Assessing the cost-effectiveness of irrigation water management practices in water stressed agricultural catchments: The case of Pinios

Agricultural Water Management 139 (2014) 31–42

Contents lists available at ScienceDirect

Agricultural Water Management journal homepage: www.elsevier.com/locate/agwat

Assessing the cost-effectiveness of irrigation water management practices in water stressed agricultural catchments: The case of Pinios Y. Panagopoulos a,∗ , C. Makropoulos a , A. Gkiokas b , M. Kossida a , L. Evangelou c , G. Lourmas d , S. Michas b , C. Tsadilas c , S. Papageorgiou d , V. Perleros e , S. Drakopoulou e , M. Mimikou a a Laboratory of Hydrology and Water Resources Management, Department of Water Resources and Environmental Engineering, School of Civil Engineering, National Technical University of Athens (NTUA), 5, Iroon Politechniou Street, 15780 Zografou, Athens, Greece b Hydroexigiantiki L.S. Lazarides & Co Consulting, 3 Evias Street, 15125 Maroussi, Athens, Greece c National Agricultural Research Foundation, Institute of Soil Mapping and Classification, 1 Theofrastos Street, 41335 Larissa, Greece d LDK Consultants Engineers & Planners SA, Off 21, Thivaidos Street, 14564 Kifissia, Athens, Greece e Consultant Hydrogeologists, Athens, Greece

a r t i c l e

i n f o

Article history: Received 6 October 2013 Accepted 18 March 2014 Keywords: Agricultural management Conveyance improvement Deficit irrigation Precision irrigation SWAT Water abstractions Water reuse

a b s t r a c t Agricultural water use in the Mediterranean region is significant, causing serious threats to water bodies that may not be able to reach the ‘good status’ required by the Water Framework Directive. This study uses the Soil and Water Assessment Tool (SWAT) model and a simple economic component developed in order to estimate the cost-effectiveness (CE) of six agricultural Best Management Practices (BMPs) in reducing irrigation water abstractions in the water scarce Pinios catchment in central Greece. Deficit irrigation, precision agriculture, waste water reuse, conveyance efficiency improvement and their combinations were evaluated and their CE was calculated for each Hydrologic Response Unit (HRU) separately and for the entire catchment. The results at the HRU scale are presented comprehensively on a map, demonstrating the spatial differentiation of CE ratios across the catchment. Based on the analysis, a catchment management solution of affordable total cost would include waste water reuse in areas adjacent to treatment installations, deficit irrigation in the least water deficient areas as well as precision agriculture in the most deficient ones. Conveyance losses reduction through the construction of piped irrigation networks was necessary for considerably decreasing groundwater overexploitation. However, since conveyance loss reduction entails significant costs, the resulting CE is not favorable. The methodology presented aims to facilitate decision making for agricultural water management by enabling modelers to combine processbased hydrological models with rapid and reliable cost estimations and use cost effectiveness metrics to identify and prioritize suitable irrigation water management practices. However, existing SWAT limitations are also discussed and the need for improving the accuracy of the representation of such practices in the future is highlighted. © 2014 Elsevier B.V. All rights reserved.

1. Introduction Agriculture is a significant water user in the European Union (EU), reaching up to 80% of the total abstractions in parts of the Mediterannean region (Wriedt et al., 2009). Especially in Greece, irrigation of crops accounts for virtually all agricultural water use, which in some cases has reached unsustainable levels (EEA, 2012). Sustainable water exploitation of aquatic systems is, inter alia, a prerequisite for achieving ‘good status’ for surface and

∗ Corresponding author. Tel.: ++30 2107722418; fax: +30 2107722879. E-mail address: [email protected] (Y. Panagopoulos). http://dx.doi.org/10.1016/j.agwat.2014.03.010 0378-3774/© 2014 Elsevier B.V. All rights reserved.

groundwater as required by the Water Framework Directive (WFD) (Directive, 2000/60/EC). Towards this end, EU Member States are required to establish a cost-effective Program of Measures (PoMs) in each river basin appropriate to water quantity and quality pressures identified in the River Basin Management Plans (RBMPs). Parts of these measures referring to the agricultural sector are known as ‘Best Management Practices’ (BMPs) (Cherry et al., 2008). For the estimation of BMPs’ cost-effectiveness (CE) in reaching multiple environmental targets at the river basin scale, highly sophisticated models are widely used. Such models are considered irreplaceable in terms of making quick water quality predictions under agricultural management changes due to their relatively fast simulation runs and increased number of watch points. Between

32

Y. Panagopoulos et al. / Agricultural Water Management 139 (2014) 31–42

several existing models, the Soil and Water Assessment Tool (SWAT) is a prominent process-based model, which is considered a robust, interdisciplinary tool appropriate for agricultural management simulation (Gassman et al., 2007; Neitsch et al., 2009). Recent examples of water management SWAT studies focus on estimating crop water productivity under alternative policy scenarios and water scarcity conditions (Faramarzi et al., 2010; Feng and Baoguo, 2010). Santhi et al. (2005) used SWAT to investigate the water savings in irrigation projects in Texas, USA, by examining crop change scenarios and irrigation water conveyance efficiency improvement. Masih et al. (2011) on the other hand, have used SWAT to investigate the impact of agricultural changes on basin streamflow in Iran. An interesting point in that study was the effort to simulate various in situ water harvesting systems and conservation techniques such as: micro-basins, terracing, bunds and mulching, with the purpose to increase soil water retention and foster plant water availability. Such interventions were indirectly represented in SWAT by increasing the Available Water Capacity (AWC) of soils. There are also studies where the SWAT code and the existing parameters could not adequately simulate specific irrigation systems. A representative recent example is the study of Xie and Cui (2011), who improved the irrigation function of ponds and reservoirs in order to reliably simulate irrigation in paddy rice fields of China, while another is this of Dechmi et al. (2012), who adjusted the model’s code to appropriately simulate irrigation water returns and total streamflow in intensively irrigated areas from outside sources. Despite those and possibly a few additional SWAT studies on agricultural water use, the water management module of SWAT has not been systematically examined for a range of practices and interventions, contrary to large sets of pollution mitigation practices, whose simulation has been adequately reported in recent studies (e.g. Arabi et al., 2008; Panagopoulos et al., 2011, 2012). Therefore, one of the major purposes of this paper is to provide guidance on how a wide range of agricultural water management BMPs could be represented at the river basin scale, taking advantage of model improvements incorporated in the most recent version of SWAT at the time of this research (ArcSWAT 2009 Version 93.7b), but also taking into consideration model representation inaccuracies that still exist. Moreover, the study contributes to the reporting on CE analysis with water management BMPs implementation, a topic of significant interest for water deficient southern European countries (Estrela and Vargas, 2012), on which only few peerreviewed studies have been published (Berbel et al., 2011). The model and approach are tested in the largest irrigated area in Greece, the water scarce Pinios catchment, which has been selected for conducting the EU funded project ‘i-adapt’ (www.i-adapt.gr), the Greek pilot project on the development of prevention activities to halt desertification in Europe.

2. Methodology 2.1. SWAT model description A catchment in the GIS-based SWAT environment is divided into subbasins and subsequently into Hydrologic Response Units (HRUs), which represent the different combinations of land use and soil types. The processes associated with water and sediment movement, crop growth and nutrient cycling are modeled at the HRU scale. Hydrological processes include surface runoff/infiltration, evapotranspiration, lateral flow, percolation, and return flow. The model considers a shallow unconfined aquifer, which contributes to the return flow and a deep confined aquifer acting as a source or sink. Agricultural management practices, such as planting, harvesting, tillage, irrigation, grazing and

nutrient applications can be simulated with specific dates. Irrigation and fertilization can be additionally applied automatically according to crop water and nutrient stress. The crop growth component of SWAT is a simplified version of the Erosion Productivity Impact Calculator (EPIC) model (Williams, 1995), which is capable of simulating a wide range of crop rotation, grassland/pasture systems, and trees. In the SWAT model, potential crop growth and yield are usually not achieved as they are inhibited by temperature, water and nutrient stress factors. As far as the irrigation routine is concerned, SWAT uses as potential irrigation sources: (a) the river, (b) a reservoir, (c) the shallow aquifer, (d) the deep aquifer or (e) an unlimited source outside the catchment. Compared to the older versions, SWAT 2009, used in this study, promises to represent more accurately the reality as the excess irrigation water applied to the HRUs does not return to the source, while new parameters have been added, with the purpose to simulate irrigation inefficiencies during water transport and application to the HRUs (Neitsch et al., 2009). 2.2. Study area The study is focusing at the Pinios basin (∼10,600 km2 ), covering almost entirely the River Basin District (RBD) of Thessaly in central Greece as shown in Fig. 1. The mean annual river flow at the outlet is reported close to 80 m3 /s and the mean annual precipitation is approximately 700 mm (Evangelou et al., 2011). The catchment is the most important agricultural producer in Greece, with fertile soils but drought climate during summer (Vasiliades et al., 2011). These conditions inversely affect both the natural vegetation and the agriculture of the region resulting in irrigation cutbacks, overexploitation of groundwater and significant losses of crop yields (Evangelou et al., 2011). Agriculture represents the 90–95% of the annual water demand of the area, with irrigated land (200,000 ha) covering half of the total cultivated area (400,000 ha). Cotton is the main crop cultivated (∼150,000 ha), followed by much smaller areas of corn and alfalfa. Irrigation water is abstracted mostly from groundwater sources; however, overexploitation has been a common practice for years leading to ever lower groundwater levels, making water more expensive to extract (deep pumping) and enhancing saline water intrusion in coastal areas (Evangelou et al., 2011; Loukas et al., 2007). It has also been estimated that the collective irrigation networks which are comprised of open canals in the study area, are characterized by significant conveyance losses and sub-optimal management of irrigation water (Mouratiadou and Moran, 2007). 2.3. Model setup and calibration The modeling study is comprised of two different simulation periods: (a) the historical period of 1975–1994 with available river flows for calibrating the model and (b) the future period 2011–2027, used as the baseline for testing irrigation BMPs. For the second period, the weather generator of SWAT (Neitsch et al., 2009) was used to produce meteorological time-series based on historical statistical measures derived from the existing stations. A crucial difference in water management between these two periods is that within the past (calibration) period crop areas were less, while irrigation water requirements could be totally satisfied by deep pumping and aquifer overexploitation. On the other hand, in nowadays and due to extremely high groundwater depths, deep pumping is not a viable practice and has been drastically reduced, while irrigated areas have been almost doubled (Makropoulos and Mimikou, 2012). The Pinios SWAT model setup was initiated with the use of a 25 m × 25 m DEM to delineate the study area (10,599 km2 ) and the river network with the catchment being divided in 49 subbasins

Y. Panagopoulos et al. / Agricultural Water Management 139 (2014) 31–42

33

Fig. 1. The Pinios river basin and its location within Greece.

Fig. 2. Subbasins (1–49), rain gauge and temperature stations, streams and reservoirs in the Pinios basin. Irrigated areas (colored) are depicted along with the irrigation source for the baseline scenario 2011–2027. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

34

Y. Panagopoulos et al. / Agricultural Water Management 139 (2014) 31–42

as shown in Fig. 2. A land use map was produced by combining a CORINE Land Cover (CLC) 2000 layer (EEA-ETC/TE, 2002) with Farm Structure Survey (FSS) data referring to the year 2000, providing the areas of each crop per municipality (Eurostat, 2000). On the other hand, the soil map of the catchment was produced by analyzing data layers provided by the Institute of Soil Mapping and Classification (ISMC) (http://www.ismc.gr/) including mainly clay and clay-loamy soils. The overlay of those maps followed by a threshold of 2% for land use classes and a threshold of 5% for soil classes led to the schematization of 361 HRUs (average area: 1 HRU = 30 km2 ), which is more attributed to land use heterogeneity. The irrigated crops included in the model were cotton (80%), alfalfa and corn. 2.3.1. Surface and groundwater exploitation The schematization of the study area as shown in Fig. 2, included subbasins 47, 48 and 49 in the eastern part, which form the lake ‘Karla’ basin (∼1100 km2 ). ‘Karla’ is a very recently constructed reservoir in the location of a natural impoundment, which existed many decades ago (Loukas et al., 2007). It was represented by a reservoir at the outlet of subbasin 47, with water inflows mainly arising from Pinios water diversions with the use of the recently constructed ‘Girtoni’ reservoir located along the river (see Fig. 2). These water transfers were simulated with the use of a point source flow in subbasin 47. At the same time, equivalent water quantities were subtracted from ‘Girtoni’ reservoir to an outside source. Other reservoirs that irrigate parts of the basin are ‘Smokovo’, which is operational since 2003 and the old ‘Plastiras’ reservoir, located outside the catchment (Fig. 2). Approximately 40,000 ha will be irrigated from these 4 reservoirs in the next years, a percentage of 20% of the total irrigated land. Groundwater will serve the rest 80% of the irrigated land in the catchment. To be able to model aquifer (over)exploitation, abstraction was conceptualized as consisting of two elements: abstraction from “renewable groundwater” that is defined as the volume that has enriched the aquifer within the period of analysis (and is assumed to be in low depths and easily accessible) or abstraction from “permanent” water reserves stored in the aquifer (assumed to be in larger depths). An HRU in SWAT can use as groundwater source for irrigation either the shallow or the deep aquifer and such a division could serve as a convenient separation of the renewable from the permanent groundwater reserves exploitation occurring in the basin. However, SWAT does not allow an HRU to be irrigated from both these sources at the same time. Hence, in order to simulate both sustainable and unsustainable groundwater abstractions in this study the exploitation of permanent reserves was also linked to the use of the shallow aquifer. Moreover, another limitation of the groundwater module of SWAT is that each subbasin has its own aquifer, which is considered as a single entity and is not extended to wider geographical areas than the subbasin’s boundaries. For a subbasin mostly or totally covered by irrigated crops it becomes obvious that a typical seasonal irrigation amount of 500–600 mm can only be abstracted from the aquifer when there is adequate water in it at the beginning of the simulation. Especially for arid or semi-arid regions such as the Pinios catchment, it is unlikely that natural groundwater recharge occurring within the wet period of the year is adequate to totally satisfy crop water requirements within the following dry period. Therefore, among the important parameters for simulating the degree of groundwater availability and exploitation in SWAT, is the initial groundwater content in each HRU defined as the initial depth of water (mm) in the shallow aquifer (parameter SHALLST in ‘gw’ files). This parameter can govern groundwater availability and thus, the actual irrigation amounts applied to the HRUs throughout the simulation period with large values ensuring that irrigation needs can be totally satisfied even for long simulation

periods. Therefore, a large value of the parameter can mimic the exploitation of permanent groundwater reserves (deep pumping), a practice which was the case for years in the study area. On the other hand, when groundwater is not considered to be an endless source, initial depths should not be set at very high levels and in the particular case of Pinios, where groundwater availability across parts of the catchment is now limited and pumping from very high depths is prohibited, a reduced irrigation amount from the theoretical one had to be applied to the HRUs to ensure a realistic simulation. Theoretical irrigation refers to the optimum irrigation amount defined by the user in the ‘mgt’ files of the SWAT project including subsequent losses (if any) during the application, and is the maximum amount that can be abstracted from the source given water availability. 2.3.2. Agricultural management practices For the optimum irrigation of cotton within a full growth cycle, 570 mm of water have to be abstracted from the source and applied from May to early September. The crop is fertilized with 195 kg/ha N and 31 kg/ha P and is harvested in October. For corn the total water amount is 620 mm and the crop is fertilized with 230 kg N/ha and 44 kg P/ha. Alfalfa is growing for a 3-year period and for its optimum irrigation 740 mm of water have to be annually abstracted. The crop is harvested 3 times per year and is fertilized only with P (46 kg P/ha/y). Irrigation is provided to crops with 9–12 doses with irrigation intervals of ∼10 days. However, a 25% of the abstracted water for irrigation in Pinios was assumed to always be lost because of significant conveyance inefficiencies. Details on how this level was estimated to represent the baseline conditions are provided later in the paper. The optimum water abstractions for irrigation are not expected to be guaranteed all around the catchment during the baseline period 2011–2027 due to the limited availability on the source (shallow aquifer). In contrast, full irrigation was the case for a long period in the past due to over-abstractions from the aquifer. The representation of both these different situations in the study area was necessary and is presented next. 2.3.3. Model calibration The developed model with the catchment schematization presented above was modified appropriately in order to reflect the water management regime existing during the historical period 1975–1994 with available measured river flows. This period was reproduced by irrigating 40% less agricultural land, proportionally across all subbasins, by ignoring the operation of the 3 recently constructed reservoirs located within the catchment and by considering groundwater as an endless source assigning large SHALLST values (5000 mm), able to mimic continuous groundwater (over)exploitation and thus ensure full irrigation throughout the 20-year calibration period. Manual adjustments on the hydrologic SWAT parameters were firstly made followed by model executions in order to set annual model predictions within reasonable water balance estimates (flows, evapotranspiration, and aquifer recharge). Especially for groundwater, the model was prompted to calculate net recharge rates similar to the long-term annual availability of renewable groundwater resources known for the catchment. The magnitude of those resources has been estimated for the study area according to meteorological and hydrogeological factors and ranged between 50 and 350 mm of water, with high values corresponding only to the western part of the catchment. Then, the model was accurately calibrated according to the observed monthly flows in ‘Amigdalia’, a river site with flow observations (Nalbantis and Koutsoyiannis, 1997), which coincides with the outlet of subbasin 33 (see Fig. 2) and drains an area of 6500 km2 . Apart from the absence of more recent observed flows for use in calibration, there was a lack of observations for the part of the catchment downstream the ‘Amigdalia’ station. However, land use and soil

Y. Panagopoulos et al. / Agricultural Water Management 139 (2014) 31–42

Fig. 3. Simulated vs observed river flows in Amigdalia station (subbasin 33 – Fig. 2).

homogeneity across the whole catchment allowed the adoption of the calibrated parameters for those areas as well. The most sensitive parameters along with their calibrated values were identified by executing a sensitivity analysis and auto-calibration with the Sequential Uncertainty Fitting (SUFI2) algorithm (Abbaspour et al., 2007), which is linked to SWAT through the calibration package SWAT-CUP4 (Abbaspour, 2011; http://www.neprashtechnology.ca/Default.aspx). In SUFI2, the degree to which all uncertainties are accounted for in the simulation is quantified by a measure referred to as the P-factor, which is the percentage of measured data bracketed by the 95% prediction uncertainty (95PPU). This is calculated at the 2.5% and 97.5% levels of the cumulative distribution of an output variable obtained through Latin hypercube sampling, disallowing 5% of the very bad simulations (Abbaspour, 2011). The maximum value for the P-factor is 100%, and ideally all measured data should be bracketed. The second measure quantifying the strength of the calibration/uncertainty analysis is the R-factor, which is the average thickness of the 95PPU band divided by the standard deviation of the measured data. The R-factor represents the width of the uncertainty interval, which should be as small as possible (smaller than a practical value of 1). The calibration of the Pinios model with SUFI2 was conducted by using the observed monthly flows of the period 1975–1984. The optimization indicated a P-factor of 0.64 and an R-factor of 0.62. Moreover, the R2 of the comparison was 0.89 and the Nash–Sutcliffe Efficiency (NSE) (Nash and Sutcliffe, 1970) 0.85. The calibrated parameter values are summarized in Table 1. As shown in Fig. 3, simulated monthly river flows had a very good convergence with the observed ones. The most recent 10year period (1985–1994) of available flows was used for validation, with NSE and R2 calculated as 0.71 and 0.72 respectively, which can be judged as satisfactory on a monthly basis (Moriasi et al., 2007). Model predictions were also compared with observations in several nested gauging stations (Nalbantis and Koutsoyiannis, 1997), giving additional evidence for the successful simulations. Another metric of calibration in this study was also the simulated crop harvest yields. The results were produced by slightly adjusting the optimum temperature, the harvest index and the total heat units of the crop to reach maturity and were compared with reported values by ISMC. The annual average harvest yield of cotton was 3.7 t/ha, while those of corn and alfalfa were estimated as 11.3 and 10.7 t/ha respectively.

35

analysis study. We executed the model for the baseline period 2011–2027 both with the SHALLST values at the natural recharge levels (50–350 mm) and with the increased by 80 mm values (130–430 mm). In the second case, an increased irrigation water abstraction occurred on a mean annual basis at the catchment scale, expressing the current overexploitation of water resources in the catchment. Accordingly, the Gwqmn and Revapmn had to receive such values, so as to be in agreement with the calibration findings of Table 1. Therefore, their values were set 75 and 10 mm lower respectively from the baseline SHALLST values (130–430 mm) in the subbasins. For the entire baseline period until the critical year 2027, the model estimated a mean annual amount of abstracted water for agriculture from all sources equal to 920 hm3 (1 hm3 = 106 m3 ). The amount that corresponds to the sustainable water resources exploitation (reservoirs water and groundwater that had entered the aquifer within the simulation period) is 700 hm3 . The difference of 220 hm3 indicates the existing 30% overexploitation of water for irrigation purposes in the entire catchment, attributed exclusively to groundwater, with the highest overexploitation occurring in the most water deficient areas. By analyzing the simulated results, it was also concluded that the ratio: water applied to HRUs/theoretical crop needs fell even below 0.5 in specific areas, in contrast to others, where irrigation needs were totally covered by the existing water resources, even without any overexploitation of groundwater. These areas included a large portion of the most western part of the catchment, as well as all areas irrigated by reservoirs (see Fig. 2). All the above estimates are close to the estimations of the true annual total abstractions, over-abstractions and water shortages across the study area, provided by local stakeholders, farmers and experts (Makropoulos and Mimikou, 2012; www.i-adapt.gr). Finally, compared to the average annual water abstraction of the historical period 1975–1994, the current abstractions are by 200 hm3 higher. The simulated abstracted water was 720 hm3 /y then and was adequate to cover entirely crop requirements, as already described. However, water use in this period should not be considered as sustainable due to its closeness to the sustainable amount of 700 hm3 /y of the future period (2011–2027) when SHALLST values of 50–350 mm were used. During the historical period, the three reservoirs within the catchment (Fig. 2) that serve as irrigation sources in the baseline were not present and the percentage contribution of groundwater to total water abstractions was even higher. Groundwater was an endless source though; overexploitation was possible all across the basin and could cover entirely the existing needs. Compared to yields of the calibration period, a 10% decreased yield per unit area was simulated at the baseline, attributed to the inefficient irrigation of large areas in the catchment due to the consideration of groundwater as a noendless source. However, increased irrigated areas result in higher total production in the catchment compared to the past and this is actually the major driving force of water shortages and overexploitation in the study area today, although new reservoirs have been constructed and operate to better manage water resources. 2.4. Irrigation BMPs formulation and costing

2.3.4. Baseline simulation The model was executed for the baseline period of 2011–2027, by increasing irrigated land to the up-to-date levels and by defining the current water transfer rules occurring in the area concerning reservoirs’ operation (see Section 2.3.1 and Fig. 2). Most importantly, we made appropriate modifications to rationally simulate groundwater availability. For the SHALLST values of the subbasins, which represent groundwater availability at the beginning of the baseline simulation (2011), we used the known longterm annual groundwater recharge levels (50–350 mm), increased by 80 mm. This level of increase was defined after a sensitivity

Four irrigation water management BMPs and 2 combinations were tested with SWAT. The unique measures are: deficit irrigation (DI), reduction of water conveyance losses or conveyance improvement (CI), precision agriculture (PA) and waste water reuse (WWR) and the combined ones consisted of the simultaneous application of DI with CI and PA with CI. The costs of all practices were calculated based on recent unit costs/prices. Specifically, the additional total cost (direct and indirect) needed on an annual basis to implement a practice compared to the baseline was estimated, taking into consideration capital and maintenance costs for a period equal to

36

Y. Panagopoulos et al. / Agricultural Water Management 139 (2014) 31–42

Table 1 Calibrated parameters by SUFI2 in Amigdalia flow station (sub. 33 in Fig. 2). ID

Parameter

Initial value

1

Cn2

2 3 4

Epco Esco Gw Delay

5

Gw Revap

60–85 depending on the land use and soil type 1 0.95 5–50 days depending on the geological formations in each subbasin 0.05

6

Gwqmn

5000 mm

−1000

1000

7

Revapmn

5000 mm

−1000

1000

8

Rchrg Dp

0

9

Sol Awc

0.05–0.20 depending on the geological formations in each subbasin 0.10–0.20 mm water/mm soil for each layer depending on the soil type 10–100 mm/h for each layer depending on the soil type

10

Sol K

Lower bound

Upper bound

Type of change in the auto-calibration

Final change (SUFI2) −12

−25

25

Percentage change

0 0 −100

1 1 100

Value selection Value selection Percentage change

0.44 0.21 10.3

−0.012

1

Add value to the initial value Add value to the initial value Add value to the initial value Percentage change

−62

−25

25

Percentage change

−17

−25

25

Percentage change

−12.4

−0.036

0.036

−75 −10

CN2, curve number; Epco, plant uptake compensation factor; Esco, soil evaporation compensation factor; GW delay, groundwater delay; Gw revap, water revap coefficient; Gwqmn, threshold depth of water in the shallow aquifer for baseflow to occur; Revapmn, threshold depth of water in the shallow aquifer for revap to occur; Rchrg dp, fraction of water in the shallow aquifer that becomes deep aquifer recharge (lost from the system); Sol Awc, available water capacity of the soil layer; Sol K, saturated hydraulic conductivity of the soil layer. Initial values of Gwqmn and Revapmn were modified accordingly with the selected initial depth of water in the shallow aquifer (SHALLST = 5000 mm) so that baseflow and water revap processes after calibration could be realistically simulated based on the fluctuations of groundwater depth.

the life time of each BMP, as well as annual gain/loss of incomes arising by their implementation. All information is shown in Table 2 and analyzed next. 2.4.1. BMPs description and representation in SWAT Net irrigation doses reaching the HRUs of Pinios are reduced due to conveyance losses. A detailed study has estimated those losses occurring within the collective irrigation networks in Pinios showing that approximately 40–48% of the provided water is lost when conveyed in open canal systems (Michas and Gkiokas, 2012). However, conveyance losses in irrigated areas served by private works (boreholes) are much less than losses in areas belonging to collective irrigation networks. As Pinios catchment in SWAT is divided into large HRUs, comprised of various polygons within each subbasin, they may correspond to irrigated areas of the real world served both by collective networks and private boreholes. Hence, in order to represent the ‘average conditions’ at the large HRU level of this study conveyance losses were reduced by half from their reported values homogeneously within the study area, becoming 25% of the abstracted water. Normally, such losses are suggested to be simulated in SWAT with the use of the IRR EFF parameter of the management (‘mgt’) files. The parameter should be set at a value less than 1.0 in order to simulate the fraction of the abstracted water reaching the irrigated HRU. In practice however, IRR EFF was found to be inactive when irrigation is applied manually (not by using the auto-irrigation routine) due to a bug in the model’s code. A different approach to simulate losses in the baseline (BMP1 in Table 2) would be to use the parameter IRR SQ in the ‘mgt’ files, which defines water losses during the application to the field (Neitsch et al., 2009). Although not precisely representing the kind of water losses we are looking for, the parameter works properly so that when it is set at 0.25, water that reaches the field is equal to 75% of the total amount abstracted, with the rest 25% becoming water lost via surface runoff from the HRUs. Another drawback that should be mentioned here is that even when the IRR EFF parameter is working (auto-irrigation), it does

not account for the amount of water that returns to the soil and the shallow aquifer via infiltration through the bed of the open canals. In contrast, conveyance losses with the use of the IRR EFF parameter in SWAT are considered complete losses from the system. What is actually occurring in the real world is that canal losses are only partly lost from the system through direct evaporation from the canals’ water surface or evapotranspiration from the vegetation growing into them. Another part of these losses called ‘canal seepage’ certainly enriches the aquifer but unfortunately, the parameters which are currently available to simulate the process (IRR EFF, IRR SQ) either remove completely the water from the system or make it runoff losses to downstream areas. Conveyance losses in both cases are not exploitable any more from the subbasins’ aquifers. Although in both cases conveyance efficiency simulation can give an accurate estimation of the net water amount directly applied to the field by a single irrigation operation, it leads to overestimations of the total groundwater abstractions needed or surface runoff produced within the multi-year simulation period. DI has been widely investigated as a valuable and sustainable production strategy in dry regions (Geerts and Raes, 2009) and has been widely tested with the use of crop growth models (Kloss et al., 2012). It represents the 2nd BMP in Table 2 and was applied uniformly throughout the crop growth cycle, with all irrigation doses being reduced by 30%, although applied at the same intervals as in the baseline. It should be noted here that this practice does not always result in a net 30% abstraction reduction. This level of reduction occurs only on those HRUs with significant water availability. In areas with no adequate water on the source, the level of reduction is inevitably smaller. The practice results in more water stress days for the crop but also in reduced runoff losses from the HRU as the reduced soil moisture contents cause lower CN values, which are updated daily by the model. The reduction of deep percolation losses, which is definitely an impact of DI on water cycle is not well simulated in SWAT as mentioned later; however, even the reduction of surface runoff losses alone constitutes this measure a good water management practice in this study, able to

Y. Panagopoulos et al. / Agricultural Water Management 139 (2014) 31–42

37

Table 2 List of BMPs selected for the analysis of the Pinios river basin. BMP

Where implemented

BMP description

BMP rationale

HOW implemented in SWAT

COSTa Additional annual cost (+) or income (−)

1

All crops (corn, cotton, alfalfa)

Conventional practices

Producing the maximum yield

No additional COST or income

All crops

30% irrigation water reduction

Saving water with a reduction in yields

50 mm dose in corn (12 doses) and cotton (11 doses), 80 mm in alfalfa (9 doses) IRR SQ = 0.25 IRR EFF = 1.0 35 mm dose in corn and cotton, 55 in alfalfa IRR SQ = 0.25 IRR EFF = 1.0

All crops

No losses during water transportation

Increasing net water applied to fields

All crops

Irrigation reduction and no transportation losses

2&3

Cotton but not in HRUs irrigated by reservoirs (subs 19, 47, 48)

Auto-irrigation and fertilization

Water and fertilizers at optimum timings Sustain yields, reduce inputs

Cotton but not in HRUs irrigated by reservoirs (subs 19, 47, 48)

3&5

3&5

In cotton HRUs of subbasins 9 and 34

No water pumping in these HRUs

Save groundwater Maximize the irrigation water amounts applied

Baseline

2 Deficit irrigation (DI)

3 Conveyance improvement (CI)

4 (2 & 3)

5 Precision agriculture (PA)

6 3&5

7 Waste water reuse (WWR)

50 mm dose in corn and cotton, 80 mm in alfalfa IRR SQ = 0 IRR EFF = 1.0 35 mm dose in corn and cotton, 55 mm in alfalfa IRR SQ = 0 IRR EFF = 1.0 Auto wstrs = 100 mm Irr mx = 50 mm IRR SQ = 0 IRR EFF = 0.75 Auto Nstrs = 0.95 Auto Napp = 20 kg/ha Auto wstrs = 100 mm Irr mx = 50 mm IRR SQ = 0 IRR EFF = 1.0 Auto Nstrs = 0.95 Auto Napp = 20 kg/ha Changing the source of irrigation from groundwater (ID = 3) to outside source (ID = 5) IRR SQ = 0 IRR EFF = 1.0 42 mm dose in cotton

+0.225 D /kg corn yield lost +0.540 D /kg cotton yield lost +0.210 D /kg alfalfa yield lost The same as in BMP 2 for yields and +400 D /ha implementation cost 2&3

±0.540 D /kg cotton yield lost/gained +200 D /ha (equipment) −1.3 D /kg N in fertilization 3&5

+100 D /ha implementation cost −0.540 D /kg cotton yield gained

a Positive values in the last column indicate additional annual equivalent cost (AEC) of BMP implementation compared to the conventional practice. Negative values indicate additional farmers’ income. Auto wstrs is the threshold of soil water deficit below field capacity that triggers irrigation, Irr mx is the dose applied when irrigation is triggered, Auto Nstrs is the threshold of daily plant growth that triggers N fertilization and Auto Napp is the N dose when N fertilization is triggered. The parameters IRR SQ and IRR EFF represent fractions of water losses at the application of water to the field and the conveyance of water from the source to the field respectively. Due to some limitations in the model’s code to always represent irrigation efficiency correctly; their usage was determined differently for simulating the various BMP options listed above. For example the 25% irrigation losses were simulated with the parameter IRR SQ (0.25), while when auto-irrigation was applied conveyance losses were directly simulated with the parameter IRR EFF (0.75). Even with the indirect representations done in some cases, the BMPs were simulated successfully. Pumping cost (not included in the last column) was also incorporated in the calculation of the total additional cost of each BMP for the HRUs which are irrigated by groundwater sources. This cost, which always corresponds to additional revenue, was calculated on the basis of the pumped water saved from the baseline with the implementation of each BMP using 0.01 and 0.02 D /m3 as unit values depending on the magnitude of water saved from the aquifer (renewable (low depths) or permanent (high depths) water reserves respectively).

represent, at least to some extent, the real effect of DI on water abstractions. On the other hand, CI was assumed to occur by the installation of piped networks. The practice is represented by BMP3 in Table 2 and was simulated by nullifying losses through the modification of the IRR SQ parameter. For simulating it, a very optimistic scenario was actually selected, with losses through underground piped systems being practically negligible. Thus, the parameter was set to zero with all the abstracted water being applied to the HRUs. DI with CI is a combination of the two previous single BMPs and represents the BMP No. 4 in Table 2. The practice leads to less net water abstracted from the source and applied to the field due to the 30% decrease; however, with the simultaneous application of CI there are no losses. Therefore, the total net water applied to the field is higher than the water applied when DI is implemented alone, leading to less water stress days for the crops.

The general concept of PA is that both irrigation water and fertilizers are provided to the crop at optimum timings and doses with the purpose to sustain or even increase yields. Such cultivation techniques had already been initiated in cotton areas of the Pinios basin (www.hydrosense.org) and their testing was continued by ISMC within the ‘i-adapt’ project period. Specifically, PA was applied in 3 cotton fields located close to the city of Larisa (see Fig. 2) and the results revealed that water use was reduced by 5–35%, without practically affecting yields (Evangelou and Tsadilas, 2012). Based on the experience gained, an efficient implementation of the PA technology is to basically use soil moisture sensors to measure soil water content and determine irrigation timing. For simulating PA in such a way, SWAT can use a critical soil moisture point, below which, irrigation is automatically applied. For the cotton areas of Pinios, auto-irrigation was simulated by defining a threshold of 100 mm soil water deficit (below field capacity) and a maximum dose of 50 mm (see Table 2). On the other hand, in

38

Y. Panagopoulos et al. / Agricultural Water Management 139 (2014) 31–42

auto-fertilization, nitrogen (N) is applied when the plant growth falls below a crop growth threshold and in this study a threshold of 0.95 was selected, which corresponds to a crop growth equal to the 95% of the optimum in each day of simulation, along with a maximum N dose of 20 kg/ha. When auto-irrigation simulation is applied in SWAT, the IRR EFF parameter is active and the 25% conveyance losses can be simulated by avoiding the indirect use of the IRR SQ parameter (BMP5 in Table 2). It is reminded however, that conveyance losses never return to the hydrologic system. Moreover, in order to simulate CI together with PA (BMP6 in Table 2) IRR EFF was simply modified to 1.0 from 0.75 and all abstracted water was applied to the HRUs. PA was not however applied in cotton HRUs irrigated from reservoirs within the study area. Unfortunately, here another bug was discovered; the model did not seem able to recognize the auto-irrigation parameters. The last BMP examined was that of WWR (BMP7 in Table 2). Such a practice would only be realistic for selected agricultural areas close to the biggest urban Waste Water Treatment Plants (WWTPs). Actually, interviewed farmers in the study area had recently shown high willingness to accept the inclusion of water reuse in water management plans for their area (Bakopoulou and Kungolos, 2009). In SWAT, WWR was simulated by simply changing the source of irrigation water from ‘aquifer’ to an outside the catchment’s area source and at the same time, by removing point source flows in the respective subbasins. Moreover, the doses of irrigation water were slightly decreased (15%) from the theoretical ones of the baseline, with water applied to the selected fields without meeting any constraints of availability. According to the waste water reuse potential that had accurately been calculated for each subbasin (Lourmas et al., 2012), the single cotton HRU of subbasin 9 with an area of 1200 ha was considered to receive wastes from the WWTP of the city of Trikala, while the cotton HRUs of subbasin 34 with a total area of 5000 ha, were considered to receive wastes from the WWTP of the city of Larisa (see Fig. 2). WWR was not considered for the 3rd big city of Karditsa as reuse potential calculated for this area included industrial wastes as well (Lourmas et al., 2012). 2.4.2. Costing of BMPs For costing the BMPs and comparing their CE in reducing agricultural water abstractions, their investment costs were first annualized based on the following formula: AEC =

r(1 + r)n

(1 + r)n − 1

× I + OMC,

(1)

where AEC is the annual equivalent cost; I represents the investment costs; OMC are the operational and maintenance costs related to the investment; r is the interest rate; and n is the useful life of the BMP (Berbel et al., 2011). The AEC was estimated by considering an interest rate of 7%, an OMC equal to 1% of the investment cost and the life time of each BMP. Indirect costs such as annual crop yield losses, annual fertilizer use and annual water pumping were subsequently added to Eq. (1). All prices were estimated with reference to the most recent years with available data and should thus be considered highly reliable. Details on the estimation of all costs used in this study are given next. PA can be implemented by using a yield monitor and 16 soil moisture sensors per 100 ha along with a data-logger and one evapotranspiration recording device per 10 ha of cotton land. The purchase cost of the yield monitor is 7000 D , of each soil sensor 35 D , of the data logger 200 D and of the evapotranspiration device 350 D . The total installation cost is calculated as 182 D /ha. This value should be however increased to include the necessary drip irrigation equipment installation required to operate the practice. This cost was calculated as 650 D /ha and the cost of the practice comes in total up to 832 D /ha for purchasing and installing the

equipment. Assuming a life time of 5 years for the equipment, the net present value of the AEC accounts for 200 D /ha/y. On the other hand, the analysis undertaken for the substitution of open irrigation canals from underground piped networks within the study area, indicated that the average construction cost was between 3000 and 8000 D /ha of irrigated land served by collective networks (Michas and Gkiokas, 2012). For the modeling study an average cost of those reported, equal to 5500 D /ha was finally adopted. Considering a 50-year life-time of the collective systems, Eq. (1) resulted in an AEC of 400 D /ha. Finally, WWR potential in Pinios was estimated by analyzing quantitative and qualitative data from WWTPs in the study area (Lourmas et al., 2012). Four different waste water reuse schemes were evaluated including restricted/unrestricted reuse, combined or not with a reservoir to store winter effluents. The term ‘restricted reuse’ includes only limited additional treatment of wastes compared to the existing ones but requires modifications in the existing irrigation methods (e.g. irrigation only with drip systems to limit the wet area in the field). The most promising reuse scheme in terms of water availability and cost of implementation seemed to be the restricted reuse with reservoir, which requires a new separated drip irrigation network and equipment, fencing and designation of irrigated areas as well as the reservoir construction. Based on the estimated costs, an average unit cost installation value for the equipment and reservoir equal to 0.2 D /m3 was chosen for the modeling study, which, for a total irrigation amount of 4500 m3 /ha can be translated to 900 D /ha. The AEC was calculated for a 20-year lifetime by adding the cost of wastes treatment on an annual basis selected with a unit average value of 0.003 D /m3 , which is approximately equivalent to 13.5 D /ha. The AEC of implementing the BMP in Pinios was calculated as 100 D /ha. Recent crop yields and fertilizer prices were also taken into consideration for the calculation of the total BMPs implementation costs and are included in Table 2. Expenses for fertilizers were calculated as 1.3 D /kg N and were correlated only with PA and PA with CI, as these were the only BMPs which could differentiate the fertilizer amount applied to the field. To complete the costing of BMPs, annual pumping costs from the aquifers had to be taken into consideration. A feasible approach was to make a correlation between energy (electricity) consumption and the volume of pumped water. The electricity cost in agriculture for the year 2012 is published on the basis of a kW h consumption (0.07 D /kW h) in the official website of the Public Power Corporation of Greece (http://www.dei.gr/). A correlation was conducted on an empirical basis based on various known local situations across Greece and the unit pumping cost was estimated between 0.01 and 0.02 D /m3 . According to the data analyzed, it was decided to correspond the lower value to the abstractions of the renewable groundwater reserves in the study area (low depths) and the higher value to the abstractions of the permanent groundwater reserves (high depths). For the HRUs that were irrigated by surface waters (reservoirs), cost savings due to less water abstractions were neglected.

3. Results and discussion The BMPs were tested one by one in the Pinios river basin. The model was executed 7 times for the period of 2011–2027 and simulated the mean annual effects of the baseline scenario and of each alternative practice on the water abstractions and crop productivity when applied entirely in the catchment. For the comprehension of the following results it is clarified that with the current SWAT capabilities, irrigation water abstracted from a source becomes loss from the subbasin as ET (crop consumption) or runoff (returns to the main channel). Based on our exhausting experimentation, returns to the aquifer via percolation still remain a very small part

Y. Panagopoulos et al. / Agricultural Water Management 139 (2014) 31–42

in SWAT. Moreover, whatever is lost via runoff should be considered a non-recoverable or not economically exploitable fraction (Perry, 2011) of the water cycle in the Pinios catchment. This is first attributed to the fact that runoff water from the irrigation of a subbasin cannot be used in another as each aquifer is a single entity that is uniquely enriched by the rainfall falling on the subbasin. Transmission losses from the river bed that could transfer runoff losses from one subbasin to the aquifer of a downstream one are also negligible in SWAT, while direct abstractions from the rivers do not occur in this modeling. As far as the operation of the ‘Smokovo’, ‘Girtoni’ and ‘Karla’ reservoirs is concerned (see Fig. 2), river flows are considerable to fill them with water even without the contribution of irrigation runoff losses from upstream subbasins both at the baseline and under a BMP implementation. Irrigation needs of the areas served by them are completely covered, so that the additional water originating from the upstream irrigation runoff losses that are recovered in reservoirs, should not be re-considered for the calculation of the net total irrigation abstractions in the catchment. Given the general water availability in reservoirs, abstractions from them would be the same even if upstream irrigation runoff losses did not occur, while the unused stored water, including these upstream losses, is released downstream to the sea either from the principal spillways by following pre-defined outflow rates or through the emergency spillways. If this was not the case and part of the upstream runoff losses was exploitable downstream, we should have subtracted it from the overall simulated abstracted water in the catchment in order to avoid a double-counting and make an accurate calculation. The abstracted water for irrigation from the reservoirs is applied to specific subbasins, then, irrigation runoff losses also occur in these subbasins and are driven directly to the sea through the stream network of the downstream subbasins. As far as the Pinios river flow at the outlet (sea) is concerned, it was simulated above the minimum required value of 10 m3 /s for almost all the months of the baseline period, a situation that was also found to be preserved when all BMPs were implemented showing that downstream water availability practically remains unaffected. Therefore, to address the CE of BMPs we use total abstractions (groundwater/reservoirs) for irrigation with the largest part becoming ET (consumed by crops) and the other runoff loss from the system. Reduced abstractions from the baseline when the BMPs are implemented correspond to the potential water saving in the catchment for use elsewhere. Although this goal is important, it does not result in saving water for another use in this work, as reduced groundwater and reservoir pumping improves the status of the water bodies but water is not exploited for other purposes. The unused water is what we call ‘saved water’ from this point onwards and has the potential either to simply improve the environmental status of the water bodies in the long-term or cause further economic benefits for the catchment with additional management. However, it is beyond the scope of this paper to evaluate the environmental status of the numerous water bodies (individual aquifers, reservoir lakes) after the implementation of BMPs or to suggest how saved water can be further exploited. Instead, we limit the analysis on calculating the CE of BMPs in reducing the abstracted water for irrigation. As was expected, there was a great variation of the BMPs’ environmental effectiveness across the catchment. To translate such spatial differentiations to comprehensive mapping results CE was calculated for each HRU and BMP implemented by using the estimated cost along with the simulated water abstraction. Fig. 4 depicts CE of the BMPs in reducing water abstraction, expressed in D /m3 saved from the baseline on a mean annual basis within the simulated period. The detailed mapping results are discussed in parallel with those calculated at the entire catchment level (Table 3). In this table, total costs are divided into 3 categories:

39

(a) cost 1, representing the direct BMP implementation cost, (b) cost 2, representing the indirect additional cost/income produced by the reductions/increases in crop yields simulated and (c) cost 3, representing the indirect income produced due to reductions in the amounts of groundwater pumped. The implementation of DI caused a 19% water saving in the study area or 176 hm3 /y. One would expect a reduction of 30% of the total abstractions as the practice was implemented in the entire irrigated land with a fixed 30% reduction of all the theoretical irrigation doses. However, there are significant areas, especially in the central and southern part of the catchment, where water needs could not be fully satisfied, thus, DI could not practically reduce the actually applied water in those areas by 30% but to a lower extent. The practice has no implementation cost; however, the lost income due to yield reductions (reduced ET as shown in Table 3) was high and approached 153 D /ha. The practice also resulted in less groundwater pumping cost, estimated near 8 D /ha of irrigated land and the CE rate of 0.17 D /m3 implies a quite cheap solution for reducing abstractions. Nevertheless, as can be seen in Fig. 4 for DI (BMP2) this ratio varies among the different HRUs in the catchment. Although in the majority of them CE was in the first and more desirable class (<0.5 D /m3 ), there are areas where DI caused significant cost because of yield reductions, followed by negligible pumping reductions. Such areas can be found in the southern and central part of the catchment where groundwater reserves were limited. On the other hand, the situation was different in the western part where a complete 30% water reduction occurred that had significant impacts on pumping cost reduction. Also, in areas served by reservoirs DI was found to be an acceptable practice with respect to CE without significantly reducing the simulated yields. The improvement of conveyance efficiency (BMP3) on the other hand, saved almost half of the water saved by DI. The practice was implemented by nullifying the 25% irrigation losses considered in the baseline but does not so much intend to save water as to increase the total irrigation efficiency. The practice resulted in a slight yield increase of almost 2–4% for all crops due to increased water consumption (ET), which is translated to an average income increase of 39 D /ha/y for the farmers. Moreover, reduced abstractions returns back a considerable amount of money spent for pumping water. Due to the significant installation cost of the practice, the total cost of implementation was high and equal to 354 D /ha on average for the irrigated areas in the catchment. The CE ratio at the catchment scale was calculated as 0.73 D /m3 of water saved, with the greatest parts of the modeled area being in that range (0.5–1.1 D /m3 ), as shown in Fig. 4 for BMP3. BMP4 is a combination of the previous 2, with water saving being 259 hm3 /y, the highest among all other BMPs tested. Installation costs (cost 1) were equal to CI costs, while crop yield lost was quite less than this caused by DI when implemented alone. The combined practice reduced water losses considerably, so that net water applications to the fields were closer, although reduced from the optimum due to DI. The reduction of pumping cost was further reduced to 15 D /ha as both practices saved considerable amounts of water, while total cost was the highest and equal to 82 MD . Nevertheless, CE was in acceptable levels (0.32 D /m3 ) due to the significant water savings. Actually, the latter is the most promising result caused by this combined BMP, which is the only practice among the ones tested, able to reduce water abstractions by more than 30%, below the sustainable levels of 700 hm3 calculated for the whole catchment. Even in this case significant areas in the central and southern part of the basin could be found where overexploitation still existed, although it was considerably reduced from the baseline. The over-30% reduction at the catchment level was achieved by computing water savings from reservoirs. The high total reduction simulated can be considered a promising result for the catchment indicating that with additional infrastructure

40

Y. Panagopoulos et al. / Agricultural Water Management 139 (2014) 31–42

Fig. 4. Cost-effectiveness (CE) of BMPs in reducing water abstractions across Pinios (D /m3 saved on an annual basis). BMP2: deficit irrigation (DI), BMP3: conveyance improvement (CI), BMP4: DI + CI, BMP5: precision agriculture (PA), BMP6: PA + CI. All BMPs are analyzed in Table 2.

(irrigation networks) across parts of it, surface water savings in reservoirs could be potentially used in areas suffering from groundwater overexploitation. The spatial differentiation of CE reveals an improvement for significant areas in relation to the single BMPs Nos. 2 and 3. However, there are still areas in the central and southern part of Pinios where the ratio lies in the second class (0.5–1.1 D /m3 ) mainly due to insignificant water savings there. On the other hand, water saving was considerable in the western part as well as in areas served by reservoirs, with the CE ratio being calculated at low levels. From the consecutive analysis of the BMPs Nos. 2, 3 and 4 it could be concluded that for reaching an ambitious environmental target of water use reduction, CI, although

expensive, should be combined with DI, even only for areas of limited extent within the catchment including first the irrigated areas served by the 4 reservoirs (see Fig. 2). The 5th BMP representing PA saved 11% of water compared to the baseline but in many HRUs resulted in a greater than 20% reduction. However, corn and alfalfa as well as cotton areas irrigated by the 3 reservoirs located within the catchment were excluded from the implementation of the practice, As noted in the table, the cost of implementing PA was close to 28 MD or 140 D /ha of total irrigated land (including the areas where not implemented). Actually, even for BMPs Nos. 5, 6 and 7, which were not tested in the entire irrigated land, all cost results in Table 3 are expressed per ha of the

Table 3 BMPs implementation results in the Pinios river basin. BMPs

Total water used (hm3 ) Water saving (hm3 ) Water saving (%) ET (mm) Cost1 (D /ha) Cost 2 (D /ha) Cost 3 (D /ha) Total cost (D /ha) Total cost (MD ) CE (D /m3 )

1 Baseline

2 DI

3 CI

4 DI–CI

5 PA

6 PA–CI

7 WWR

916 0 0 504.8 0 0 0 0 0

740 176 19.2 493.5 0.00 152.69 −7.65 145.04 29.30 0.17

818 98 10.7 508.6 400.43 −39.02 −7.20 354.20 71.55 0.73

657 259 28.3 503.8 400.43 19.67 −14.70 405.40 81.89 0.32

818 98 10.7 501.3 105.67 40.49 −8.11 138.06 27.89 0.28

793 123 13.4 504.9 370.38 15.42 −10.76 375.04 75.76 0.62

902 14 1.6 506.9 3.03 −20.72 −0.79 −18.48 −3.73 −0.26

DI, deficit irrigation; CI, conveyance improvement; PA, precision agriculture; WWR, waste water reuse. ET is the mean annual evapotranspiration. Positive cost values indicate additional annual cost of BMP implementation (AEC) compared to conventional practices (baseline). Negative values indicate additional farmers’ income arising from BMP implementation. BMPs 5 and 6 were implemented only in cotton areas excluding those HRUs irrigated by reservoirs located inside the basin (total area of such HRUs ∼30,000 ha), BMP 7 only in 3 cotton HRUs (∼6000 ha), while all other BMPs were applied to all irrigated crops (alfalfa, corn and cotton), which cover in total 202,000 ha. Cost 1 is the direct BMP implementation cost (BMP installation cost including fertilization expenses reduction in the case of PA application), cost 2 the cost arising from yield changes and cost 3 the reduced cost because of pumping water reductions from the baseline. All cost results are expressed per ha of the total irrigated land (202,000 ha) in order to give the magnitude of cost at the catchment scale when each BMP is adopted for implementation at the greatest possible extent. CE is expressed in D /m3 saved and was calculated by dividing the total cost needed to implement the BMP (MD ) with the total water saved (hm3 ). 1 hm3 = 106 m3 .

Y. Panagopoulos et al. / Agricultural Water Management 139 (2014) 31–42

total irrigated land (202,000 ha) in order to give the magnitude of cost at the catchment scale if each of them was to be solely selected for implementation at the greatest possible extent. A notable result is that the 1st type of cost is almost half than the cost of installation (200 D /ha). This is not just the result of including the total irrigated land for the ‘per ha’ cost calculation but also the result of fertilization savings that were incorporated in this type of cost. PA saved more than 30 kg N/ha in fertilization, which approximately resulted in a cost saving of 40 D /ha of land where the practice was applied. On the other hand, notable cotton yield increases could be simulated only in areas with adequate groundwater availability. Thus, due to the great extent of groundwater deficient cotton areas, simulated yields at the catchment level appear to have been slightly reduced from the baseline resulting in the positive value of the 2nd type of cost in Table 3. In total however, the practice increases crop water productivity throughout the study area by reducing water abstractions and crop consumption (ET) for almost the same level of crop yield, something very important, especially for the most deficient areas. The reduction in the cost of pumping on the other hand, was calculated greater than 8 D /ha of the total irrigated land, indicating that PA, even when implemented to a lower spatial extent can save more groundwater than the single DI and CI BMPs. The average CE of PA at the catchment level was calculated as 0.28 D /m3 and is considered highly acceptable, while as can be seen in Fig. 4 for BMP 5, almost the whole irrigated land belongs to the first and most acceptable CE class. BMP No. 6 is the combination of PA and CI with water saving being further increased to 13.4% from the baseline. With the reduction of conveyance losses the maximum available water up to the cotton theoretical dose of 50 mm was applied at the timings when soil moisture was depleted. As soil moisture was reaching the threshold point less frequently, less water was totally spent within the growth cycle of the crop. Therefore, the pumping cost saving was higher than that of the single PA BMP. The total cost at the catchment level was 76 MD or 375 D /ha of total irrigated land and is lower than that caused by the combined DI-CI BMP (BMP No. 4) mostly because of the fertilizers saving. It should be however noted that the practice was implemented only in cotton areas irrigated by groundwater resources, so that unit area costs inevitably appear lower when expressed per ha of total irrigated land. The CE of the practice is rather undesirable (0.62 D /m3 ) and its spatial differentiation in Fig. 4 (BMP6) reveals that it lies within the range 0.5–1.1 D /m3 for almost all of the groundwater fed cotton HRUs. Finally, there were significant benefits for the farmers of the HRUs where waste water reuse was implemented. Irrigation was full and crop needs were completely satisfied, while pumping cost was zero, with water being provided from an external source. The values in Table 3 had been again calculated for the whole irrigated land in the catchment and as such, both unit area costs and the total water saving (14 hm3 or 1.6%) are small. However, it can be concluded that waste water reuse may be preferred among all other practices for implementation in the particular 3 HRUs of 6200 ha area in total (see Table 2) as the optimum irrigation conditions maximized yields, leading to 3.73 MD total cost reduction. If expressed per ha of land where only implemented, the annual income increase for the local farmers would be 600 D /ha. Due to the limited area of implementation a CE map is not provided here. Potential limitations in the CE predictions described here may be linked with the ability of SWAT to accurately represent natural processes and BMPs application. For instance, the groundwater routine is a rough approximation of the real world with the area of each groundwater body (aquifer) coinciding with the subbasin’s boundary. Therefore, the outputs of a SWAT study may be sensitive to the schematization of the catchment and its division into subbasins. It is for example obvious that in the case of the study area of this paper, the less the area occupied by irrigated crops in a

41

subbasin, the higher the possibility for their needs to be satisfied from groundwater, given the fact that groundwater was not considered to be an endless source. On the other hand, the new SWAT routines for the simulation of irrigation conveyance efficiency have still bugs/errors preventing the ‘direct’ representation of water losses in some situations with the use of the ‘IRR EFF’ parameter. Moreover, even when the parameter is active, one should consider the fact that irrigation conveyance losses do not completely disappear from the system as SWAT assumes but partly return to the aquifers. The use of the ‘IRR SQ’ parameter to represent them as runoff losses in the baseline achieved to simulate appropriately the amount of water applied to the HRUs; however, on top of no aquifer return simulation, this way caused a 5–10% reduction of the average annual river flow in all cases when conveyance improvement was implemented, something not very realistic. However, river flows at the outlet were significant implying a good status under all BMPs implemented due to the considerable unexploitable flows still occurring in the catchment. From our experimentation with irrigation in SWAT, we also identified inaccuracies in the water balance simulation. For instance, differences in the percolated water simulated under various irrigation practices were expected higher, while, in general, when irrigation is applied, the model seems to give a relatively increased priority to the satisfaction of the ET requirements. More studies are strongly encouraged to support our findings. Secondary limitations in the CE predictions may arise from the BMPs cost calculations and the degree of uncertainty, which is always included in absolute values. However, the cost estimation of each type of water management intervention was conducted within the desertification project ‘i-adapt’ (www.i-adapt.gr) by experts in each field using the most recent unit cost/prices and relevant data. It is thus believed that possible small deviations, especially due to the usual economic data uncertainties and the approximations made for the modeling study, cannot be able to affect the ranking of the different BMPs with respect to their total cost of implementation.

4. Conclusions and recommendations This study tested four irrigation water management BMPs and two sensible combinations of them in the water deficient Pinios river basin in central Greece demonstrating a feasible simulation with SWAT. The cost of BMPs implementation was calculated based on recent unit costs/prices including direct and indirect types of costs related to crop yields and water pumping. From the analysis undertaken, it can be concluded that a cost-effective management intervention in the study area would include: (a) waste water reuse for the adjacent areas to WWTP installations, (b) deficit irrigation in the western part of the catchment as well as in areas irrigated by reservoirs and (c) precision agriculture technologies in the highly water scarce central and southern parts of the catchment. Conveyance losses reduction was also examined, but was identified as the most expensive BMP drastically increasing total costs when implemented; at the same time however, it considerably increases the efficiency of both deficit irrigation and precision agriculture when implemented in the same areas. It is suggested that its incorporation within an integrated management scheme of Pinios would depend only on funding availability. From the tabulated and mapping results presented in this paper, one could clearly conclude that the CE of irrigation BMPs in reducing water abstractions for irrigation at the local or the catchment level can be considered an informative index for choosing between different interventions. Clearly the work would also benefit identification and inclusion of socio-economic constraints related to site-specific BMPs implementation, to allow for a more realistic

42

Y. Panagopoulos et al. / Agricultural Water Management 139 (2014) 31–42

design of solutions. What is even more needed for a less uncertain decision making based on SWAT predictions is a detailed review of the irrigation water management module and the relevant parameters in particular by the SWAT-developers team, making the tool entirely operational in water management studies. The consideration of groundwater connectivity between subbasins should definitely be considered in a future version. Also, improvements in the simulation of irrigation returns and groundwater return flows from conveyance inefficiencies in particular are necessary, while the correction of bugs in the code will allow all practices to be implemented under all combinations of irrigation application ways (dates/irrigation amounts or auto-irrigation) and irrigation sources (e.g. groundwater/reservoirs). It is however, believed that, despite limitations and needs for future improvements of the processbased model, the current study presents a useful approach for the cost-effective development of Programs of Measures, to be embedded in RBMPs for predominantly agricultural catchments in water scarce areas, such as the Mediterranean. Acknowledgments The current research was conducted under the project ‘i-adapt’ (http://i-adapt.gr), which is one of the Pilot projects on Development of Prevention Activities to Halt Desertification in Europe, partly funded by DG Environment of the European Commission, Grant agreement number: 07.0316/2010/581799/SUB/D1, Period of implementation: 1st Feb 2011–31st May 2012, Target country/region: Pinios River Basin, Thessaly, Greece. References Abbaspour, K.C., Yang, J., Maximov, I., Siber, R., Bogner, K., Mieleitner, J., Zobrist, J., Srinivasan, R., 2007. Modelling hydrology and water quality in the prealpine/alpine Thur watershed using SWAT. J. Hydrol. 333, 413–430. Abbaspour, K.C., 2011. User Manual for SWAT-CUP 4.3.2. SWAT Calibration and Uncertainty Analysis Programs. Swiss Federal Institute of Aquatic Science and Technology, Eawag, Duebendorf, Switzerland, pp. 103. Arabi, M., Frankenberger, J.R., Engel, B.A., Arnold, J.G., 2008. Representation of agricultural conservation practices with SWAT. Hydrol. Process. 22 (16), 3042–3055. Bakopoulou, S., Kungolos, A., 2009. Investigation of wastewater reuse potential in Thessaly region, Greece. Desalination 248, 1029–1038. Berbel, J., Martin-Ortega, J., Mesa, P., 2011. A cost-effectiveness analysis of watersaving measures for the water framework directive: the case of the Guadalquivir River Basin in southern Spain. Water Resour. Manag. 25, 623–640. Cherry, K.A., Shepherd, M., Withers, P.J.A., Mooney, S.J., 2008. Assessing the effectiveness of actions to mitigate nutrient loss from agriculture: a review of methods. Sci. Total Environ. 406 (1/2), 1–23. Dechmi, F., Burguete, J., Skhiri, A., 2012. SWAT application in intensive irrigation systems: model modification, calibration and validation. J. Hydrol. 470/471, 227–238. Directive 2000/60/EC of the European Parliament and of the council of 23 October 2000 establishing a framework for Community action in the field of water policy. Official Journal of the European Communities L327/1. EEA, 2012. Towards efficient use of water resources in Europe. EEA Report No 1/2012. EEA, Copenhagen, ISBN 978-92-9213-275-0, ISSN 17259177, DOI 10.2800/95096. Available online: http://www.eea.europa.eu/ publications/towards-efficient-use-of-water EEA-ETC/TE, 2002. CORINE Land Cover Update. I&CLC 2000 Project, Technical Guidelines. http://etc-lusi.eionet.europa.eu/ Eurostat, 2000. http://epp.eurostat.ec.europa.eu (accessed April 2010). Estrela, T., Vargas, E., 2012. Drought management plans in the European Union. The case of Spain. Water Resour. Manag. 26, 1537–1553. Evangelou, L., Vlouchos, C., Moustaka, K., Tsadilas, C., 2011. DPSIR Study for Desertification in Pinios. Final Report. Deliverable for the Task B of the Project i-adapt: Innovative Approaches to halt Desertification in Pinios. DG Environment.

Evangelou, L., Tsadilas, C., 2012. Pilot of a Precision Irrigation Experiment of Cotton in Pinios River Basin. Final Report. Deliverable for the Task D of the Project i-adapt: Innovative Approaches to Halt Desertification in Pinios. DG Environment. Faramarzi, M., Yang, H., Schulin, R., Abbaspour, K.C., 2010. Modeling wheat yield and crop water productivity in Iran: implications of agricultural water management for wheat production. Agric. Water Manage. 97 (11), 1861–1875. Feng, H., Baoguo, L., 2010. Assessing grain crop water productivity of China using a hydro-model-coupled-statistics approach. Part I: Method development and validation. Agric. Water Manage. 97, 1077–1092. Gassman, P.W., Reyes, M.R., Green, C.H., Arnold, J.G., 2007. The soil and water assessment tool: historical development, applications, and future research directions. Trans. ASABE 50 (4), 1211–1250. Geerts, S., Raes, D., 2009. Deficit irrigation as an on-farm strategy to maximize crop water productivity in dry areas. Agric. Water Manage. 96, 1275–1284. Kloss, S., Pushpalatha, R., Kamoyo, K.J., Schütze, N., 2012. Evaluation of crop models for simulating and optimizing deficit irrigation systems in arid and semi-arid countries under climate variability. Water Resour. Manag. 26, 997–1014. Loukas, A., Mylopoulos, N., Vasiliades, L., 2007. A modeling system for the evaluation of water resources management strategies in Thessaly, Greece. Water Resour. Manage 21, 1673–1702. Lourmas, G., Papageorgiou, S., Evangelatos, F., Mamais, D., 2012. Report on the Waste Water Reuse Potential for Pinios. Final Report. Deliverable for the Task C1 of the Project i-adapt: Innovative Approaches to Halt Desertification in Pinios. DG Environment. Masih, I., Maskey, S., Uhlenbrook, S., Smakhtin, V., 2011. Impact of upstream changes in rain-fed agriculture on downstream flow in a semi-arid basin. Agric. Water Manage. 100, 36–45. Michas, S., Gkiokas, A., 2012. Investigation of Irrigation Network Losses in Thessaly. Final Report. Deliverable for the Task D1 of the Project i-adapt: Innovative Approaches to Halt Desertification in Pinios. DG Environment. Makropoulos, C., Mimikou, M., 2012. ‘i-adapt’: Innovative Approaches to Halt Desertification in Pinios: Piloting Emerging Technologies. School of Civil Engineering, National Technical University of Athens, Available online at: http://i-adapt.gr/downloads/i-adapt-monograph.pdf Moriasi, D.N., Arnold, J.G., Liew, M.W., Bingner, R.L., Harmel, R.D., Veith, T.L., 2007. Model evaluation guidelines for systematic quantification of accuracy in watershed simulations. Trans. ASABE 50 (3), 885–900. Mouratiadou, I., Moran, D., 2007. Mapping public participation in the Water Framework Directive: a case study of the Pinios River Basin, Greece. Ecol. Econ. 62, 66–76. Nalbantis, I., Koutsoyiannis, D., 1997. Final Report, Upgrading and Updating of Hydrological Information of Thessalia, Report 4. Department of Water Resources, Hydraulic and Maritime Engineering, National Technical University of Athens, Athens, 78 pp., Available online at: http://itia.ntua.gr/getfile/186/1/ documents/er5 4.pdf (in Greek). Nash, J.E., Sutcliffe, V., 1970. River flow forecasting through conceptual models. I. A discussion of principles. J. Hydrol. 10, 282–290. Neitsch, S.L., Arnold, J.G., Kiniry, J.R., Williams, J.R., 2009. Soil and Water Assessment Tool – Theoretical Documentation-Version 2005. Blackland Research Center – ARS, TX, USA http://twri.tamu.edu/reports/2011/tr406.pdf Panagopoulos, Y., Makropoulos, C., Mimikou, M., 2011. Reducing surface water pollution through the assessment of the cost-effectiveness of BMPs at different spatial scales. J. Environ. Manage. 92 (10), 2823–2835. Panagopoulos, Y., Makropoulos, C., Mimikou, M., 2012. Decision support for diffuse pollution management. Environ. Modell. Softw. 30, 57–70. Perry, C., 2011. Accounting for water use: terminology and implications for saving water and increasing production. Agric. Water Manage. 98, 1840–1846. Santhi, C., Muttiah, R.S., Arnold, J.G., Srinivasan, R., 2005. A GIS-based regional planning tool for irrigation demand assessment and savings using SWAT. Trans. ASAE 48 (1), 137–147. Vasiliades, L., Loukas, A., Liberis, N., 2011. A Water Balance Derived Drought Index for Pinios River Basin, Greece. Water Resour. Manag. 25, 1087–1101. Williams, J.R., 1995. The EPIC model. In: Singh, V.P. (Ed.), Computer Models of Watershed Hydrology. Water Resources Publications, Highland Ranch, CO, USA, pp. 909–1000. Wriedt, G., Van der Velde, M., Aloe, A., Bouraoui, F., 2009. Estimating irrigation water requirements in Europe. J. Hydrol. 373, 527–544. Xie, X., Cui, Y., 2011. Development and test of SWAT for modeling hydrological processes in irrigation districts with paddy rice. J. Hydrol. 396, 61–71.