Hydrological principle behind the development of series of bunds in ancient tank cascades in small catchments, Sri Lanka

The utilisation of water harvesting systems of ancient origin can be seen in many regions of the world even today. Iran, Syria, Morocco, Spain, Oman, and Jordan are some countries in the drylands of the Mediterranean and western Asia (Beckers et al. 2013). India (Palanisami 2006; Okhravi et al. 2014; Alaguraj et al. 2017), Sri Lanka (Zubair 2005; Bebermeier et al. 2017), Cambodia and China are some examples from Asia. Water supply systems that rely on locally generated intermittent sources are called autogenic, and they are classified into groundwater harvesting, floodwater harvesting and runoff harvesting (Beckers et al. 2013). Runoff harvesting techniques in micro catchments consist of small pits, contour bench terraces and check dams or earthen bunds. In general, water harvesting systems consist of four components: catchment, conveyance or deflection device, storage facility and the target of water supply (Beckers et al. 2013). Water harvesting is inherently related to hydrology, which is not a new area of endeavour, and it is as old as man's efforts at survival. Throughout the ages, the well-being of the civilisation has been, and will forever be, dependent on available and usable water (Carter et al. 1963). Runoff harvesting efforts have played a vital role in water resource management over millennia, and they have been based on the knowledge gained through historical and experimental events of hydrology along with the available technologies. The development of a large number of traditional tanks in many Asian countries, including Sri Lanka, Burma, Thailand, Cambodia, and India (Begum 1987), are such practices.

These traditional tank systems have helped in agricultural production in the dry zones of India, Sri Lanka and Southeast Asia for the past 2,000 years. Around 208,000 tanks exist in India (Nagarajan 2013), and around 30,000 such tanks, termed village or minor tanks, have been estimated to exist in Sri Lanka. However, a considerable number of such tanks remain unrestored in Sri Lanka, and only around 13,500 tanks are in working condition providing agricultural and other water needs of the community (DoAD 2012).

The importance of restoration of ancient tanks in an abandoned state has recently been considered as one of the adaptation options to face the impacts of climate change in Sri Lanka (MoMD&E 2016). Nevertheless, the prevailing selection criteria of tanks for restoration are based on the agricultural performance, and this restricts the selection of tanks to be restored. On the other hand, the experiences and observations on restored tanks during the last several decades also show that agriculture under these tank systems has not been very productive (Somasiri 2000) and agricultural performance and productivity are low compared to major irrigation systems (Department of Census and Statistics 2016). Besides, the occurrence of frequent water scarcities and breaching of around one thousand tanks in a single flood event (MIWR&DM 2018) are practical problems associated with these systems. However, they are considered as one of the most important elements in the dry zone landscape of the country. In 2018, they were declared as a Globally Important Agricultural Heritage Systems by the FAO of the United Nations (FAO 2018).

The incongruous behaviour of the tanks makes the understanding of the principles behind the development of these ancient systems vital for their incorporation in modern water resources management. Several authors have highlighted the necessity to identify development principles for tank systems. Kennedy (1933) proposed acquiring the sense and substance of individual village irrigation tanks. Maddumabandara (1985) and Panabokke (2009) have emphasised the same requirement for the tank cascades in Sri Lanka while highlighting the importance of these systems as the smallest hydrological unit for water resource management in river basin development. Tennakoon (2000) raised the need for identifying the sequence of tanks built along the rivulets and the maximum number of tanks in a cascade as research areas to be studied. Dharmasena (2004) stated that the most pronounced reasons for not achieving the anticipated impacts of rehabilitation and development of village tanks lie in the failure to take proper account of deep-rooted customs and traditional wisdom in planning rehabilitation and development.

With this background, this study attempts to find a solution to this practical problem by investigating the principles behind the development of a series of bunds in small catchments in the form of tank cascades by developing a hypothesis and testing it using the HEC-HMS model.

Sri Lanka is an island situated in the Indian Ocean between latitudes 5° 55′ and 9° 50′N and longitudes 79° 42′ and 81° 53′E. The total area of the country is around 65, 635 km². The island receives rainfall mainly from two monsoons, namely the Northeast and the Southwest. The country is divided into different climatic zones based on the mean annual rainfall. The mean annual rainfall in the dry zone is below 1,750 mm while that in the intermediate zone is between 1,750 mm and 2,500 mm. Due to the seasonality of rainfall, an alternating wet and dry climate prevails in the two zones where the tank densities are high.

The majority of these tanks are located in series called tank cascades (Maddumabandara 1985), and around 1,150 such tank cascades are found in the country (DoAD 2007). According to cascade classification, the cascades of an area less than 2,500 Ac (1,012 ha) are in the small category, and those between 2,500 and 5,000 Ac (1,012–2,023 ha) belong to the medium category (Panabokke 2000). The southeastern low country region, which has a high abandoned tank density, consists of 28 main watersheds and 15 sub-watersheds and the region was selected as the study area. Over 200 tank cascades have been identified in the area, and around 87% of the cascades are in the small and medium categories (Perera et al. 2020a). Hence, a medium cascade located in the area was selected as a specimen cascade for analysis. Figure 1 shows the tank distribution in the country and a close up of a bund setting along rivulets in the study area.

The specimen tank cascade consists of five abandoned tanks located along the upstream part of the Alakola ara rivulet in Kumbukkan Oya river basin. The location map of Kumbukkan Oya basin and the layout of the specimen tank cascade in the 1:50,000 topographical maps (Survey Department 2002) are shown in Figure 2.

The layout of the bund series identified in the 1: 50,000 topographical maps was further investigated using the STRM30 (NASA 2015) Digital Elevation Model (DEM) of 30 m resolution and satellite images available in Google Earth (Figure 3). The cascade parameters and catchment characteristics were determined using the tools available in Arc GIS 10.3.1 software. Flow accumulation paths, gross and net catchments at each bund were delineated using the software, which determines the contributing area above a set of cells in a raster map. The marginal shift appearing between overlaid bund series and the delineated flow path, especially at Tank C, can be due to limitations associated with the two different data sources.

Flow accumulation paths and five bunds in series denoted from A to E in the catchment area are shown in Figure 3. Elevation – storage relationships of the tanks were estimated based on the topography, water spread area of the tanks and empirical formula, V = 0.4AD which was proposed by Kennedy (1933), where V is the capacity in acre-feet of the tank at full supply depth D in feet, and A is the water spread area in acres at that depth. The linear approximation was used to estimate the tank storages at different elevations considering estimated capacities of the tanks and assuming that the area at zero depth is zero. Baker (1914) developed velocity, discharge and afflux diagrams for 100 ft (30 m approximately) long natural spillways of village tanks, which are adopted for village tanks in general.

The rainfalls of 2, 5, 10, 25 and 50 year return periods were determined using regional depth-duration-frequency relationships presented in ‘Rainfall depth – Duration – Frequency studies for Sri Lanka’ (Baghirathan & Shaw 1977), the oldest available for the country as the study is on the hydrological behaviour of the tank system in ancient times.

A uniform distribution of rainfall intensity was assumed for the entire catchment as the catchment area is small.

L is the longest river course of the catchment in miles, and L_ca is the length from the point of interest to the point on the river course closest to the centroid of the catchment in miles. C_t was taken as 1.17 (Ponrajah 1989) for the specimen catchment of the study. The catchment parameters, the tank parameters and storm characteristics used in the modelling of the bund series are shown in Table 1.

The methodology adopted to identify the development principle behind the formation of a series of bunds consists of two steps: development of hypothesis and testing of the hypothesis by illustrating the hydrological principle.

Development of the hypothesis on the system principle behind the cascades was carried out based on the historical evidence, definitions, terminologies and studies on the development and operation of the tank cascades. The first definition of the tank cascade is ‘a connected series of tanks organised within the micro catchments of the dry zone landscape, storing, conveying and utilising water from an ephemeral rivulet’ (Maddumabandara 1985). Later Panabokke suggested replacing the terms micro with meso and ephemeral rivulet with either second-order inland valley or first-order ephemeral stream (Sakthivadivel et al. 1994). The functions of the cascades stated in both definitions are similar and relate to the general functions of storage in a tank. As such, none of them has addressed the reason behind the development of tank series instead of a single tank, which is comparatively large. However, the indigenous term Ellangawa given for a cascade means a series of successive tanks where Ellan means ‘hanging’, and gawa means ‘one after another’ (Tennakoon 2000). It also implies the hydrologically dependent and interlinked nature of the functions of the tank series.

However, the cascade classifications available (Panabokke 2000) are mainly based on the geometry of the cascades, and the hydrological characteristics within the cascades have not been addressed. Mendis (2002) presented a concept on the existence of different types of water bodies associated with bunds such as wewa, wila, and vatiya, which gives an insight into the different functions of the bunds. In this concept, wewa means tank; wila means pond and vatiya means a small flood protection dyke. Accordingly, all bunds that exist in a series would not be tanks in terms of the functions as present-day terminology. The inherent characteristic of different degrees of water storage in the tanks with a higher temporal variation (Perera et al. 2020b) also reinforces this presumption.

Presently the water recycling or re-use concept of water management in tank cascades is considered as a functional principle of cascades, and it has been studied using a water balance simulation model to predict tank water availability (Jayatilaka et al. 2001). However, it has been revealed that there are large storage fluctuations and high dependency of tank storage behaviour on rainfall events. Also, there are some instances of reduction of inflow to downstream tanks due to restoration of tanks in the upstream (Kariyawasam et al. 1984). Further, upgrading the tanks in the upstream of cascades by introducing sluices and developing lands for cultivation under them have caused frequent water shortages (Tennakoon 2005). These facts prove that the system characteristics substantially govern the inflows to the tanks in addition to the depth of rainfalls that generate the catchment runoff.

On the other hand, there is historical and recent evidence of breaching of tank bunds during flood inflows. Ievers (1899) has recorded two events of breaching of a large number of tanks which provided water for agriculture in the Anuradhapura and Vanni areas during the cyclone in 1829 and breaching of 180 tanks in North Central Province in a single day due to floods in 1887. Kennedy (1933) has stated that the reason behind the first systematic study on village tanks was breaching of tanks near Madawachchiya in Anuradhapura District in 1923. Udawattage (1985) has noted the breaching of 78 tanks out of 488 tanks in Puttalam District during the high rainfall event that took place on 22 April 1984 (Figure 1). Recently, there was the breaching of 982 minor tanks and diversions in 2011 and 967 in 2014 during floods (MIWR&DM 2018). Accordingly, breaching of tank bunds during high-intensity rainfall events could be considered as an inherent risk associated with these tanks.

Limited spillway capacity is the primary cause for breaching of bunds and Baker (1914) reported that the most common form of spillway used for village tanks in the country was a channel cut in the ground on the right or left bank of the tank site, called the natural spill. The discharge capacity of this type of spillway is low, and it limits the evacuation of the excess water from the tank. As a result, the afflux builds up, and it causes breaching of tank bunds at times due to overtopping. Accordingly, the spillway technologies of ancient village tanks have functional limitations.

The sluice or the outlet is also an important structure of a tank that is used to take the stored water out from the tanks when required. Knox (1681) wrote of his experience on the methodology used by farmers to draw water to the paddy fields from the tanks by cutting open the tank bund at one end and then refilling the cut section later. The absence of sluices in village tanks was also mentioned by Abbay (1877). The same was confirmed by Ievers (1899) and indicated that there were no sluices in village tanks to draw the water from the tanks in earlier days. The limitation on sluice-technology must have influenced the construction of tanks with shallow bunds, which increases the risk of overtopping during floods.

With this background, a hypothesis on the principle behind the formation of ancient tank cascades was developed based on the following facts:

• Construction of tanks with shallow bunds was due to the absence of sluice technology during ancient times.

• High risk of breaching of bunds was associated with village tanks due to limited spillway technologies in the past and at present.

• Bunds are located closely along the rivulets.

• Experiences of reduction of inflows to the tanks in the downstream due to restoration or upgrading of tanks in the upstream of the series.

• High variability and uncertainty in year-round storage availability of the tanks.

• There are incidences of leaving upstream bunds unrestored when the tanks in the downstream of the series are upgraded or restored with adequate spillway capacities.

• Meaning of the indigenous term Ellangawa used to denote the tank cascade.

Accordingly, the following hypothesis was developed:

Hypothesis: Ancient water storages formed with a series of bunds in shallow valleys of a small catchment is a regulating technique developed over time from the experience of peak flows with available technologies.

HEC-HMS 4.2.1 (Hydrologic Engineering Center – Hydrologic Modeling System) software (USACE 2016) was used as a tool to simulate the hydrological behaviour of the specimen tank cascade taking into account its ability in the simulation of short-time events and suitability for the purpose. The software includes a number of infiltration models, unit hydrograph approaches, and hydrological routing. A model of the specimen tank cascade was developed incorporating the catchment and tank parameters to investigate the behaviour of the tanks during inflows generated from rainfall events of different return periods.

Figure 4 shows the hydrologic model developed using HEC-HMS 4.2.1 as a simplification of the tank cascade that aids in understanding its flood behaviour. The tank cascade was modelled incorporating five different scenarios of bund arrangements for the inflows generated from 2 year, 5 year, 10 year, 25 year and 50 year return period rainfalls. The scenarios tested with the model are given in Table 2.

The catchment is ungauged, and there are no field observations of overtopping events and discharges. Selecting a suitable model, its calculation options and improving the quantity and quality of hydrological data are essential if a model is to be fit for purpose (Beven 2019). Out of nine different loss methods provided in HEC-HMS, the Initial and Constant Loss Method was selected. The method requires fewer input parameters and is generally used in Sri Lanka. The Snyder unit hydrograph method was selected as the transform method for the catchments and lag method was selected as the routing method to model the streams between two bunds.

The model was run to check the overtopping condition of the bunds, which was considered as the failure of the system. Initially, the model was run for scenario 1, where only the downstream-most tank exists starting from inflows generated from 2 year return period rainfalls. When the bund was overtopped with the inflow generated from a particular rainfall of a given return period, the immediate upstream bund was added to the series. The model was rerun with rainfalls of different return periods until the system failure occurred due to overtopping of at least one bund. The test was continued with the addition of bunds to the model one by one sequentially from downstream towards the upstream. The model of specimen cascade was run for two initial conditions, namely, the tanks empty and the tanks full, to understand the hydrological behaviour of the bund series and their degree of survival during floods. Finally, the results were interpreted and used to verify the hypothesis while making suggestions to modify the existing definition of the tank cascade.

The inflow and outflow hydrographs of each tank obtained from the respective runs of HEC-HMS model for the five scenarios of the bund series show the behaviour of the tanks. When the pool elevation is equal to the bund top level, the spillway discharge reaches the maximum with which the excess flow cannot be discharged without overtopping the bund. This discharge is considered as the limiting spillway capacity of the tanks, which is around 40 m³/s for all tanks studied. When the discharge exceeds the limiting spillway capacity, the bund acts as an overtopping section in the model, and this situation is considered as a failure of the tank system. Accordingly, when at least one bund is overtopped, it is considered as a failure of the bund series and the lowest return period flood it occurred at was considered as the failure return period.

The model was first run with the initial condition of the tanks empty, and Figures 5–9 show the inflow and outflow hydrographs of the different scenarios during the failure return period floods.

The results of Scenario 1, which has only bund A in place at the most downstream of the series, shows that it could safely discharge the runoff generated from the catchment for 2-year return period rainfall (Figure 5). The bund was overtopped with the inflows generated from the 5-year return period rainfall. The estimated peak inflow of the catchment for the 50-year return period is around three times the inflow generated from 2-year return period rainfall.

As Bund A was overtopped with the inflow generated from the catchment for the 5-year return period rainfalls, Bund B was then introduced to the series while forming the first attempt at developing a tank cascade (Scenario 2).

Although both bunds A and B were not overtopped for inflows generated from 2-year return period rainfall, they were overtopped with the inflow generated from the rainfall of 5-year return period (Figure 6). The peak inflow at Bund A was attenuated while the inflow hydrograph consists of two peaks.

As both the bunds were overtopped with the inflow generated from the rainfall of 5-year return period, Bund C was introduced to the series (Scenario 3) as the third bund. The modelling results after the introduction of Bund C showed that the bund arrangement is safe for the flood generated from the rainfall of 5-year return period. The effects of routing were observed at bunds A and B due to existence in a series (Figure 7). However, overtopping of bund C during the 10-year return period flood indicates the risk associated with Bund A and B due to a sudden flash flow that can occur due to possible breaching of Bund C.

The model was then run with Bund D introduced to the series, increasing the number of bunds to four in the tank cascade (Scenario 4).

Addition of Bund D further improved the detention capacity of the bund series, and the modelling results show that the bund arrangement is safe for the flood generated from the rainfall of 10-year return period (Figure 8). However, Bund D was overtopped with the inflow generated from the rainfall of 25-year return period as the spillway capacity is not adequate to safely discharge the inflow generated by sub-basin D. Overtopping of Bund D indicates the risk associated with the entire cascade during the inflow generated by rainfall of 25-year return period.

Finally, Bund E, the most upstream bund, was introduced to the series, increasing the number of bunds to five in the tank cascade (Scenario 5).

The addition of Bund E to the series prevents the overtopping of Bund D while improving the robustness of the system during the inflows generated from the rainfall of 25-year return period (Figure 9). However, Bund A was overtopped with the inflow generated from this rainfall event. When Scenario 5 was checked for the 50-year return period rainfall, it was revealed that the four upstream bunds, except Bund A, are safe even for the inflows generated from the rainfall of this return period.

The modelling results clearly demonstrate the routing mechanism, which consists of peak attenuation and shifting of the time of the peak. The existence of a bund series reduces inflows into all but the uppermost tank.

The model was rerun for the same scenarios by changing the initial storage condition to the tanks at full supply level state, and the results are given in Table 3.

The results show that the initial storage state of the bund series governs the occurrence of overtopping of the bunds. The tank water levels being at full supply level prevents permanent detention, which reduces the regulating capacity while limiting the attenuation of peaks of inflows. Therefore, the series of small tanks at full supply level can withstand only the inflows of lesser return period rainfalls than when the tanks are in an empty state.

Accordingly, the capacity to discharge excess water from the tank is a crucial factor when harvesting the runoff of a small catchment with relatively small storage infrastructures in the form of tanks. Therefore, the limitations of the technologies in ancient times had made it necessary to find alternative solutions. The development of a series of bunds in small catchments could be such a solution in addition to the primary focus of storing water. For instance, the restored tanks experiencing frequent water scarcities may not have been developed as reliable water sources. Hence, the development of series of bunds in the form of tank cascades in small catchments would be the alternate solution to have a robust system to withstand flood inflows due to the limitations of the technologies that existed in the ancient times.

The study attempted to identify the principle behind the development of ancient tank cascades for water resource management in small catchments in Sri Lanka. A hypothesis was developed logically based on the facts presented in studies on the subject and historical evidence. A hydrological model was used to test the hypothesis for identifying the behaviour of a cascade, and the results verified the hypothesis by filling the gaps that existed in understanding the hydrological aspect of the tank cascades.

Runoff harvesting of small catchments must have been a challenge with shallow bunds, and limited spillway and sluice technologies that had been practised in these systems. Higher discharges generated from the catchment with the rainfalls of different return periods would have made the situation more aggravating. The building of more than one bund in a shallow valley divides the catchment into sub-catchments, and the series of bunds attenuate the peak inflows to the tanks. Attenuation of peak inflows prevents the overtopping of bunds even during floods of high return periods, avoiding the risk of breaching. The results prove that a strategically located series of bunds has a higher potential to withstand floods of a longer return period than a single bund.

Accordingly, a series of bunds must have been built by the ancestors as an autogenic runoff harvesting technique on a trial and error basis using the knowledge gathered from both successful and unsuccessful experiences of local flood events. Hence, the thousands of tank cascades in the country can be considered as time-tested systems developed as a replication of the principle to suit the climatic conditions prevailing in the dry and intermediate zones.

Therefore, the definition of the tank cascade is suggested to be modified as ‘village tank cascade is an autogenic runoff harvesting technique consisting of strategically located series of bunds in shallow valleys of a small catchment for storing water while increasing the ability to regulate peak flows experienced over time’. As such, the principle behind the development of tank cascades is generic while the hydrological behaviour of the cascades is unique, depending on the shape of the catchment, the number of bunds, the capacity of tanks, magnitudes of the flood inflows and the water storage requirement.

Hence, it is recommended to assess the potential of cascades to withstand floods of the design return period, adopting the principle behind the development of these systems with field observations of flood events when restoring them for present-day requirements. Further, the identification of a development principle would facilitate the restoration of existing cascades as water resource management systems with modifications such as relocating, resizing, inclusion, disregarding and elimination of tanks by taking a holistic approach. For instance, increasing tank capacities, upgrading spillways and outlets of potential tanks, disregarding or elimination of some tanks in cascades would be the more prudent option when rehabilitating these ancient systems in the modern era. Accordingly, the findings provide a solution for the problem on the restoration of ancient tank cascades in Sri Lanka and give insights into the crucial need for the identification of development principles of ancient hydrological systems in any region of the world in order to utilise them sustainably.

The research presented in this paper forms a part of the study carried out under the title of ‘restoration of ancient Ruhunu Rata tank systems to meet the future water demand and livelihood developments as an adaptable strategy to overcome the impacts of climate change’ with a research grant from the National Research Council, Sri Lanka (Grant no. 12–104). The authors would like to thank the anonymous reviewers for their valuable comments and suggestions.

All relevant data are included in the paper or its Supplementary Information.