METHODOLOGICAL SCHEME FOR DESIGNING THE MONITORING ...

Environmental Monitoring and Assessment (2005) 111: 1–26 DOI: 10.1007/s10661-005-8044-5

c Springer 2005

METHODOLOGICAL SCHEME FOR DESIGNING THE MONITORING OF GENETICALLY MODIFIED CROPS AT THE REGIONAL SCALE 2 ¨ F. GRAEF1,2,∗ , W. ZUGHART , B. HOMMEL3 , U. HEINRICH1 , 1 U. STACHOW and A. WERNER1 1

Centre for Agricultural Landscape- and Land Use Research (ZALF), Department for Land Use Systems and Landscape Ecology, Eberswalder Str. 84, D-15374 Müncheberg 2 Federal Agency for Nature Protection (BfN), D-14191 Bonn 3 Federal Biological Research Centre for Agriculture and Forestry (BBA), D-14532 Kleinmachnow (∗ author for correspondence, e-mail: [email protected])

(Received 9 August 2004; accepted 22 December 2004)

Abstract. According to EC regulations the deliberate release of genetically modified (GM) crops into the agro-environment needs to be accompanied by environmental monitoring to detect potential adverse effects, e.g. unacceptable levels of gene flow from GM to non-GM crops, or adverse effects on single species or species groups thus reducing biodiversity. There is, however, considerable scientific and public debate on how GM crops should be monitored with sufficient accuracy, discussing questions of potential adverse effects, agro-environmental variables or indicators to be monitored and respective detection methods; Another basic component, the appropriate number and location of monitoring sites, is hardly considered. Currently, no consistent GM crop monitoring approach combines these components systematically. This study focuses on and integrates spatial agro-environmental aspects at a landscape level in order to design monitoring networks. Based on examples of environmental variables associated with the cropping of Bt-Maize (Zea maize L.), herbicide-tolerant (HT) winter oilseed rape (Brassica napus L.), HT sugar beet (Beta vulgaris L.), and starch-modified potato (Solanum tuberosum L.), we develop a transferable framework and assessment scheme that comprises anticipated adverse environmental effects, variables to be measured and monitoring methods. These we integrate with a rule-based GIS (geographic information system) analysis, applying widely available spatial area and point information from existing environmental networks. This is used to develop scenarios with optimised regional GM crop monitoring networks. Keywords: assessment procedure, GIS, GM crops, spatial monitoring design

1. Introduction After the lifting of the de facto moratorium on the approval of genetically modified organisms (GMOs) by the EC in 2003 new GM crops are likely to be cultivated in large areas of the EC. To identify their potential adverse effects the regulatory framework 2001/18/EC of the EC (2001) specifies a step-by-step approval process for the deliberate release of GMOs into the environment. For market-approved GMOs it regulates (a) a case-specific monitoring that focuses on potential effects on human health and the environment, identified in the preceding risk assessment process and limited to a maximum time period of ten years in which to obtain

2

F. GRAEF ET AL.

results and (b) a general surveillance, which aims to identify and record indirect, delayed and/or cumulative adverse effects that have not been anticipated in the risk assessment and that should be carried out possibly over a longer time period and a wider area. Some countries have regulatory systems for GMO management in place (Kinderlerer, 2003). However, there is little guidance and scientific background information on how GM crop monitoring should be put into practice. Information and practical examples are given by EC (2003), Mellon and Rissler (1995) and Perry et al. (2003). The concerns about potential adverse effects of GM crops or their use on the agroecosystem and the environment include gene escape by outcrossing and hybridisation with non-GM crops and wild relatives, increasing crop and weed management efforts (Wolfenbarger and Phifer, 2000), modified fitness parameters and invasiveness of crops (Snow, 2003), effects on decomposers and soil organisms (Griffiths et al., 2000; Saxena et al., 1999), enhanced mortality of non-target organisms such as phytophagous arthropods, e.g. with Bt maize (Pleasants et al., 2001), tri-trophic interaction (Birch et al., 1999; Schuler et al., 1999) and decreased countryside biodiversity due to altering of current management regimes (Hails, 2002; Werner et al., 2000). Herbicide-tolerant crops, for instance, are expected to allow more efficient weed control, leading to fewer surviving flowering plants to provide food for various organisms along the food chain (Firbank et al., 2003; Krebs et al., 1999). Monitoring adverse GM crop effects requires a highly complex approach because (a) different spatial and temporal scales (Mellon and Rissler, 1995) and (b) different land use and cultivation systems are concerned (Crawley et al., 1993; Firbank et al., 2003), (c) potential long-term GMO effects e.g. on farmland biodiversity have not been fully investigated (Perry et al., 2003) and (d) unforeseen effects may surface as the frequency and scale of GM crop cultivation increases. Consequences of wide-scale cultivation cannot be drawn from limited-scale studies because on the field scale many effects turn out to be negligible (Pleasants et al., 2001; Sears et al., 2001). Assessing the impact of GM crops is also challenging, because it is important to have a baseline against which to compare the environmental effect of GM crops. Because GM crop monitoring is preceded by one kind of ecological risk assessment, typical components are to be included, e.g. integrating available information on exposure, sensitivity, adaptive capacity and adverse ecological effects (EPA, 1998; Turner et al. 2003). A critical issue is to define the monitoring scope. This includes identifying potential hazards to parts of the environment and to define, which environmental values or standards for instance biodiversity or ecosystem functions are to be protected (EPA, 1998; Müller, 2001). These standards are ultimately axiological statements on how ecosystem functions are valued. Functional environmental standards are not yet indicated, e.g. by the framework 2001/18/EC, so they may be derived from other sources such as national environmental laws, if existing. The GM crop monitoring procedure generally includes a number of steps (Figure 1): (1) characterisation of the market-ready GM crop from release-related

METHODOLOGICAL SCHEME FOR DESIGNING THE MONITORING

3

Figure 1. Stepwise GMO monitoring and assessment approach (Umweltbundesamt, 2001, modified).

risk research information; (2) information on potential adverse environmental effects. Both (1) and (2) are part of the environmental risk assessments prior to environmental release; (3) selection of indicators for anticipated adverse effects, in the following referred to as variables. The choice of variables to be monitored should be scientifically based, in particular depending on their exposure and on their indicator value. Various research groups, e.g. Firbank et al. (2003) and Züghart and Breckling (2003), have proposed concepts for selecting monitoring variables for potential agro-environmental effects of GM plants; (4) analytical methods by which these can be measured and evaluated; (5) adapted spatial design of a GM crop monitoring network. Once GM crop monitoring is carried out, (6) measurement data are produced and analysed. From the results (7) subsequent risk (re)assessments are carried out to derive (8) decisions on risk containment, approval or refusal of the GM plant and to possibly adapt the monitoring methodology (Figure 1). This paper focuses on step (5), however, the methodological approach requires integrated information from the preceding steps. Our aim and scope is to develop a systematic methodology that can be used to design a GM crop monitoring network on the regional scale. Our work is based to some extent on known methodologies and experience, e.g. from long-term environmental monitoring programs (Brandt et al., 2002; Hoffmann, 1998), GMO risk assessments (Eastham and Sweet, 2002) and sampling designs (Leigh and Johnston, 1994; Perry et al., 2003). Certain aspects, however, are new, such as potential consequences of – again potential – horizontal

4

F. GRAEF ET AL.

TABLE I Environmental standards, potential adverse environmental effects and examples of variables for monitoring of Bt-Maize in Brandenburg (Germany) (no distinction between case specific monitoring or general surveillance) Environmental Potential adverse standards concerned1 environmental effects2 a, d b, c, e, f

Bt toxin persistence Shift in trophy levels and diversity

b, c d, c, d d, e

Examples of variables to measure2 Bt toxin in soil Species diversity and frequency of herbivore arthropods and specific antagonists Species diversity and frequency of butterflies close to the cropped fields3 Saprophages in the soil (dipterous larvae) Microbial basal respiration

Change in biomass decomposition Change in soil communities Microbial sum indicators Consequences for best Change of cropping techniques management practice Change in animal pests Resistance of corn borer (Ostrinia nubilalis)

b, c, d e, f e, f

1 General environmental standards: GMO persistence and/or invasion (a), ecological interactions (b), biodiversity (c), soil functions (d), sustainable agricultural practice (e), plant health (f). 2 Bt-Maize requires also differentiation for specific transformation events with different levels of toxin exprimation (Schmitz et al., 2003). 3 Bt-toxin effects on butterflies close to cropped fields through pollen movement.

gene transfer and the potential persistence of modified genes in different organisms and media (Nielsen et al., 2001). In monitoring potential GM crop effects we have to consider a range of unevenly spatially and temporally distributed ecological variables (e.g. Table I). We therefore integrate spatial data into the process of designing a monitoring network. We use available spatial information that should be integrated in order to establish a representative and reliable GM crop monitoring network. We also combine existing environmental networks, which provide useful baseline data over years and different sites. For a systematic and convenient spatial data management with scenario-driven decision support, we use a rule-based GIS analysis as spatial planning tool. We develop a transferable framework and assessment scheme based on examples of environmental effects and variables to be measured, which may also be used in other parts of the EU. 2. Methods 2.1. S TUDY

AREA

Our study focuses on the State of Brandenburg (29,500 km2 surface), located in the deciduous forest zone of Central Europe in the lowlands of Northeast Germany


5

around Berlin. The region has a subcontinental to subatlantic climate and was formed by the last glaciation, is undulated, and dominated by sandy or slightly loamy soils (Marcinek and Zaumseil, 1993). Mean precipitation is between 490 and 610 mm (1961–1990) and mean annual temperature is 7–8 ◦ C. Different forest communities predominate as potential natural vegetation. Land use according to CORINE (“COordination of INformation on the Environment”) Land Cover data is 55% farmed land, 37% forests, 6% settlement areas and 2% water surfaces and wetland. 2.2. A SSESSING THE RELEVANCE AND VALIDITY OF SPATIAL AGRO - ENVIRONMENTAL INFORMATION FOR VARIABLES To develop a systematic methodology for designing a regional monitoring network and test its feasibility, we selected examples of environmental variables for four GM crops (Bt-Maize, HT winter oilseed rape, HT sugar beet, and starch-modified potato), which covered a certain ecological spectrum, respectively. Table I presents a set of examples. Two German GMO monitoring expert groups (“Bund-LänderAG: Monitoring von Umweltwirkungen gentechnisch veränderter Pflanzen” and “AG Anbaubegleitendes Monitoring”) assumed them to be potentially relevant for environmental monitoring of Bt-Maize. Their utilization for this work, however, is for methodological purpose only. A careful consideration of the individual steps performed within the monitoring scheme in Figure 1 shows that, among the respective driving factors and decisionmaking steps, inherent space-relevant issues need to be addressed. Each monitoring step contains an inherent uncertainty in terms of space, which accumulates among the subsequent steps. Selecting the analytical detection methods (Figure 1, step 4) is one of the most challenging tasks within the conceptual monitoring process. At this step all elements of the preceding steps need to be combined for the spatial monitoring design in step 5 to give answers to the questions: which potential adverse effects, which variables, which monitoring methods, which number and location of monitoring sites and which measurement frequency (why, what, how, where and when is to be monitored). These components we systematically integrated to attain a regional, spatially representative monitoring network. In general, a combination of environmental factors is responsible for the spatial distribution of variables to be monitored in the agro-environment (Leigh and Johnston, 1994; Perry et al., 2003; Werner et al., 2000). Moreover, GM crop monitoring has to integrate a set of different agro-environmental variables, each with their specific distribution in space and time. We therefore considered different scales along with the monitoring components (EC, 2003; Stein and Ettema, 2003). Table II summarises spatial information layers, each representing the spatial and temporal heterogeneity of agro-environmental components. The currently most valuable agro-environmental information for the case study Brandenburg is (a) CORINE Land Cover, which indicates different types of land use (Brandt et al.,

6

F. GRAEF ET AL.

TABLE II Super-regional spatial information layers relevant to GM crop monitoring in Brandenburg Spatial layers 1

Land cover

Landscape ecology2 Crop species distribution3 a) Cropping intensity 4 or b) Crop cultivation systems5 Present corn borer infestation6 Corn borer infestation potential7 Crossing partners8

Contents e.g. arable land, forests and natural areas, wetlands, urban areas Elevation, climate, soil and soil water site characteristics Cultivation areas (km2 ) of major arable crops a) Various levels in arable land use intensity or b) regional variation of agricultural systems (not yet available) Information from plant protection service Regional crop infestation survey of plant protection service Plant species distribution maps of potential crossing partners

No. of classes Resolution 5

250 × 250 m

6

1 × 1 km

3

District level

2–5

250 × 250 m

2

Regional level

4

Approx. district level

2

10 × 10 km

1 CORINE Land Cover data: Europe-wide aerial survey and inventory from 1995, describing land cover and land use according to a nomenclature of 44 classes organised hierarchically in three levels. We used the most aggregated level III (five general land use classes) and modified class two (arable land and heterogeneous agricultural areas) and three (forest, semi-natural areas and, permanent crops and pastures). 2 Landscape ecology: Ecoregion map for the Land Brandenburg. 3 Crop species area distribution: Regional census on district level on a five year period. Each crop classified within three quantiles in three intensity levels (1: low; 2: medium; 3: high). 4 Variability in cultivation intensity based on farm surveys (compare Firbank et al., 2003). 5 Regional cultivation systems’ map: Expert-based (in construction by the authors). 6 Regional annual corn borer infestation survey of plant protection service. 7 Corn borer infestation potential maps, derived from climate factors and pest surveys (Kluge et al., 1999). 8 Potential wild living crossing partners (BfN, 2003).

2002); (b) landscape or ecoregion maps with defined, homogeneous site characteristics that determine the distribution of ecosystems and habitats (Mueller-Dombois and Ellenberg, 1974). They largely integrate the ecological site conditions (Schröder and Schmidt, 2001) and thus are not limited to a specific variable, but can be used for a range of different variables along the food chain; (c) crop species distribution and respective area cultivated on the district level. Crop species distribution information of previous years is important where Bt GM crops are often grown continuously on the same field, because there might be cumulative effects over time; (d) arable land use intensity information derived from detailed farm surveys


7

(Hails, 2002; Perry et al., 2003). Information on different crop cultivation system types are desirable but not yet established; (e) present corn borer (Ostrinia nubilalis) infestation to pinpoint present GM insect-tolerant maize cultivation areas; (f) corn borer infestation potential (Kluge et al., 1999) in order to pinpoint potential GM insect tolerant maize cultivation areas; (g) wild plant species distribution, which enables determining areas with potential crossing partners. Due to the low resolution (10 × 10 km), this information may only serve as a rough basis and, if relevant, would require more detailed mapping. The choice of relevant information layers depends on multiple factors such as GM crop trait, environmental setting or data availability and is thus likely to vary according to circumstances. It is crucial to systematically determine the relevance of available spatial information for the variables. This we do in an assessment procedure that subsequently enables the spatial monitoring design (step 5). The key element of the assessment procedure is an expert-based assessment of the relevance of spatial information layers focussing on the question: “How strong, based on the existing state of knowledge, is the influence of the environmental factors within the spatial information layers on the spatial distribution of the variables to be monitored?” The following example using the potential variable to monitor “persistence of Bt toxin” with the GM crop Bt-Maize in Brandenburg, Germany, illustrates the assessment approach (Table III): The variable will be monitored only on arable land (ranking of 3 = highly important for spatial information “land use”). The toxin decomposition strongly depends on site factors (“landscape ecology”: ranking of 3). We expect no correlation between Bt-toxin decomposition and any high concentration of maize fields (“crop species distribution”: 1). We expect no influence of the corn borer (“infestation potential”: 0). Cropping techniques such as soil tillage and rotation systems will have an influence (“regional cultivation systems”: 2). There is no relation between crossing partners and Bt toxin persistency (“crossing partners”: 0). We made these assessments for Bt-Maize, herbicide-tolerant (HT) oilseed rape, HT sugar beet, and starch-modified potato for the respective variables selected for environmental monitoring by the expert groups mentioned above for all spatial information layers. From the single assessment results we calculated an overall ranking of each spatial information layer using the number of times a layer attains the highest ranking 3 (Table III); the specific order of this ranking is the basis for the subsequent geographical analysis of spatial information. 2.3. EVALUATING

ENVIRONMENTAL MONITORING NETWORK INFORMATION

Existing environmental monitoring networks provide useful baseline data over many years and different sites. These data are therefore very useful to interpret GM crop monitoring data. The selection of GM crop monitoring sites should be based on their infrastructure (Umweltbundesamt, 2001). Therefore, existing

8

F. GRAEF ET AL.

TABLE III Examples of variables for monitoring of Bt-Maize in Brandenburg (Germany), assessment and ranking of the relevance of spatial information layers for the spatial distribution of variables Variables measured Transgenic DNA in soil Species diversity and frequency of herbivorous arthropods and specific antagonists Species diversity and frequency of butterflies outside the cropped fields Saprophages in the soil (dipterous larvae) Microbial basal respiration Microbial sum indicators Change of cropping techniques Resistance of corn borer Overall assessment: number of times a layer attains the max. ranking 3 Overall ranking of spatial information layers2

Landscape Crop species Pot./pres. Crop. intensity/ Land use ecology distribution infestation cultiv. systems 31 3

3 3

1 1

0 1

2 2

3

3

1

0

1

3

3

1

0

2

3 3 3

3 3 2

0 0 1

0 0 3

1 1 3

3 8

1 6

3 1

3 22

3 22

1st

2nd

5th

4th

3rd

1

Assessment and ranking of the relevance of spatial information layers for the spatial distribution of variables on the following scale: 0 = no relevance, 1 = low relevance, 2 = medium relevance, 3 = high relevance. 2 Ranking is carried out for the number of times a layer attains the maximum score 3. Two times the score 2 equals the score 3. In case of equivalence the number of times a layer attains the score 2 decides upon the overall ranking.

networks were screened with regard to their suitability using the criteria (a) spatial representativeness of the agro-environment in terms of site numbers and spatial distribution using standard deviations of the frequency of sites located on agricultural land of different ecoregions, (b) amount and relevance for GM crop monitoring of measured biotic and abiotic variables in the respective monitoring networks (Table IV). There are four operative static monitoring networks suitable for GM crop monitoring, which dispose of geographically unchangeable sites. The PPA (Plant Protection Agency) network is flexible, with annually changing sites, and thus cannot be used within spatial planning but is likely to support the static network planning. The monitoring networks dispose of various numbers of periodically measured variables that we combined in groups of variables. We rated the measured variables’ suitability for GM crop monitoring and cumulated them for an overall network

9


TABLE IV Assessment of environmental monitoring networks in Brandenburg (Graef et al., 2004, modified) Monitoring network1

LSM

IEM

EMBR

FOPV

PPA

No. of sites in BB Farmland sites in BB Spatial representativeness 3 Groups and resp. No. of measured variables in monitoring networks Soil characteristics Soil analyses Climate factors Fauna Flora Agricultural management Total No. of variables Suitability assessment of variables4 Overall ranking

31 24 high

12 3 low

89 36 low

23 23 high

3 8 2 0 3 4 19 27

7 11 5 3 4 4 33 46

5 9 6 2 1 7 30 42

4 3 2 0 0 8 17 25

20–30, flexible – 20–30, flexible – medium – GM crop mon-itoring relevance2 1 1 1 1 1 1 2 2 2 8 2 13 – 24 –

3rd

1st

2nd

4th

–

–

1

LSM: long-term soil monitoring; IEM: integrated ecological monitoring; EMBR: ecosystem monitoring of biosphere reserves; FOPV: trial sites of Federal Office of Plant Varieties; PPA: Plant protection agency. Not relevant groups of variables not shown. 2 Rating of variables for GMO monitoring (results from working groups of the Federal Environmental Agency (UBA) and the Federal Biological Agency (BBA)): 0 = no relevance, 1 = medium relevance, 2 = high relevance. 3 Representativeness analysis using standard deviations of the frequency of sites located on agricultural land of different ecoregions. 4 Suitability assessment of variables for GM crop monitoring; index number of measured variables × GM crop monitoring relevance.

suitability assessment, using a simple index (number of measured variables × GM crop monitoring relevance, (Table IV)). With respect to the variables’ suitability, the IEM (Integrated Ecological Monitoring) and EMBR (Ecosystem Monitoring of Biosphere Reserves) network are more favourable than the LSM (Long-term Soil Monitoring), FOPV (trial sites of Federal Office of Plant Varieties) and PPA networks. With respect to spatial representativeness among different ecoregions, the LSM and FOPV are preferred. 2.4. I NTEGRATING

A RULE - BASED GEOGRAPHICAL ANALYSIS FOR DETERMINING MONITORING SITES

We defined GM crop monitoring network scenarios individually for Bt-maize, HToilseed rape, HT-sugar beet, starch-modified potato and combined for all four GM

10

F. GRAEF ET AL.

crops. For each, we developed optimisation procedures using Arc/Info GIS. The GIS analysis sequence scheme was implemented in AML (Arc Macro Language). This parameter input device included queries to produce (a) an optimised spatial design from available network sites, (b) potential additional sites and (c) tables outlining rejection reasons for network sites during the site selection procedure. The results can be regarded as parts of a spatial decision support system (SDSS), which is integrated into the GIS. The GIS analysis (Figure 2) is carried out using routine GIS procedures described in the following. It is applied to all scenarios using a layer selection sequence defined by the preceding rankings of spatial area information (Table III) and network point information layers (Table IV). It incorporates the three modules A, data preparation and decisions on monitoring intensity (Table II, Figures 3 and 4); module B, hierarchical rule-based optimisation of existing and potential additional sites (Tables II, IV and V); and module C, decision support (Figures 5–8). For the spatial area information the analysis basically constitutes a multiple sequential overlay of different geographical information types starting with the

Figure 2. Geographical analysis scheme for determining GM crop monitoring sites in Brandenburg (area and network information in Tables II, IV and V, illustrated in Figures 3 and 4; decision support illustrated in Figures 5–8).


11

Figure 3. Overlay of landscape ecology map with CORINE land cover class “arable land” resulting in six major landscape classes.

most relevant (high ranked) layers (Figure 3), mandatory for the spatial design, and ending with the less relevant ones. These low ranking layers are optionally overlaid, only if sufficient existing monitoring sites are located in the overlay results. This rule, implemented as a recursive mechanism following the overlay, helps avoiding a shortcoming of potentially suitable sites for the network design. Generally, the overlay of two layers results in an intersection layer containing more precise information on more regions, however with less overall surface of interest (Figure 3). The intersection layer again constitutes the basis for the next overlay, thus leading to a step-wise increase in information accuracy for the design of a representative network.

12

F. GRAEF ET AL.

Figure 4. Static agro-environmental monitoring networks and sites in Brandenburg (LSM: long-term soil monitoring; IEM: integrated ecological monitoring; EMBR: ecosystem monitoring of biosphere reserves; FOPV: trial sites of Federal Office of Plant Varieties).

Module A, data preparation and decisions on monitoring intensity: The highranked area layers (land cover and landscape ecology) were selected, and the respective classes were determined and overlaid to create a basic number of classes (BNC), all of which have to contain monitoring sites (Figure 3). The ranked network point information layers were overlaid (Figure 4). Because we lacked data on the expected spatial variability and error probability of variables in the landscape classes (Pilz et al., 2004; Stein and Ettema, 2003), we tested different sites frequencies and subsequently elaborated on three and six sites per class, taking into account the present network setting and comparable monitoring network designs, e.g. of Perry et al. (2003). With respect to network quality and representativeness, we selected


13

Figure 5. Spatial monitoring design: Monitoring sites for HT-oilseed rape combining all networks (landscape description see Figure 4).

three relevant networks and/or network combinations: LSM; LSM together with FOPV; and LSM together with FOPV, IEM, and EMBR. We then overlaid the network combinations with the BNC layer; as a mandatory rule, we excluded all those sites that were not a class member (Figure 2). Module B, hierarchical rule-based optimisation of existing and potential additional sites: For Bt-maize we overlaid the lower-ranked area information layers “present infestation”. For HT-oilseed rape we additionally overlaid the “land cover” layer, applying the GIS rule “select only areas with maximum distance