An economic and environmental assessment of a glyphosate ban for the example of maize production

Abstract

We aim to contribute to a more informed discussion of the economic and environmental effects of a glyphosate ban in European agriculture. As real-world observations of weed control under a glyphosate ban are not available, we develop a normative modelling approach based on damage abatement functions considering production risk and farmers’ risk preferences. Different sources of risk are included by incorporating uncertainty of both attainable yield level and weed pressure. Results for a case study of silage maize cultivation in 377 municipalities in North Rhine-Westphalia, Germany, show that a glyphosate ban causes a shift towards more mechanical weed control, but not to more pronounced use of selective herbicides. The ban slightly reduces net profits and yields, leads to a significant reduction of the overall toxicity of pesticide use, but increases energy consumption of the agricultural system. The magnitude of these effects is found to be critically dependent on output price levels and yield expectations.

1. Introduction

The relicensing of glyphosate in European agriculture has received massive societal and political attention in recent years. Those advocating a ban used as arguments potential social costs in the form of human health risks (e.g. Guyton et al., 2015) and environmental effects, for instance negative effects on biodiversity through a decline in fodder plants for butterflies or through the accumulation of metabolites (e.g. Brower et al., 2012; Helander, Saloniemi and Saikkonen, 2012; see Tarazona et al., 2017, for an overview of the ongoing debate). In contrast, arguments for the continued use of glyphosate mainly relate to private and social economic benefits, such as lower production costs and consequently lower food prices. Others warn of potential trade-offs in the environmental and human health dimension when glyphosate is substituted with other forms of weed control, such as more intensive selective herbicide application or increased tillage intensity (see e.g. Williams, Kroes and Munro, 2000; Duke and Powles, 2008). Despite the fact that glyphosate was relicensed by the European Commission for 5 additional years at the end of 2017, the debate continues. Additionally, we observe that actors in the foods sector have started to demand glyphosate-free products from suppliers.¹ However, scientific information on trade-offs between environmental, human health and economic implications of a glyphosate ban are limited (cf. Finger, 2018).

We contribute to a more informed debate by focusing on key agronomic and economic aspects of a glyphosate ban in a state-wide case study for a major crop, based on a highly detailed bio-economic simulation model. More specifically, we test for the following potential consequences of a glyphosate ban: (i) yield losses, (ii) a higher tillage intensity and consequently changes in diesel consumption and thus energy efficiency and (iii) increased use of post-emergence herbicides with higher toxicity and thus stronger adverse effects on the environment than glyphosate. Our paper fills a gap in the literature as so far only limited scientific evidence on the consequences of a possible glyphosate ban is available (cf. Schulte and Theuvsen, 2015; Böcker, Britz and Finger, 2018a).

As real-world observations of weed control under a glyphosate ban are not available, we employ a normative modelling approach based on damage abatement functions (Karagiannis and Tzouvelekas, 2012). The methodology presented in this article contributes to the literature by combining state-contingent decisions on pesticide application, depending on weed pressure, with an expected utility (EU) framework, using a highly detailed and spatial explicit representation of weeds, weed control strategies and their economic and environmental implications. More specifically, we quantify optimal alternative weed control strategies (including mechanical and chemical pre- and post-sowing strategies) under a glyphosate ban, considering production risk and farmers’ risk preferences. To this end, we extend the model and analysis by Böcker, Britz and Finger (2018a) in two important directions. First, we account for different sources of risk by incorporating uncertainty with respect to attainable yield levels and weed pressure. Attainable yield levels are unknown at the time of pesticide application, so that the production system contains a high level of uncertainty. For example, in a dry year with low yield levels the benefit of a more intense weed control strategy is rather low because the yield reduction stems from water scarcity and not from weed competition for light or nutrients. By incorporating such risks and farmers’ risk preferences in our model, we therefore account for a major determinant of pesticide use decisions (Horowitz and Lichtenberg, 1993). Second, we assess the environmental implications of the chosen weed control strategies with different environmental indicators. Environmental consequences of a glyphosate ban are core arguments in the political debate, and should be considered. A pesticide risk indicator is used to evaluate adverse effects of pesticide application on human health and on the ecosystem. Here, we choose the very detailed Pesticide Load Indicator (PLI) that has been developed and applied in Denmark (Kudsk, Jørgensen and Ørum, 2018). Furthermore, we quantify the process energy demand of weed control strategies and thus account for all physical material flows. This allows assessing the environmental effects of potential increases in machinery use in response to a glyphosate ban. In summary, our approach provides a holistic assessment of potential effects of a glyphosate ban, integrating potential economic as well as the most important environmental effects. Revealing and quantifying possible trade-offs between different goal functions are crucial for a well-informed policy debate on glyphosate. The here-developed model may further serve as an important building stone for assessments concerning future debates on pesticides.

The model is applied to silage maize cultivation in the federal state of North-Rhine-Westphalia (NRW), Germany (Figure 1). Maize is one of the major crops in Germany, grown on 21 per cent of the arable land and 15 per cent of the agricultural area (2.5 million hectares [1 ha = 10,000 m²] in 2017; Statistisches Bundesamt, 2017). In our case study region NRW, maize is even the dominant crop with a share of almost 30 per cent of the arable land, mainly used for cattle feeding and for biogas production, either as silage maize or as grain maize or corn-cob-mix (Information und Technik Nordrhein-Westfalen, 2017). Glyphosate is mainly applied before sowing in cultivation of maize (Figure 2).² Between 22 and 35 per cent of the maize growing farmers have been found to apply glyphosate, either in combination with mechanical strategies, such as chisel ploughing, or with direct sowing of maize (Julius Kühn-Institut, 2017; Wiese et al., 2018). The remaining farmers use mechanical strategies without herbicide application before sowing, for example mouldboard ploughing or one or two passes of chisel ploughing and/or rotary harrowing. Direct sowing in combination with a glyphosate application is especially relevant on light soils where reduced traction need leads to lower costs of sowing (direct sowing is usually more expensive than conventional sowing). It is also found on heavy soils where direct sowing can have a cost advantage because of high traction needs for mechanical tillage strategies.³ After sowing, a large share of farmers apply selective herbicides. Mechanical strategies such as hoeing are poorly established.

Our paper is structured as follows: we first present the modelling approach based on an output damage function combined with EU maximisation. Moreover, we give some detail on the PLI and the energy process analysis and present the data used. Finally, we present and discuss results and conclude.

2. Methodology

The spatially explicit optimisation approach applied in this article draws on Böcker, Britz and Finger (2018a),⁴ where model construction and application are based on two major steps. First, the maize yield is estimated as a function of weed pressure and weed control. Net profits for each weed control strategy are calculated accounting for output prices and costs, which are also dependent on weed pressure. Second, the optimal pre- and post-sowing weed control strategies are selected. However, the approach of Böcker, Britz and Finger (2018a) is deterministic: it is assumed that farmers know for certain both the weed pressure and the realised yield. We expand the approach by accounting for the uncertainty associated with attainable yield distributions and weed pressure, which allows introducing a detailed stochastic production function and accounting for risk preferences of farmers. We differentiate weed pressure and weed control across space to analyse environmental, ecological and economic effects. This output damage control approach is applied to each of the 377 silage maize-producing regional units of NRW, m = 1, …, M (Figures 1 and 2).

2.1. Modelling structure

The production function consists of three parts. The first part accounts for the pre-sowing weed control activities. Secondly, $ỹm,χa$ is the attainable yield distribution for a regional unit m in year χ. The third part accounts for the post-sowing weed control. In the latter, $e−(β0+β1·zm,j)2$ is the yield loss based on different control strategies (denoted further as D), I is the percentage yield loss as D approaches 0, A is the percentage yield loss as D approaches infinity, $Ψ̃$ is the distribution of the time of maize emergence in relation to weed emergence (in growing degree days; this measure can be seen as an expression for the overall weed pressure in a certain year t) and C is the rate at which the yield loss I decreases as Ψ becomes larger.

Farmers face several uncertainties when deciding on an optimal weed control strategy (e.g. Auld, Menz and Tisdell, 1987): (i) the level of weed infestation, (ii) the effectiveness of the weed control strategy, (iii) prices, yield improvement and quality effects and (iv) reinvasion, spill-overs on own crop and time-interval effects from delays of receiving benefits. We here focus on (i) the level of weed infestation because differing yearly growing conditions, mainly climatic ones, lead to varying weed pressure. In contrast, other aspects are either not important for maize production in our case study (e.g. quality effects) or assumed to be less relevant because of the high level of information provided to farmers (e.g. effectiveness of weed control strategies; see also section 2.2.). Moreover, (v) there is uncertainty regarding the attainable yield level, which is of high importance for our application to silage maize. Weed control usually occurs relatively early in the growing season of maize and, accordingly, the same weed control can lead to different yield outcomes. To reflect this uncertainty, the attainable yield $ym,χa$ is introduced as a random variable in our model (see Figure 2). We do not account for price risks because silage maize output prices are not characterised by high volatility. This is the case, because the biogas boom in Germany has stimulated the widespread use of long-time supply contracts (Reise, Liebe and Mußhoff, 2012; Britz and Delzeit, 2013).

Thus, we introduce risk in our model by accounting for the stochasticity of weed pressure and attainable yield. The attainable yield in each year χ is assumed to be stochastic, e.g. due to stochastic weather conditions that are independent of weed pressure and weed control (e.g. Tembo et al., 2008). This yield variability is captured by an empirical yield distribution quantified for each regional unit m. Concerning weed pressure, farmers face uncertainty with respect to the time of weed emergence relative to maize, Ψ, which is a key indicator of weed-induced potential yield losses. Herbicide strategies are chosen and applied early in the growing season of maize based on the observed weed pressure. If maize has a growth advantage over weeds in a certain year, cheaper or no herbicide strategies might be favoured. In contrast, strategies with higher efficacy that are often also more complex and expensive might be more promising if weeds have larger growth advantages over maize. An overview of the methodology is also given in Figure 2.

2.2. Goal function

The simultaneous consideration of the years t for which observations on weed pressure are available allows to reflect that farmers aim to avoid resistances of weeds against herbicides. Specifically, it is assumed that farmers need to change the active substances they use. More precisely, the strategies were classified into groups according to the Herbicide Resistance Action Committee and a constraint was added to prevent control strategies from the same group being used in 2 consecutive years. Furthermore, strategies containing the active substance nicosulfuron are only allowed to be applied every second year.

To address the research questions, we compare a baseline scenario, in which glyphosate is licensed, to a counterfactual scenario, in which glyphosate is banned. We report in the main body of the paper results for slightly risk averse behaviour with r_a- = 0.5, reflecting recent empirical evidence for German farmers (Maart-Noelck and Musshoff, 2014; Meraner and Finger, 2017). We conduct additional sensitivity analyses with respect to the partial risk aversion coefficient, considering values of −2.0, 0.0 and 0.8, which reflect risk loving, risk neutral and more risk averse preferences (results presented in the Appendix in supplementary data at ERAE online). This is relevant as farmers are found to be on average slightly risk averse, but at the same time a large heterogeneity in the population exists (ibid.). Furthermore, we assume expected output prices for silage maize of €4.00, 4.60 and 5.20/dt (dt denotes a deciton, i.e. 100 kg), reflecting the range of currently observed silage maize prices. It is assumed that harvesting and ensiling are done by the selling farmer. In addition to the four levels of risk aversion, we also include sensitivity analysis with regard to the expected attainable yield by considering a 10 per cent higher attainable yield level because yield expectations may increase in the future due to new varieties and/or better fertilisation (the results of this analysis can be found in the Appendix in supplementary data at ERAE online). In total, we consider therefore 4 (risk aversion) × 3 (maize price) × 2 (attainable yield) = 24 different scenarios. For each scenario, we check for significant differences in net profits and environmental impacts at the municipality level (see below).

2.3. Pesticide load analysis

Values for each sub-indicator are computed from a broad range of potential effects on the environment and human health. More specifically, Λ_toxy assesses short-term effects on eight different families of animals and plants (birds, mammals, fish, earthworms, bees, daphnia, aquatic plants and algae). Additional long-term effects are taken into account for fish, earthworms and daphnia. Λ_fate considers biodegradability, bioaccumulation and mobility in soil. Λ_heal is calculated based on Hazard- and Risk Statements with regard to human health of the specific substances as well as product formulation. For a specific pesticide product, the load per kilogram or litre is calculated based on the load of each single active substance and its concentration in the product. For details, see Kudsk, Jørgensen and Ørum (2018). PLI values for the used products are presented in detail in the data section.

2.4. Energy process analysis

In order to assess the energy use related to a specific weed control strategy, we employ the methodology and definitions of Jones (1989) and of Hülsbergen et al. (2001) (for a recent application see Jankowski, Budzyński and Kijewski, 2015). The aim of this approach is to ‘trace all the energy inputs into an agricultural system, based on physical material flows’ (Hülsbergen et al. (2001: 306f.), excluding energy flows from human labour and solar energy (Uhlin, 1999). Direct energy input (E_d) refers in our case study to the consumption of diesel whereas indirect energy input E_i quantifies the energy needed to produce the different inputs: seed, mineral fertilisers (we treat manure as waste from livestock production, i.e. assigning zero energy content), pesticides and machinery. The overall energy input E is equal to E_d+ E_i. The energy output EO is equal to the energy content of the harvested maize minus the inherent energy in seed (which is lower than the energy needed to produce the seed). The net energy output NEO is equal to EO – E. All energy values are given in calorific values [MJ/ha].⁷

2.5. Hypothesis testing

With respect to the applied herbicides in case of a glyphosate ban, we test for differences in (i) weed control costs, (ii) yields and (iii) the expected net profit. Furthermore, we test for potential decreases of pesticide load in the categories (iv) toxicity, (v) environmental fate, (vi) human health and (vii) overall pesticide load (sum of all load indicators, see data section). In the energy process analysis, we test the hypotheses that (viii) the energy output EO decreases (i.e. the yield decreases), (ix) the net energy output NEO decreases, (x) more direct energy is used (E_d increases), (xi) more indirect energy is used (E_i increases), (xii) more energy is used in general (E increases) and finally (xiii) the energy efficiency decreases (EO/E). Wilcoxon–Mann–Whitney tests are used to test our hypotheses on differences between regional unit averages of results over all years t.

3. Data

The first part of the data section gives an overview of the most important data sources of the above presented model. The focus lies on weed control strategies, weed spread and yield data. The subsequent two sections present the data underlying the application of the PLI and the energy balance.

3.1. Weed data, weed control strategies and yield data

The model consists of m = 1, …, 377 regional units, which represent the maize-producing municipalities of NRW (Figure 1). The complete data sources on weed spread, weed abundance, yield losses and herbicide efficacy are documented in Böcker, Britz and Finger (2018a, 2018b). We account spatially explicitly for the influence of i = 1, …, 32 (grass-)weeds on yield in the model. Each regional unit has a certain probability that a specific weed occurs. These data are taken from the FloraWeb database, an open GIS-based platform (NetPhyD and BfN, 2013). In our model, we account for the heterogeneity of weed pressure across space and time. However, we do not account for inter-annual or -regional dynamics of weed abundance because geographical position and soil conditions have been found to be more important for the composition of weed species than management factors (Lundkvist et al., 2008; Hanzlik and Gerowitt, 2011). In NRW, small to medium size fields are present. Weed seed import is therefore likely, for instance, by wind, unsprayed field edges or machinery, which further motivates refraining from modelling explicitly weed dynamics. This would require the consideration of all crops in which any of the 32 considered weeds could occur.

For pre-sowing weed control, g = 1, …, 19 strategies are considered and for post-sowing h = 1, …, 55 (see the Appendix in supplementary data at ERAE online). This selection includes both the currently dominating strategies and strategies that are currently not yet economically viable but might become relevant under a glyphosate ban. Pre-sowing strategies consist of different combinations of glyphosate application, mouldboard ploughing and non-inverting strategies such as chisel ploughing/rotary harrowing. Post-sowing strategies consist of selective herbicide application (once or twice) and/or mechanical strategies such as harrowing or hoeing. The costs for herbicide application and for machinery are treated as deterministic in the model because farmers know input prices in the moment the weed control decisions are made (see Table A2 in the Appendix in supplementary data at ERAE online).

Whereas Böcker, Britz and Finger (2018a) conducted a sensitivity analysis on weed pressure, we now incorporate a distribution of weed pressure in the model. We thus focus on the distribution of the value describing the difference between the emergence of maize and weeds in a specific year and regional unit, i.e. we consider if and to what degree weeds have a growth advantage over maize. Ψ can be positive or negative, depending on the time of maize and weed emergence (see also Figure 2). Here, we consider the period t = 2006–2015 with yearly, changing values of Ψ depending on the regional unit m. In order to determine this distribution, we make use of the growing degree-day (GDD) concept (McMaster and Wilhelm, 1997) and use spatially and temporally specific information on weather (temperature) and phenology (starting dates of sowing and emergence) data of silage maize⁸ for our study region. The weather data are provided by the German Weather Service (Deutscher Wetterdienst) from six weather stations in NRW and we assigned each regional unit to the closest weather station.⁹ The detailed assessment of the Ψ-values can be found in the Appendix in supplementary data at ERAE online.

With regard to the expected attainable yield level, we make use of raster data (1 × 1 km) on water-limited potential yields of silage maize for χ = 1999, …, 2011 that were gratefully provided by Ganga Ram Maharjan and Thomas Gaiser from the Crop Science Group of University of Bonn. The raster data were created by a crop model that is presented and documented in Hoffmann et al. (2015) and Zhao et al. (2015). This raster data were aggregated to municipality levels.¹⁰ Water-limited yields are chosen, as irrigation is irrelevant in silage maize production in the region.

The parameters C and I in equation (2) are taken from Bosnic and Swanton (1997). A is defined as 63.8 per cent, which was found to be the maximum potential yield loss in field trials (Böcker, Britz and Finger, 2018a). The key parameters of the production function in equation (2), α₀, α₁, β₀ and β₁, are estimated by determining those parameter values that minimise the error term between the observed yields and the yields simulated with the control strategies used in current silage maize production (more details can be found in Böcker, Britz and Finger, 2018a: 184ff., 2018b). Expert knowledge from the Chamber of Agriculture of NRW was used in order to get information about the currently used practices of maize cultivation (ibid.). In addition, recent Ψ-values were included for the time period 2013–2015. The finally estimated parameter values are α₀ = 1.266, α₁ = 0.683, β₀ = 0.747 and β₁ = 0.543.

3.2. Pesticide load

Information about ASs to calculate the PLI is taken from the Pesticide Properties DataBase (Lewis et al., 2016), which draws on publicly available sources, such as pesticide admission and regulation procedures. In addition, we obtain complementary information about ASs per product as well as their specific concentration from product specification sheets of the herbicide manufacturers and from herbicide recommendations, such as from the Chamber of Agriculture NRW (Landwirtschaftskammer Nordrhein-Westfalen, 2015) and the Bavarian State Research Centre for Agriculture (Bayerische Landesanstalt für Landwirtschaft, 2016). The overall load and the load in the three sub-categories can be found in Table 1 for all herbicide products included in our analysis. Note that the selection and weights of the indicators and the indices underlying the calculation of the load values in the PLI reflect the focus on environmental problems and preferences in Denmark (see the Appendix in supplementary data at ERAE online for further details). The PLI covers a broad range of environmental and health effects on a product level. It is aligned with European pesticide regulations, is already implemented, has been scientifically tested as a pesticide risk indicator in Denmark for several years (Kudsk, Jørgensen and Ørum, 2018) and can be easily applied to other countries (e.g. Möhring, Gaba and Finger, 2019). Thus, its application to the German study region seems reasonable, especially as no comparable indicator is available for Germany.

3.3. Energy balance

Process analysis based on energy balances is a widespread method in agricultural sciences (e.g. Deike, Pallutt and Christen, 2008; Jayasundara et al., 2014), typically drawing on literature providing general information on energy use in agricultural and industrial processes (e.g. Green, 1987; Hülsbergen et al., 2001; Audsley et al., 2009). For indirect energy consumption of producing and maintaining machinery, Aguilera et al. (2017: 341) report for the year 2010 values of E_tr = 156 MJ/kg of machinery for tractors, E_hr = 102 MJ/kg for harvesters, E_tm = 72 MJ/kg for tillage machinery and E_om = 62 MJ/kg for other machinery. The gross calorific values of the different production inputs and steps are presented in Table A3 (direct energy E_d) and Table A4 (material use for indirect energy E_i) in the Appendix in supplementary data at ERAE online.

Following Jayasundara et al. 2014: 83 (citing Nagy, 1999), we add 6 per cent to the estimated production requirements to account for formulation, packaging and delivery requirements. The assumed energy requirements used in our analysis are presented in Table 2.

4. Results

Results are presented as averages over the range of simulation years, T, and over the regional units/municipalities, M. In the main text, we consider the scenario with moderate risk aversion (r_a = 0.5) and the three output price levels of €4.00, 4.60 and 5.20/dt. The sensitivity analyses with respect to different levels of the partial risk aversion coefficient and a potential increase of the attainable yield level by 10 per cent are presented in the Appendix in supplementary data at ERAE online. This section first illustrates descriptive results on the weed control strategies used. In the following, we present the main results on the trade-offs between economic (yield, net profit) and environmental consequences (herbicide load and energy process analysis) of a glyphosate ban.

4.1. Descriptive results

Three characteristics mainly influence the choice of pre- and post-sowing weed control strategies in our model: (i) the distribution of the net profit (determined by the output price and depending on the distribution of the attainable yield), (ii) the weed occurrence and pressure in a certain region and year and (iii) the soil type influencing the weed control costs. These mechanisms will be described in detail below.

In the base scenario without policy intervention, we find that applying glyphosate is the utility maximising strategy in about 28.5 to 38 per cent of the regional units, depending on the output price (Figure A2 in the Appendix in supplementary data at ERAE online). This number is in line with surveys on how many farmers apply glyphosate in Germany (see introduction). A higher price of silage maize, Ceteris paribus, increases glyphosate application. The major alternative to pre-sowing weed control using glyphosate is mechanical weed control based on two passes of chisel ploughing (or similar machines). Other strategies, such as one pass of chisel ploughing, chisel ploughing in combination with harrowing or conventional tillage with ploughing are less frequent; mouldboard ploughing is not found to be optimal in any scenario. The level of risk aversion only has a minor and insignificant impact on the choice of the pre-sowing and post-sowing weed control strategies.¹¹ This does not mean that the incorporation of risk effects in our model is obsolete, as the consideration of stochastic weed pressure and attainable yield levels have effects on the optimal decision-making of farmers compared to a deterministic approach.

The average number of applied active substances per ha per year [AS/ha] excluding glyphosate is presented in Table 3 and in Figures A4 and A5 in the Appendix in supplementary data at ERAE online. At the low output price level of €4.00/dt 2.6 AS/ha and year are applied. At the higher output price level of €5.20/dt, the number increases to 3.0–3.1 AS/ha and year on average. At the low price level, nicosulfuron, profulfuron and pyridate have a relatively high share of the applied active substances. Terbuthylazine, S-metolachlor, flufenacet, iodosulfuron, foramsulfuron and thiencarbazone gain in importance at higher expected revenues. The model results show that the most frequently applied active substance is terbuthylazine, which is in line with observations on farm practices in the case study region (Julius Kühn-Institut, 2017). The use of different active substances is relatively constant over the four different risk aversion levels, with slightly but not significantly lower shares of active substances if a risk affine decision maker is assumed (r_a = −2.0). In case of a glyphosate ban, the composition of the active substances does not change significantly (Table 3).

4.2. Economic consequences

We find that expected revenues (i.e. expected output price levels and the expected attainable yield level) are the main drivers of weed control expenditures and the choice of applied active substances (Figure 3, Table 3 and Figures A2–A5 in the Appendix in supplementary data at ERAE online). More specifically, the average costs of the optimal weed control strategies increase with higher expected revenues. For example, at lower levels of expected revenues, cheap mechanical weed control that has a low damage control is more frequently used compared to glyphosate, even in the baseline scenario without a ban. This is due to lower sowing costs after two passes of chisel ploughing compared to glyphosate application and direct sowing. If glyphosate would be banned, costs for herbicides and herbicide application would decrease significantly at all three output price levels, but we observe a significant but relatively small increase in total weed control costs regardless of the output price. Note that this mainly results from higher costs for mechanical weed control (significant at the 0.01 level), which is the main substitute for glyphosate. The enforced change in weed control also impacts the yield level, it would therefore be possible that both revenues and costs increase, but clearly not their difference because such a choice would have already constituted an optimum in the reference scenario. In most geographical units, we find a reduction of the yield and consequently of the turnover. This yield reduction is described in the energy output EO in Section 4.4 in detail. The average expected net profit varies between €555/ha €981/ha depending on the output price (without direct payments, Figure 4). On average, this is reduced by €2–3/ha, but in single regional units higher losses also occur of about €10/ha. The reduction is small and statistically insignificant (Figure 4).

4.3. Pesticide load analysis

The frequency distribution of pesticide load indicator values for the optimal strategies (averages per municipality) is illustrated in Figure 5. The figure shows the results for moderate risk aversion (r_a = 0.5) and the glyphosate licensed-scenario. Pesticide load values differ strongly between the regional units and range from about 0.5–3.8 load/ha. Differences are especially large for the environmental fate load, which is due to high load values of herbicides containing prosulfuron (e.g. contained in the product ‘Peak’). Load values increase in output prices and expected attainable yield levels. The levels are higher than the average herbicide load for a hectare of maize in Denmark (where the PLI is used in policy analysis and documented), which could be due to higher taxation of products with high load values (Böcker and Finger, 2016).¹²

The analysis of the four different hypotheses regarding the pesticide load is presented in Table 4 for r_a = 0.5 and in Table A5 (Appendix in supplementary data at ERAE online) for all risk aversion coefficients. We find significant load reductions under a glyphosate ban in all scenarios because chemical weed control is substituted by mechanical control. This holds for the total load indicator, as well as all sub-indicators (i.e. environmental toxicity, fate and human health). The decrease is strongest with respect to the human health load. The sub-indicators for environmental fate and toxicity show lower reductions, reflecting the low environmental load of glyphosate-based products (Table 3).

The spatial distribution of the total potential load reductions in case of a glyphosate ban is presented in Figure 6 (see Figure A6 in Appendix in supplementary data at ERAE online for the energy consumption plotted against the pesticide load). The load reduction is highest in two types of regions, in which glyphosate is more likely to be used. These are firstly regions with heavy soils (i.e. high shares of clay) where mechanical alternatives to glyphosate require higher traction power and thus are more expensive. Here, glyphosate application is, under current conditions, often the optimal weed control strategy. Secondly, high load reductions are found in regions dominated by high shares of sandy soils where direct tillage or strip-till practices in combination with glyphosate application are relatively cheap. In contrast, on medium soils (i.e. with balanced mixtures of clay, silt and sand), glyphosate is applied less frequently in our model. Changes in the expected output price level have small and insignificant impacts on the pesticide load reduction per hectare. However, since we find that glyphosate application is optimal in more regional units when maize price levels are higher, higher prices would also imply more widespread load reductions under a ban.

4.4. Energy process analysis

The frequency distribution and the mean values of the energy output and input are presented in Figure 7 for a risk aversion coefficient of 0.5 in the glyphosate licensed-scenario. The expected energy output EO across the regional units ranges from 141 GJ/ha to 278 GJ/ha, the energy input ranges from 2.7 to 12.2 GJ/ha. For higher output prices, we find higher energy outputs because of increased weed control and consequently higher energy inputs, but also due to the fact that higher maize prices incentivise the use of other yield increasing inputs such as fertiliser. Reflecting slightly decreasing marginal returns, the energy efficiency is highest in the low price-scenario with a ratio of 26.2 compared to 25.3 in the high price-scenario (Figure 7). Similar trends can be observed under other risk aversion coefficients and higher expected attainable yield (not shown).

Table 5 presents the differences in results for the energy indicators between the glyphosate licensed and the glyphosate banned scenarios for three different output price levels. Regarding the energy output EO, we observe in most scenarios only a small reduction by about 650–850 MJ/ha due to lower yields (around − 1 dt/ha). However, this difference in EO is overall not significant. Similar results are found for net energy output NEO. Direct energy use E_d increases significantly by about 300 MJ/ha due to a glyphosate ban. Indirect energy consumption E_i tends to decrease, but differences are not significant. The change in the total energy use (ΔE) is positive, ranging from +91 to +167 MJ/ha. Finally, we find that a glyphosate ban decreases energy efficiency (ΔEO/E), reflecting that mostly mechanical weed control is used as a substitute. This decrease is small and insignificant for the low output price level of €4.00/dt, but larger and highly significant for the higher price levels of €4.60/dt and €5.20/dt.

Figure 8 includes the maps of the increase in direct energy consumption (E_d) and the reduction in energy efficiency (EO/E) due to a glyphosate ban (see Figure A6 in Appendix in supplementary data at ERAE online for the scatterplot). The emerging picture resembles the one for the PLI. We find a relatively strong increase on heavy soils, where mechanical strategies require high energy input, but also an increase on light soils due to current application of no till or strip-till practices in combination with glyphosate. The trends are similar in most of the regional units. Energy efficiency gains after a ban can only be observed in very few locations.

5. Discussion

We find that a glyphosate ban would lead to a significant, but relatively small increase in weed control costs, independently of expected output prices, stemming mostly from increases in more expensive mechanical weed control measures, which outweigh lower herbicide expenditures. Furthermore, we find reduced yields such that a glyphosate ban leads to lower net profits, on average by €2–3/ha and in some cases by up to €10/ha. The small change also reflects cost savings in sowing when switching from direct sowing – used mainly in combination with glyphosate – to mulch sowing. This income loss is lower than what was found in other studies. For example, Kehlenbeck et al. (2015) found potential losses of €−88 to +22/ha, depending on the scenario, but they also included unprofitable strategies in their calculations. Of course, results depend strongly on cost assumptions, which reflect the assumed machinery and plot size. Larger machinery and plot sizes reduce the per-hectare-costs of weed control and tend to increase the attractiveness of mechanical weed control compared to chemical control. Moreover, reduced herbicide prices decrease the usage of mechanical strategies. Nevertheless, the fact that under current conditions only one third of the farmers apply glyphosate in silage maize production in Germany (Julius Kühn-Institut, 2017), a share close to what is simulated by our model under current conditions, supports the result of our analysis that mechanical strategies are a feasible alternative to glyphosate application in our case study.

Our results thus indicate no large economic implications of a glyphosate ban in silage maize production in North Rhine-Westphalia. However, our findings point to a trade-off between reducing adverse effects of pesticides on human health and the environment, and on energy consumption as a driver of climate change.

In our case study, we find an average increase of total energy consumption of up to 170 MJ/ha and year, but the change depends strongly on expected output prices. If glyphosate is assumed to be used on 66,500 ha of silage maize in NRW (~190.000 ha silage maize according to Information und Technik Nordrhein-Westfalen, 2017, and 35 per cent of farmers applying glyphosate; Julius Kühn-Institut, 2017), our analysis indicates that the energy demand of silage maize cultivation would increase by 11 TJ in case of a glyphosate ban. Relative to the total final energy consumption of agriculture in the state of NRW (IWR, 2018), this would indicate an increase in energy demand of 0.03 per cent (assuming that 2.2 per cent of the total final energy consumption is consumed by agriculture; Eurostat, 2018). The climate change impact of this additional energy consumption depends on energy sourcing and production. Indirect energy, i.e. for factor production, is consumed for the most part in form of electricity; direct energy relates to diesel use. The model also simulates a loss in energy efficiency if glyphosate would be banned, also due to somewhat lower yields. This means that more silage maize will have to be cultivated in case of a glyphosate ban to meet the demand.

We find a significant decrease in the overall toxicity, expressed in pesticide load indicator units, of on average 0.2 load units (11 per cent) per hectare. Yet, pesticide load levels (per ha) in maize production before and after the ban are low compared with load levels in other crops such as potatoes or vegetables.¹³

The level of risk aversion only has a small influence on the choice of weed control strategies. Note that this does not imply that the introduction of risks in our model was not important for outcomes. It rather reflects the importance of observing the level of weed infestation on the field. Farmers are thus able to apply stage-contingent weed control. This finding is also in line with the ambiguous findings on the risk effects of pesticides in the literature. A major source of uncertainty is the level of the attainable yield that leads to a stochastic marginal value product of weed control. This can even create incentives for risk-averse decision makers to use less than profit maximising levels of herbicides (e.g. Horowitz and Lichtenberg, 1993). Alternatives to the EU framework used here could provide additional insights. For example, Carpentier (2017) recently applied prospect theory to pesticide application to explicitly consider farmer’s reference situation to the protected or the unprotected crop.

Our results relate to short- to mid-term and not long-term changes in weed control strategies in silage maize cultivation under a glyphosate ban. Furthermore, extensive margin and farm effects such as changes in crop rotation or other adjustments at a farm level have not been not included so far, but could be addressed in future research. However, extensive margin and farm effects of a glyphosate ban in cultivation of maize in our study area are likely to be quite limited for three major reasons. Firstly, the simulated changes in net profits are quite small and glyphosate use in other important crops is even lower than in maize, rendering significant adjustments in cropping patterns under a ban unlikely. Secondly, conservation tillage, i.e. non-inverting tillage, without applying glyphosate is feasible in silage maize production. Due to high investment cost for specialised machinery, a problem in increased conservation agriculture without glyphosate could arise if the necessary machinery is neither owned by the farmer, nor contractors can be found. In our case study region, however, the density of contractors is quite high, so the problem should not be a major concern. Thirdly, total maize production is unlikely to change much as policy measures promoting biogas production from silage maize generate a stable and high demand (Information und Technik Nordrhein-Westfalen, 2017). The slight yield reduction on areas that currently apply glyphosate would decrease feedstock availability for the existing biogas plants, facing large sunk cost – and thus further intensify the recently observed price increases for silage maize. That would oppose the incentive to reduce the maize acreage in response to the slightly decreased net profit under the ban. Although the short-term economic consequences would be rather small, further research should address possible unwanted ecological and socio-economic side-effects, e.g. via changes in land use or trade-patterns. Stricter regulatory measures or economic incentives to reduce glyphosate application are for this reason more suitable than bans. Our results provide a quantitative basis for political debates on glyphosate. They show clear trade-offs in decision-making and reveal the need for flexible and tailor-made agri-environmental policy solutions. This could, for example, comprise solutions such as pesticide taxation (Finger et al., 2017). Pesticide taxation incentivises those farmers to reduce glyphosate use, which are able to substitute glyphosate with little net profit reduction, while still enabling glyphosate use for those facing higher reductions. Similarly, farmers who need large amounts of energy to substitute glyphosate use should be able to further apply glyphosate, while those who face lower increases in energy use should be incentivised to reduce glyphosate applications.

Regarding the weed control implementation in the model, additional temporal dimensions could be introduced. More specifically, temporal interdependencies of applied weed control strategies could be considered, e.g. if weed pressure spills over different periods. For example, if only conservation tillage strategies without glyphosate application are chosen, the size of the weed seed bank grows (Bàrberi et al., 1998). In addition, it could be interesting to include a behavioural algorithm relating to the choice of weed control strategies (cf. Hüllermeier, 2005) and farmers' experiences. Furthermore, other environmental dimensions could be included in the model. For example, no till practices – possibly economically viable only if glyphosate is licensed – reduce soil erosion, especially on hilly grounds (Montgomery, 2007). Moreover, we do not take into account that a change to mechanical weed control maybe has negative effects on ecological soil conditions. Along these lines, problems with nitrogen surpluses might increase due to slightly lower yield levels if glyphosate is banned and manure application is not adjusted. New technologies could be a game changer for the future role of pesticides in agriculture (e.g. Finger, 2018). For example, autonomous weeding robots might allow to reduce pesticide use dramatically without facing trade-offs of high emissions or soil compaction (e.g. Walter et al., 2017). Thus, future modelling attempts shall also incorporate new technological options.

6. Conclusions

We develop and employ a detailed bio-economic model focusing on weed damage control in order to analyse the potential effects of a glyphosate ban. The model simulates the optimal choice from a set of pre- and post-sowing weed control strategies. Production risks and farmers’ risk preferences are considered in an EU framework. The main findings for a case study for silage maize production in 377 municipalities in North Rhine-Westphalia, Germany, are that a glyphosate ban would cause a shift towards more mechanical weed control measures, but not to a more pronounced use of selective herbicides. Economic impacts on maize yields and net profits are small. The weed control strategy set chosen in response to a glyphosate ban is less toxic (as measured using the Pesticide Load Indicator) but more energy intensive (based on a detailed energy process analysis). The magnitude of these effects is found to be critically dependent on output price levels and yield expectations. Thirteen different hypotheses are tested with regard to a glyphosate ban (see Table 6 for summary results). Our analysis thus quantifies in detail trade-offs between different policy goals and can inform policy debates, as well as regional and private initiatives for alternatives to glyphosate use. Furthermore, the modelling approach developed here could be an important starting point for the assessment of economic and environmental trade-offs in future debates on restrictions of pesticide use.

Acknowledgements

We thank Ganga Ram Maharjan and Thomas Gaiser from the Crop Science Group of University of Bonn for providing yield data for this research. Furthermore, we thank the plant protection consultants of the Chamber of Agriculture of NRW and of the Chamber of Agriculture of Lower Saxony for providing valuable information supporting this research.

Footnotes

Review coordinated by Ada Wossink

References