A quantitative approach on environment-food nexus: integrated modeling and indices for cumulative impact assessment of farm management practices

Methodology

This study follows a 4-step combined methodology. In the first two steps, data is gathered and a basin is simulated by the SWAT model with the perspective of water quality and quantity. Here, the effectiveness of different BMPs on exporting pollution loads (kg/ha), pollutants concentration in lake (mg/L), crop production yields (ton/ha), nutrition production (Kcal/yr), and water footprint (m³) are evaluated. Thus, the modeling provides a quantitative framework for further environmental-food analysis in basin. In this study, the first two steps, except the nutrition production, follows the previously developed SWAT model by Jamshidi, Imani & Delavar (2020).

In the third step and to quantify the CIAs of BMPs, a combined method is developed to convert the modeling results into equivalent environmental damages. An excel-base LCIA method according to ReCiPe (2016 v1.1) is used including related characterization midpoints (water consumption and eutrophication) and endpoints (human health and ecosystem damages) with normalization coefficients. In this step, some new approaches are considered to develop LCIA analysis. For example, the embodied water consumption, directly analyzed by the SWAT model (WF), is introduced as a reliable water consumption index for the LCIA of food crops. This is due to the fact that food crop’s WF includes both consumed (blue and green) and polluted (grey) water. These items fit to life cycle assessment of available water in the ecosystem (Bigdeli Nalbandan et al., 2022). In addition, this step considered two different weighting approaches for integrating health and ecosystem damages (endpoints) as a single index. The entropy analysis uses a mathematical equation to calculate the weights of health and ecosystem, while environmental performance index (EPI) applies predefined weights for the two endpoints.

In final step, a state-of-the-art index is introduced as “environmental footprint of food production” (FEF) that calculates the cumulative environmental damages of nutrition production in basins. This new index is applicable for quantifying the equivalent environmental damages related to food production. It also compares the cumulative impacts of BMPs and farm management practices by considering different perspectives like WF, pollution emissions, crop nutrition, and ecosystem protection. The main contribution of this research is in its combined method, particularly the third and fourth steps. Here, an environment-food nexus analysis compares the cumulative impacts of BMPs in a basin. The methodology steps are illustrated in Fig. 1.

Zrebar Basin, western Iran, was chosen as the study location to verify the method. It doesn’t mean this approach is developed for a specific basin. On the contrary, the SWAT-ReCiPe is applicable in any basin for comparing farm management strategies.

Study area

The proposed methodology is verified in Zrebar Lake basin, western Iran. Zrebar basin encompasses 90 km² including 20 km² of irrigated and rain-fed farmlands (22%). Its lake meets eutrophication problem mainly due to the agricultural discharges, particularly irrigated farmlands (Imani, Delavar & Niksokhan, 2019). Main rain-fed (RF) crops in this area are wheat, barley, grape and peas with average nutrition values of 3,640, 3,540, 670 and 420 cal/kg, respectively. The irrigated crops include tomato, tobacco, alfalfa, apple with average nutrition values of 180, 0, 230, and 520 cal/kg, respectively in addition to irrigated wheat and barley. Figure 2 shows the dominant LUs in the study area with its geographical conditions.

Simulation-calibration

In the proposed methodology, the SWAT model is used for basin simulation before accounting environmental damages and footprints of agricultural productions. This model can simulate complicated systems by considering management practices in farmlands, interactions between water quality and quantity, pollution transport, and production yields (Abbaspour et al., 2015; Arnold et al., 2012b; Rivas-Tabares et al., 2019). Therefore, required data such as topography, soil properties, LU type, management practices, and weather/climate were inputted to the model (Table 1). The basin was split into 26 sub-basins and 1,100 HRUs. This model was calibrated and validated based on available data (2006–2013) of monthly lake inflow, nitrate and phosphate concentrations simultaneously. Production yields and evapotranspiration rates were also controlled with the observation data (Jamshidi, Imani & Delavar, 2020). Table 2 outlines the calculated regression coefficient (R²) and RMSE-observations standard deviation ratio (RSR) in the calibrated model.

It is noteworthy that the main idea of this research is to develop an integrated method for accounting environment-food nexus. Accordingly, authors used the outputs of the already calibrated SWAT model previously developed for BMP and WF assessment in the study area (Jamshidi, Imani & Delavar, 2020). Thus, simulation-calibration details are skipped here as they can be fully retrieved in the cited reference.

BMP scenario

This study uses the SWAT outcomes for BMP analysis in three scenarios as defined in Table 3. Base is the scenario without using any BMPs. In BMP1, the application of fertilizers, manure and chemical, and water for irrigation are reduced 25% for farmlands. In BMP2, the reduction equals 50%. In both BMP scenarios, FS is assumed to be implemented in the vicinity of lake. Slim FS represents 10–12 m width, while moderate FS has 20–25 m width. All scenarios are analyzed by the SWAT model in the same period from 2007–2013.

Water footprint

The WFs of agricultural productions are accounted by the standard method and include the three main elements of green, blue and grey water (Franke, Boyacioglu & Hoekstra, 2013; Hoekstra et al., 2011). It should be noted that WFs calculate the direct embedded water of farmlands and exclude indirect water embodied in further processing of agricultural productions.

(1) $$WF=GnWF+BWF+GWF$$

(2) $$GnWF=10E{T}_a$$

(3) $$BWF=10(E{T}_b-E{T}_a)$$

(4) $$GWF=max{\left({\displaystyle \frac{L}{C_{max}-C_{nat}}}\right)}_i$$

In these equations, GnWF, BWF and GWF are green, blue and grey water footprints (m³), respectively. ET_a refers to the evapotranspiration from soil and vegetations when there is no irrigation (mm). ET_b includes the total evapotranspiration during irrigation (ET_b > ET_a). Thus, the SWAT models evapotranspiration twice with and without irrigation. It uses the climatic data of minimum and maximum daily temperature with precipitation as mentioned in Table 1. Afterwards, it estimates the actual evapotraspiration for crops by Hargreaves equation (Kisi, 2007; Majidi et al., 2015). L is the exported pollution loads (ton/ha) of pollutant i to the receiving water body. In modeling, output.hru file shows the pollution loads per each HRU and sub-basin. L is the net pollution loads transported from LUs into the 26^th sub-basin, the Zrebar Lake in this study (Arnold et al., 2012a). C_max is the maximum allowable concentration of pollutants. C_nat equals pollutants concentration in the receiving water on the condition there is no human interference. Here, the C_max of TN and TP are assumed constant according to the global limits for controlling the trophic state of lakes (Jamshidi, 2021) and they equal 1.5 and 0.035 mg/L, respectively. C_nat of TN and TP are also assumed 0.4 and 0.01 mg/L, respectively (Jamshidi, Imani & Delavar, 2022).

Environmental impact assessment

The quantification method of environmental damages in basin is compatible with the indices of LCIA. In the current research, LCIA characterization coefficients are derived according to the ReCiPe method, which was previously developed by some collaborations in Europe (Huijbregts et al., 2017). In this method, normalized data at the European and global level are available for 16 midpoint and three endpoint indices. In later updates, ReCiPe considered several conversion coefficients based on global scale, instead of the European scale. However, it preserved the possibility of using these coefficients on the continental and country scales. Another feature of ReCiPe is to expand environmental damages for evaluating the impacts of water consumption on human health, aquatic and terrestrial ecosystems (Huijbregts et al., 2017). However, the current study proposes to use WF for accounting the water consumption of food crops in LCIA. This is due to the ability of WF in calculating water consumption including both water quality and quantity.

In this method, all effective environmental factors derived from the SWAT model are initially converted into the equivalent units. Table 4 illustrates eutrophication midpoint coefficients that convert NO₃, NO₂, NH₃ and PO₄ to the equivalent environmental damages. The average WFs of crops are considered (m³) for water consumption midpoint in aquatic, terrestrial and marine ecosystems. Equation (5) shows how conversions are carried out.

(5) $$Q_j={\left(T\times M\right)}_j$$

Q is the midpoint index, T represents the output of the SWAT model such as water footprint or pollutant concentration, M is the conversion coefficients, and j is environmental component such as aquatic, terrestrial, and marine. By this equation, it is possible to calculate the equivalent environmental effects of each pollutant in the life cycle period of the product or activity. It should be noted that these coefficients represent average values. It means that they do not need supplementary conversion coefficients for shallow or deep waters as they are free from adjustment for the conditions with different vegetation or trophic state. Moreover, pollution discharge to freshwater has indirect impacts on other ecosystems in long-term. Thus, marine impacts are also considered in calculation even the pollution is not directly discharged to the sea.

Since the midpoint indices are calculated based on equivalent units, such as kgN-eq or m³ water consumed, it is necessary to accumulate these environmental impacts with different units under a single index. This is the most challenging step in conventional CIA. ReCiPe uses equivalent damage-based indices for integrating midpoints into endpoints by Eq. (6).

(6) $$D_j={\left(Q\times E\right)}_j$$

Here, the calculated midpoint indices (Q) are converted into endpoint damage-based indices (D) according to conversion coefficient of E (see Table 5). Here, human health and ecosystem (non-human) damages are two endpoint indices. The former is based on disability-adjusted life years (DALY) and the latter is based on probable number of harmed species in year (species.yr). DALY represents the equivalent years of human life lost by death or being disabled due to illness caused by existing pollutants in the environment. On the other hand, the unit of measuring ecosystem damage is the total number of species lost over time. Table 5 shows the conversion coefficients that turn each equivalent midpoint indices into the two endpoints. ReCiPe model also recommends that endpoints (D) should be normalized by specific coefficients that turn the calculated damages into dimensionless indices per person (Sleeswijk et al., 2008).

Normalization and weighting

Calculated endpoints are normalized by Eq. (7) on a global scale based on reference coefficients (Table 5). They are finally aggregated according to their weights by Eq. (8). In this study, entropy and EPI are the weighting methods of normalized endpoints.

(7) $$R={\displaystyle \frac{D}{N}}$$

(8) $$C=\sum (W\times R)$$where, C is the annual environmental damage per person, W is the weight of each endpoint, N represents the normalization value and R is the normalized endpoint. Weights can be calculated based on different mathematical methods, such as entropy or fuzzy (Chen et al., 2019; Zeng, Luo & Yan, 2022), or based on expert opinions and references (Chen et al., 2022). In this study, EPI determines health and ecosystem weights as 0.4 and 0.6, respectively (Hsu & Zomer, 2016), whereas entropy method (W_En.) calculates the weights of endpoints through a probabilistic function as Eq. (9).

(9) $$W_{En.}=-{\displaystyle \frac{1}{\mathrm{l}\mathrm{n}\left(t\right)}\sum _{\mathrm{z}=1}^{\mathrm{t}}{\left(\mathrm{R}\times \mathrm{l}\mathrm{n}\mathrm{R}\right)}_{\mathrm{z}}}$$

In which, t is the number of available data. In entropy, factors with more data dispersion gain higher weights (Imani, Delavar & Niksokhan, 2019). Here, the weights of endpoints (R) are evaluated based on C variations from 2007–2013 in each BMP scenario. Accordingly, the ecosystem and health endpoints weigh 0.44 and 0.56, respectively in the entropy method.

Environmental-food index

This study introduces a new index for food and nutrition production in farmlands. It is quantified based on the environmental damages calculated by the SWAT-ReCiPe. This index quantifies the CIA per food production in any area or BMP as Eq. (10).

(10) $$FEF={\displaystyle \frac{C}{S}}$$

In this equation, FEF is a dimensionless index that represents the CIA of food production. In other words, FEF is the environmental footprint of nutrition production. It can be calculated by the proposed method for comparing major environmental concerns in food production, including water-food nexus. Low FEF (~0) means that strategies used for food production is rather clean, while higher FEF (>1) indicates their destructive condition. C is defined earlier that notes environmental damages (CIA), and S is calculated by Eq. (11).

(11) $$S={\displaystyle \frac{T_{Cal}}{B\times P}}$$

In which T_Cal is the daily total nutrition (calories) of food production in the study area, B equals the malnutrition baseline of humans assumed 2,000 cal/day (Liu et al., 2022), and P is the global population (7.75 billion) to convert and normalize S per person in global scale.

SWAT outcomes

The basin simulation by the SWAT model could calculate the annual pollution loads exported by HRUs in different management scenarios. Figures 3 and 4 respectively show the cumulative N and P loads discharged by farmlands in three scenarios. Here, the annual variations are due to (1) the precipitation variation influencing pollution transport from RF farms, and (2) temporary water transfers from upstream for irrigation development. The average N pollution of all irrigated and RF farms ranges between 1,176 and 3,985 kg yr⁻¹. This value for P is between 20 and 82 kg yr⁻¹. For 20 km² farming area, the average export coefficients for N and P are 0.6–2 kg ha⁻¹ yr⁻¹ and 0.01–0.04 kg ha⁻¹ yr⁻¹, respectively. This range implies that nutrient export coefficients can increase 3–4 times greater than dry periods in the study area. On an average for 2007 to 2013, BMP1 can reduce 33.8%N and 7.7%P pollution exported from all agricultural LUs. BMP2 can improve these removals to 59.9% and 20.9% for N and P, respectively. It implies that basin response to BMPs’ is not linear. In addition, P removal requires stricter BMPs than N removal. Yet, nutrient pollution reduction may have different ecological impacts on marine, aquatic and terrestrial systems which are accounted through the combined method. BMPs are also effective on crops production yields, WFs and nutrition productions (Table 6).

Environmental impacts

For the base scenario, the combined method calculates the environmental midpoint impact (Q) of farming in Zrebar basin as Fig. 5. It shows that freshwater eutrophication is the most critical item during the study period. The embodied water consumed is also significant for damaging the terrestrial ecosystem and human health. This conclusion remains unchanged in BMP1 and BMP2 despite 25–50% FR (Fig. 6). Since TP concentration in lake is the main driver of freshwater eutrophication, consuming less ammonium-based fertilizer can hardly solve eutrophication problem in short term in the combined method. On the contrary, controlling erosion and sediment transport by FS from upstream is more efficient for TP mitigation in the lake.

Figure 7 shows the cumulative ecological damages. Their values in Zrebar basin are relatively larger than health problems in all management scenarios. It is noteworthy that human health indices are mostly reliant on toxins and heavy metals. These pollutants were hardly traced in the study area. The results present that farm management strategies can mitigate the average ecological impacts from 1.41E−6 to 1.34E−6 (4.9%) for BMP1 and 1.28E−6 (9.2%) for BMP2. Likewise, these strategies can diminish human health risk from 2.58E−7 to 2.4E−7 (6.8%) for BMP1 and 2.22E−7 (13.9%) for BMP2. It means that using 50% less fertilizer with a FS in this area may totally reduce 9% ecological and 14% health risks (Fig. 8). Here, the cumulative impacts are low but not negligible as they range between 1E−6 and 1E−7 per person. However, these values are meaningless unless they are used as quantitative tools for comparative analysis.

Figure 8 summarizes the impacts of BMPs on food production (S) in addition to normalized environmental impacts (per person) on the ecosystem and health. Since nutrition production is intrinsically a favorable action with environmental perspective, their related impacts are negative. The overall environmental impact of farming and related management practices are finally calculated by the weighted average of normalized ecosystem and health damages. The weighing step is carried out with different methods. Since EPI gives higher weights to ecological items, the related results are relatively more than entropy method. Despite different weighting approaches, the overall CIA (C) reduction for BMP1 ranges between 5–8%, while it ranges between 10–13% for BMP2. It implies that using strict BMPs may not necessarily have significant improvement. On the contrary, S is reduced 1.66% and 3.73% by BMP1 and BMP2, respectively. It points an important conclusion that although farm management practices may reduce environmental damages, they can adversely reduce the nutrition production. This conclusion highlights an environment-food nexus index for more comprehensive understanding of management impacts.

Figure 9 draws the environmental footprint of nutrition production (FEF) in Zrebar Basin. Here, the conventional pattern (base scenario) can quantitatively generate 0.61 (entropy) and 0.78 (EPI) environmental impacts. In other words, 0.61–0.78 environmental units would be damaged for one unit nutrition production. This is a footprint mainly accounted by the impacts of consumed water and eutrophication in the study area originated by the agriculture. Using BMP1 and BMP2 can reduce FEF 6.5–9.1% (entropy) and 4–6.4% (EPI), respectively. It means that 50% FR combined with FS (BMP2) can reduce 6.4–9.1% of FEF in Zrebar basin. Obviously, this new index is more helping for policy makers rather than conventional approach on pollution reductions in a basin. For example, this index can present criteria to compare two alternatives of implementing vegetated FS or changing crop patterns in a basin. The first alternative only reduces pollution loads and consequently environmental impacts, while the second option emphasizes nutrition improvement despite pollution discharges.