaito Island [86], and Irabu Island). A floating-type subsurface dam (partially closed subsurface dam) that does not penetrate the foundation with a cut-off wall has been proposed [84,87,88], but has not yet been realized. Future development is expected.