Tackling climate change: Agroforestry adoption in the face of regional weather extremes

The cultivation of agroforestry systems is regarded as an effective strategy that can help synergistically mitigate and adapt to the adverse impacts of climate change and regional extreme weather events. This study addresses the question of whether, and under what conditions, farmers are likely to adopt agroforestry systems in response to regional weather extremes and presents a novel research approach to tackle this question. A discrete choice experiment was conducted to elicit farmers’ preferences for agroforestry and wood-based land use systems. The results were combined with geospatial weather information. Assuming adaptive weather expectations, land users’ dynamic responses to extreme weather were simulated in terms of adoption probabilities. Farmers in the case study region in southeastern Germany were found to have a negative preference for alley cropping systems (i.e. agroforestry) and short rotation coppice, compared to an exclusively crop-based land use system. However, the results from the simulation of a 2018-like extreme weather event showed that alley cropping has a high probability of being adopted in the long-term. This study provides novel insights into the adaptive uptake of climate-resilient agroforestry systems. This information can be used to develop more effective policies and programs to promote agroforestry as a climate-resilient land-use


Introduction
The latest assessment report of the Intergovernmental Panel on Climate Change (IPCC) reiterates the fact that climate change poses exceptional challenges to various social and economic sectors on a global scale (IPCC, 2021).In its 2020 Global Risks Report, the World Economic Forum, listed climate-related concerns among the top five long-term risks for the first time (WEF, 2020).In addition to affecting annual mean temperatures and precipitation, climate change also increases the frequency of occurrences of regional extreme weather events such as droughts, heat waves, heavy rain and floods (IPCC, 2021;Lüttger and Feike, 2018;Mann et al., 2018;Westra et al., 2014).In this context, agriculture is often seen as one of the most susceptible sectors to such changes (IPCC, 2007), negatively affecting, for example, crop yields (e.g.Lesk et al., 2016;Schlenker and Roberts, 2009;Haqiqi et al., 2021), total factor productivity (e.g.Chambers et al., 2020;Chambers and Pieralli, 2020;Stetter and Sauer, 2021), and ultimately farm income and viability (e.g.Kawasaki and Uchida, 2016;Dell et al., 2014;Dalhaus et al., 2020).Prime examples of years with extreme weather conditions are the 2003 European heat wave, the 2018 European drought and heat wave, or the 2010-2013 southern United States and Mexico drought.Despite these impacts, agriculture is also regarded as one of the most important anthropogenic contributors to climate change (Lynch et al., 2021).Overall, farmers need effective adaptation and mitigation strategies to deal with the challenges of climate change.
One major channel through which agriculture can actively address climate change impacts is land use (Pielke, 2005).A promising pathway in this direction is the adoption of agroforestry and woodbased land-use systems, which are recognized to play a key role in synergistically approaching adaptation and mitigation (Verchot et al., 2007;Cardinael et al., 2021;van Noordwijk et al., 2011;Duguma et al., 2014;Noordwijk et al., 2014).These systems mitigate the effects of climate change through their carbon sequestration potential (Albrecht and Kandji, 2003;Schroeder, 1993;Oelbermann et al., 2004;Cardinael et al., 2017).There are also indirect mitigation effects created by the planting of trees and other woody perennials on agricultural land.These may also effectively reduce deforestation (Schroeder, 1993;Verchot et al., 2007) and help replace fossil fuels with fuel wood (Kuersten and Burschel, 1993).Simultaneously, the resulting positive regulating effects on hydrological cycles, soil, and the microclimate, may https://doi.org/10.1016/j.ecolecon.2024.108266Received 27 September 2023; Received in revised form 29 April 2024; Accepted 31 May 2024 lead to more climate-resilient farming practices (Lasco et al., 2014).Furthermore, agroforestry and its provision of multiple ecosystem services (Brown et al., 2018;Wolz et al., 2018) is also seen as a key component in the realm of ecosystem-based climate change adaptation (Pramova et al., 2012;Hernández-Morcillo et al., 2018).Provided the various merits of agroforestry systems, there remains a large untapped potential for the introduction of agroforestry and its expansion across the globe (Noordwijk et al., 2014), and this may become even more important in the face of an increased frequency of regional extreme weather events (Duguma et al., 2014;van Noordwijk et al., 2021).
This paper addresses the question of whether, and under what conditions, farmers are likely to adopt agroforestry and wood-based land-use systems against the background of regional weather extremes.
To answer this question, we conducted a discrete choice experiment to elicit farmers' preferences for alley cropping and short rotation coppice.We combined the results with geospatial weather information.Assuming adaptive weather expectations, the dynamic land-use responses of farmers to extreme weather years were simulated in terms of adoption probabilities utilizing the approach of Ramsey et al. (2021).The paper then discusses these land-use responses in the broader context of climate change resilience (OECD, 2020;Meuwissen et al., 2019).
We found that farmers in our case study region of Southeast Germany have a negative preference for alley cropping and short rotation coppice compared to an exclusively crop-based land use system.However, the results from the simulation of extreme weather under different scenarios show that alley cropping systems (agroforestry) might have a high probability of being adopted in the medium to long term, thus strengthening farmers' resilience to extreme weather events and climate change.
The elicitation of farmers' preferences for agroforestry and woody perennials has been the subject of multiple studies, including Gillich et al. (2019) and Pröbstl-Haider et al. (2016), who analyzed farmers' preferences for short rotation coppice in Germany and Austria using discrete choice experiments.Other studies focused on the adoption of agroforestry systems, mostly in the context of the Global South (Bayard et al., 2007;Amusa and Simonyan, 2018;Beyene et al., 2019;Schaafsma et al., 2019;McGinty et al., 2008;Dhakal et al., 2015).Furthermore, multiple authors have simulated the (economic) potential of agroforestry cultivation under different circumstances (e.g Paul et al., 2017;Frey et al., 2013).Lasch et al. (2010) and Gomes et al. (2020), projected the cultivation potential for short rotation coppice in eastern Germany and coffee-agroforestry in Brazil, taking into account various climate change scenarios until 2050.The problem with such scenarios is that they are usually conducted on a global scale and likely do not represent local farmers' actual and perceived experiences with extreme weather and climate change.This is why they are usually not well-suited for farm-level based simulations (Morton et al., 2015;Ramsey et al., 2021).Overall, studies on the effects of weather shocks on land-use change are scarce (Girard et al., 2021).To fill in this gap, Ramsey et al. (2021) developed a novel framework to simulate how farmers dynamically adjust their cropping decisions in response to specific weather patterns.
This study contributes to the literature in several ways.It quantifies the link between adverse weather and farmers' preferences for agroforestry and short rotation coppice accounting for short-to longterm adaptation responses.While many of the aforementioned studies address why integrating woody perennials into farms' cultivation plans might be useful for mitigation and adaptation, they frequently ignore whether and how farmers respond to weather patterns.Establishing this link is particularly important in light of the increased frequency of extreme weather events resulting from climate change.Furthermore, our research design, which combines a discrete choice experiment, geospatial weather information, and the simulation framework of Ramsey et al. (2021), allows us to provide novel insights into farmers' responses and resilience in the face of a changing climate.The approach presented in this study further allows us to evaluate two important aspects that have widely been neglected in the literature on the climateland-use nexus so far.This nexus has usually only been studied ex-post for already established land uses.With the approach of this study, it is possible to evaluate this relationship for more recent, not yet established land use systems.Also, by taking choice-specific attributes into account, it is possible to simulate alternative scenarios reflecting the role of legislation, market conditions, and technological progress.
The remainder of the article is structured as follows.First, a short description of agroforestry and wood-based agricultural land use systems is provided before presenting our conceptual framework (Section 2).In Section 3, the data collection and empirical strategy is presented.Section 4 describes the results from the discrete choice experiment and the weather simulations, followed by a discussion of the most important findings (Section 5).The paper closes with a summary and several concluding remarks in Section 6.

A short description of agroforestry and wood-based agricultural land use systems
Agroforestry systems are land-use systems in which woody perennials are integrated with agricultural crops and/or livestock on a piece of land, either in a spatial arrangement or in a temporal sequence (Nair, 1985;Leakey, 2017;Cardinael et al., 2021).This definition encompasses a wide range of diverse systems including silvopastoral (the combination of trees with livestock), silvoarable (planting crops between rows of trees), forest farming (food, herbal, botanical, or decorative crops under a forest canopy), home gardens; as well as hedgerows, windbreaks, riparian buffer strips, and many more (Pantera et al., 2021;USDA, 2019).As can be seen by this diverse list, agroforestry is not a new concept and goes back a very long time in many regions of the world (Pantera et al., 2021).
As for the integration of trees on agricultural land, short rotation coppice systems have been identified as an attractive land-use alternative from both economic and ecological perspectives (Wolbert-Haverkamp and Musshoff, 2014;Baum et al., 2009).Short rotation coppices usually consist of fast-growing tree species such as poplar, willow, paulownia, robinia, or eucalyptus with short rotation periods and frequent harvests (every three to five years) (Rödl, 2017).Unlike agroforestry, short rotation coppices are typically associated with a single use on the same field.
More recently, alley cropping systems that integrate strips of short rotation coppices into agricultural fields have received increasing attention (Tsonkova et al., 2012).In such a system, farmers produce crops and woody biomass in the same field at the same time.This can result in multiple benefits across several domains.Many previous studies, including Paul et al. (2017), Gosling et al. (2020) and Schoeneberger et al. (2017) have found that alley cropping can generate higher economic returns than single-crop land uses.Diversifying production output can increase economic stability (Tsonkova et al., 2012), and alley cropping can contribute to a more sustainable bio-based economy by simultaneously providing food and renewable raw materials (Gillich et al., 2019).Numerous studies have also found positive effects on crop yield and land-use efficiency (see e.g.Schoeneberger et al., 2017).Alley cropping also provides a range of environmental services due to its multifunctional nature.It has the ability to break up large-scale structures, increase biodiversity through increased habitat, increase species diversity and their connectivity throughout agricultural landscapes, and it can reduce soil erosion and nutrient leaching (Langenberg et al., 2018;Tsonkova et al., 2012;Schoeneberger et al., 2017).
Finally, agroforestry systems and short rotation coppice can, to some degree, play an important role in synergistic approach to climate change mitigation and adaptation.In terms of mitigation, alley cropping and short rotation coppice can store large amounts of carbon in aboveground and belowground biomass (Albrecht and Kandji, 2003) as well as in soil (Cardinael et al., 2017), thus reducing atmospheric carbon dioxide (CO 2 ) (Cardinael et al., 2021;Tsonkova et al., 2012;Schroeder, 1993).Regarding adaptation, integrating trees into agricultural lands provides a buffer against weather extremes through regulating hydrological cycles, improving nutrient and water-use efficiency, and modifying microclimates (Wolz et al., 2018;Ashraf et al., 2019;Pramova et al., 2012).Agroforestry can also diversify farmer income by hedging financial risk (Wolz et al., 2018), and can make production more resilient to the adverse effects of climate change (van Noordwijk et al., 2021).Despite these multiple benefits, silvoarable agroforestry systems are still relatively rare in Europe (den Herder et al., 2015;Langenberg et al., 2018).Noordwijk et al. (2014) notes that there is a huge potential for the introduction and expansion of agroforestry areas around the globe.

Land-use, random utility maximization and weather expectations
Given the large potential for the introduction and expansion of agroforestry, this study seeks to elicit farmers' preferences for agroforestry and short rotation coppice compared to conventional crop farming against the background of climate change.The theoretical concept for this analysis is based on random utility theory following Lancaster (1966) andMcFadden (1973). 1 In planning the use of their land, farmers face a choice among a set of alternative land uses in different decision situations under varying conditions.Each farmer receives a certain level of indirect utility from each land-use alternative.In a given decision situation , she will select alternative  if and only if   >   ,  ≠ .The indirect utility of an alternative cannot be directly measured but it can be expressed by a systematic (deterministic) component  , which reflects specific characteristics as well as individual and location-specific features of the farmers, and a random component , which represents unobserved decision-relevant elements (Mariel et al., 2021).A farmer  receives a certain level of indirect utility   from a land use alternative  in a choice situation : 1 See Supplementary Material S.1 for a discussion on the use of random utility theory in the context of land use and production decisions.
As usual, it is assumed that farmers' utility for a land-use alternative varies with a set of decision-relevant characteristics (, see Section 3.2.1).Furthermore, it is assumed that farmers' utility also depends on expected weather  at the time of the planting decision: where  and  are coefficients to be estimated.Following Nerlove (1958) and Ramsey et al. (2021), we assume that farmers have adaptive weather expectations that are based on their past local weather history, where both short-and long-term trends may affect land use choices.
The fact that past weather creates different types of subjectively experienced uncertainties (Singh et al., 2020) provides the direct link between local weather history and individual decisions.Following this line of argument, we assume that farmers' filter the signal coming from their local weather history through a subjective uncertainty space (see Fig. 1), where uncertainty is defined as ''[. . .] a cognitive state wherein a person lacks perfect information, time, and resources that could enhance the individual's ability to evaluate all possible outcomes of a decision and choose the optimal outcome.Uncertainty reflects the level of incomplete understanding, knowledge, or information regarding a phenomenon, activity, individual, or action that may influence an individual's decision-making [. . .]'' (Singh et al., 2020(Singh et al., , p.1050)).Thus, a change in local weather, e.g.due to an extreme weather event leads to an update of individuals' expectations, which are processed through their uncertainty filter and affect their utility coming from different land uses, and ultimately influence their adoption decision.Furthermore, it is realistic to assume that farmers do not assign equal importance on each past weather event, which is why a simple average of past weather would not properly reflect farmers' expectations.Ramsey et al. (2021) express the expected-weather-formationprocess as follows: where c− are actual past weather events. 0 is a reference expectation,   reflects a farmer's weight assigned to the recent past,   is the weight assigned to the more distant past, and  (⋅) is a weighting function (e.g. annual mean).Hence, weather expectations are formed by two components, one reflecting longer term weather patterns and one reflecting short term weather variations.With this approach, we take up recent conceptual developments in the context of expectation  , 2021;Ji and Cobourn, 2021;Wimmer et al., 2024).While a discounting mechanism (e.g., exponential, (quasi-)hyperbolic, etc.) is an alternative approach to considering historical weather data, it suffers from the conceptual limitation that short-term experience is inherently considered more important than long-term weather experience.While some studies suggest that individuals may place more weight on shortterm weather patterns when making decisions, research on climate change adaptation suggests that longer-term weather patterns may play a more significant role in influencing land use decisions (Mérel and Gammans, 2021;Hsiang, 2016;Burke and Emerick, 2016).
In light of these theoretical considerations, past (extreme) weather events are expected to influence farmers' decisions to adopt more climate-resilient methods and mitigating options.These include agroforestry or short rotation coppice by affecting farmers' weather expectations, which ultimately influence farmers' preferences for their selected options.

Material and methods
This study first provides information on the case study region, Bavaria, before describing the discrete choice experiment (DCE) setup used to collect data on farmers' preferences.Then, the data that are used to describe the weather are presented.By combining the experimental data with local weather data and utilizing a correlated random parameter logit model (RPL) approach, it is possible to estimate farmers' preferences and probabilities for the cultivation of each landuse option and to obtain coefficient estimates reflecting the influence of land-use characteristics and (anticipated) weather.Finally, the simulation approach used to model the adaptive adjustment behavior of farmers in response to an extreme weather year based on the estimates from the RPL model is described.

Study area
The DCE was conducted in Bavaria, a federal state of Germany in central Europe (Fig. 2).Located in the southeast of Germany, Bavaria belongs to the core regions of agricultural production within the European Union (EU).It reflects the diversity of European farming (conditions) to a high degree, which is why this site was chosen for conducting this study.In terms of natural conditions, farming takes place along an elevation gradient of 1500 m (from 100 m in northwestern Bavaria to 1600 m in southeastern Bavaria) and a macroclimatic gradient with a mean annual temperature range between 3 • C and 10 • C and an annual precipitation of 470-1592 mm (from 1960 to 2020).Its natural conditions range from pre-alpine and alpine areas with high precipitation and rather clayey, limestone and dolomite based soils to regions with flat land and fertile loess soils to dry, marlstone, limestone and dolomite based hillside locations.

Choice experiment setup
A DCE was used to elicit the influence of land-use characteristics on farmers' decisions about whether or not to cultivate agroforestry.Each farmer was repeatedly confronted with a choice situation, in which the attributes of three land-use alternatives (namely short rotation coppice, alley cropping, and status quo crop farming) varied.

Attribute selection and levels
After a careful literature review and feedback from agricultural experts, the attributes used to describe the land-use alternatives are as follows: expected yearly contribution margin, yearly contribution margin variability, minimum useful lifetime, payments for ecosystem services and a dummy if the alternative qualifies as an ecological priority area (see also Supplementary Material S.2).
The primary monetary attribute, the expected contribution margin, measures yearly revenues (yield times price), minus variable costs per hectare.Fixed costs and subsidies are not considered.Moreover, because revenues and costs are spread over the entire production period of short rotation coppice and alley cropping, a contribution margin equivalent, which corresponds to the annualized form of the net present value is introduced.
Previous studies show that uncertainty plays an important role when it comes to farmers' decision-making processes in general (see e.g.Menapace et al., 2013) and land allocation in particular (El-Nazer and McCarl, 1986;Knoke et al., 2015).This study expresses outcome uncertainty in terms of contribution margin fluctuations.It is defined as the average annual percentage rate of variation in the contribution margin.Since farmers tend to be risk averse (Menapace et al., 2013), we expect an increase in variability to negatively affect preferences.
The minimum useful lifetime (in years) of a land-use alternative is closely related to the entrepreneurial flexibility of farm businesses.Being tied to a land-use type for a longer period of time means a loss of flexibility (Musshoff, 2012).2This is expected to have a negative impact on farmers' preferences.As short rotation coppice and alley cropping provide a wide range of environmental services, agri-environmental payments (as compensation for the provision of ecosystem services) could provide a positive incentive for farmers to cultivate one of these land-use options (e.g.Layton and Siikamäki, 2009).Agri-environmental payments are expressed in terms of euros per hectare.
Finally, Langenberg et al. (2018) find that a key driver for farmers to engage in alley cropping may be the designation of the area as an ''ecological priority area''.To obtain this designation, farmers must allocate a certain amount of land to ecological priority areas (which are considered environmentally friendly).They then receive area-based ''greening'' payments, which account for approximately 30% of the farmers' total basic payment.

Conducting the choice experiment
Once the choices, attributes and corresponding levels were determined, the actual choice experiment commenced.A choice experiment was created with three labeled alternatives, namely ''Short Rotation Coppice'', ''Agroforestry'', and ''Status Quo''.Given its labeled nature, we followed Viney et al. (2005) and created an L MA design for the DCE, resulting in 36 choice cards.To reduce the psychological burden of answering all the choice tasks, they were randomly blocked into three sets of twelve choice cards each.An example of a choice card can be found in the Supplementary Material S.3.Each respondent was then randomly assigned to one of the three blocks.Before the participants began the choice experiment, they were given an explanation of how the DCE would work, along with descriptions of the alternatives and attributes relevant for the task (see Supplementary Material S.2 and S.4).By maintaining a neutral framing, we aimed to elicit participants' true preferences without influencing their choices a priori.While we recognize that framing matters, the literature on framing in discrete choice experiments is inconclusive regarding the extent to which it affects choice results (Rolfe et al., 2002;Kragt and Bennett, 2012; extended investment horizon can increase susceptibility to unforeseen events such as changes in market conditions, technological advances, or regulatory changes that affect the stability of profits (Brach, 2003).
3 To ensure a relatable and realistic status quo option, we utilized data from the Bavaria-specific LfL (Bavarian State knowledge and service center for agriculture) contribution margin database (LfL, 2020), validated through consultation with local experts and farmers' pre-test feedback.A post-test comparison with empirical data (LfL, 2021) confirmed its appropriateness.For agri-environmental payments, we examined Bavarian schemes, encompassing climate and biodiversity protection, soil, and water conservation (BStELF, 2018).The status quo reflected the minimum requirements for direct payments, with short crop rotations of about three years prevalent in arable farming in the case study region (Leteinturier et al., 2006;Glemnitz et al., 2011;Stein and Steinmann, 2018).Respondents affirmed the appropriateness of the status quo option during the pretest phase.Bujosa et al., 2018;Pelletier et al., 2022).Hence, results should be interpreted with caution in light of the existing ambiguity on the role of framing in discrete choice experiments.
Furthermore, we used ''cheap talk'' (Landry and List, 2007;Vossler et al., 2012) and reminded the participants about the hypothetical nature of the experiment, and that they should nonetheless answer truthfully. 4he survey consisted of several parts.After some general information and consent to participate, respondents were asked about general (socio-economic) characteristics of their farm, followed by the DCE.Finally, participants were asked to provide further information on their local perceptions of climate change and several personality traits.
After an extensive pre-test phase, which took place in the early summer, the survey was conducted online in October 2020.The survey was facilitated through agriEXPERTS, an agricultural market research platform affiliated with Deutscher Landwirtschaftsverlag (dlv), a specialist publishing house for agriculture.The survey was administered to their online access panel, which consists of pre-selected farmers who have expressed an interest in participating in research surveys.In total, 58% of the panel members (out of 361) participated in the survey.The survey included an invitation to take part in a lottery to win one of ten vouchers for a popular agricultural clothing shop worth 50 EUR each.It took approximately twelve minutes to complete the questionnaire.

Weather variables
To accurately describe the local weather history of these farms, five common weather indicators were selected, including average temperature, precipitation sum, number of dry days, number of hot days and the number of heavy rain days, during the local growing season (March-October). 5The variables were derived from 0.1 degree gridded daily data from the European Climate Assessment & Dataset (ECA&D) project (Cornes et al., 2018).Following ETCCDI (2018) and DWD (2022), dry days are defined as days with precipitation of <1 mm and hot days are defined as days with maximum temperature >30 • C. On heavy rain days precipitation exceeds 20 mm (DWD, 2022).Using the postal codes provided by the respondents, the meteorological indices were aggregated and linked to the farm locations and responses from the questionnaire. 6s described in Section 2.2, farmers form their weather expectations based on historical weather patterns, which can be distinguished into short-and longer-term weather patterns.To capture this distinction, short-and long-term weather variables were defined by different lag structures.The baseline specification for short-term weather patterns for the five indicators were based on the average of years  − 1 to  − 3 (more recent past) and longer-term weather patterns are based on the average of years  − 4 to  − 10 (more distant past).Fig. 3 summarizes these variables (see also Supplementary Material S.5).Further lag structures were computed reflecting several candidate time horizons of expectation formation, which were later tested against the base structure.All weather variables were mean-centered, which proved

Econometric approach
For the econometric analysis, the RPL model was used to account for preference heterogeneity in the utility function of the investigated sample (Hensher et al., 2015;Train, 2001).The utility function (2) was parametrized with alternative specific constants (  ), land-use specific attributes () and individual-specific weather parameters (): The model formulation is a one level multinomial logit model, for individuals  = 1, … ,  in choice setting  and alternatives .It assumes a Gumbel distribution of the error term   , and the probability of each choice  is as follows (Hensher et al., 2015): In the RPL framework, the coefficient vectors   ,   , and   are considered random draws from a distribution whose parameters need to be estimated.Under this assumption, we can use maximum simulated likelihood estimation to obtain coefficient estimates for   ,   , and   (Train, 2001).A total of 1000 Halton draws were used for each model estimation.As for the parameter distributions, it was assumed: where  () and   describe random unobserved preference variation, with mean zero and covariance matrix with known values on the diagonal, and fixed by identification restrictions. is a lower triangular matrix that allows correlation across the attribute-related random parameters and  = ( 1 , … ,   ).This specification followed the approaches of Hess and Rose (2012) and Hess and Train (2017), who showed that only by allowing for correlation across attribute-related random parameters, is it possible to capture scale heterogeneity alongside heterogeneity in utility coefficients.Ignoring this correlation could severely bias parameter estimates.
It is assumed that all random parameters were normally distributed except for the coefficient of the contribution margin, which was assumed to be log-normally distributed because economic theory states that the sign for the profit attribute should always be positive.Following this line of argument, the same logic could apply to the ''Agrienvironmental payments'' attribute.However, previous research showed that farmers might respond negatively to support payments, because they might evoke a sense of reliance and undermine farmers' notion of autonomy (Stock and Forney, 2014), invoke high transaction costs (Mettepenningen et al., 2009), or because farmers might not have sufficient trust in agricultural regulations (Stuhr et al., 2021).Therefore, restricting the sign of this attribute might be too restrictive.Nevertheless, we tested our preferred model against two alternative models: (a) All random parameters were normally distributed (including contribution margin) and (b) all random parameters were normally distributed except for contribution margin and agri-environmental payments, which were log-normally distributed.

Post-estimation simulation
To evaluate the short-and longer-term adjustment dynamics to an extreme weather period, farm-level responses to one-to five-year weather shocks were simulated over a period of 10 years using the method of Ramsey et al. (2021).The simulation is based on the estimated parameters from the fitted RPL model in Section 3.4.These simulations are primarily based on the drought year 2018, which caused severe damage to German crop farming (Webber et al., 2020).Following the argument of Girard et al. (2021), that different weather shocks have different impacts on land-use responses, we also present simulation results for a 2003-like heatwave (Ciais et al., 2005).Note: CM = Contribution margin (Euro), CMV = Contribution margin variation (%), MUL = Minimum useful lifetime (years), AEP = Agri-environmental payments (Euro), Green = Cultivated area eligible for greening premium.
Given the lag structure of the weather variables, farmers' adoption probabilities can be simulated each year during and after a weather shock based on the formula for land use probabilities (Eq.( 5)).In the baseline scenario, the values of the weather variables  were replaced for each farm in years 0-10 by their respective long-term averages (LTA) over the 30-year period 1991-2020.For a one-year shock scenario, the 2018 (2003)-like event is assumed to occur in period  = 0, and then the weather returns to the LTA.This shock will affect the values of the short-term weather variables (lags 1-3) in periods 1-3, and then they return to the LTAs.The longer-term weather variables (lags 4-10) remain at the LTAs for periods 1-3 before changing to a ''shocked'' level in years 4-10 after the shock (compare Ramsey et al., 2021, p.13, and Supplementary Material S.6).Fig. 4 illustrates the composition of each weather variable over time for a one-, two-and three-year weather shock as they enter Eq. ( 5) in the simulation.
We ran simulations for the full sample as well as for each district separately to explore more potential regional adaptation paths.Table 2 summarizes the respective values used for the construction of the weather variables.Regarding the levels of the land-use attributes , we constructed several scenarios, reflected by different attribute levels used in the simulations.The respective levels and scenarios are summarized in Table 3.

Sample summary statistics
In total, we received 210 responses.After performing plausibility checks, twelve responses were removed. 8In Table 4, summary statistics for key farm characteristics in this sample are described and compared with the population means for Bavaria.Approximately half of the sample are full-time farmers, which is only slightly higher than the Bavarian average (45%).Several characteristics of this sample are similar, on average, to the Bavarian average, namely cropland and grassland shares, farmers' ages and the participation rate in agri-environmental programs.
At the same time, these sample farms manage more land on average, have a smaller share of rented land, and have a smaller workforce than the population mean.Also, the sample share of organic farms with 10% is very similar to the population share (12%).Overall, the descriptive statistics show that our sample reflects the Bavarian farmer population reasonably well, except for a few dimensions including farm size and labor. 9These deviations from the population mean are not necessarily negative in light of a dynamic trend towards fewer but larger farms within the EU (Wimmer and Sauer, 2020).Nearly all farmers stated they had already experienced negative consequences due to climate change related extreme weather events, especially in the form of yield and quality losses.

Choice experiment results
The model estimation results are summarized in the Supplementary Material S.7.In a first step, we compared the model of our choice -the correlated RPL model (model 4) -to a multinomial logit model (model 1), a correlated RPL without weather variables (model 2) and an uncorrelated RPL model with weather variables (model 3).Likelihood-ratio tests showed that the correlated RPL model with weather information was a significantly better fit to the data than the alternative models.We also tested alternative distributional assumptions regarding the random parameters (Section 3.4, Supplementary Material S.8), which were also rejected in favor of our preferred model.S.9 shows the correlation structure of the parameters.Additionally, these tests have empirically confirmed that the weather (history) variables jointly have a significant impact on farmers' land use decisions as assumed in the theory section.From Supplementary Material S.7 (Model 4), we can see from the negative attribute specific constants (ASC) that arable crop cultivation is preferred to both short rotation coppice and alley cropping under average weather conditions.Expressed as probabilities, this means that on average, crop farming has a 49% probability of being adopted by farmers, followed by alley cropping (30%) and short rotation coppice (21%).Finally, significant preference heterogeneity was found (indicated by the highly statistically significant estimates of the standard deviations of the random parameters). 109 Such discrepancies can potentially be compensated through weighting.While weighting can potentially enhance sample representativeness, it necessitates accurate population data and can compromise statistical power (Bollen et al., 2016).Furthermore, the interpretation of weighted analyses can be challenging, particularly for complex statistical models, as the weighting process can distort standard errors and complicate the interpretation of results (Gelman, 2007).In the present study, these limitations precluded the use of weighting. 10The low standard deviation estimate for ASC:SRC suggests that there is little (but statistically significant) variation among individuals in terms of their We calculated the average marginal effect of a ceteris paribus change in the land use attributes on farmers' adoption probabilities to gain further insight into farmers' land-use preferences (Fig. 5).We found that, on average, a e10 increase in the contribution margin for alley cropping would lead to a 1.8% increase in the probability of cultivating alley cropping, a 1.4% decrease in the probability of adopting short rotation coppice, and a 0.5% decrease in the probability of cultivating arable crops only.This implies that a more substantial increase, say e100, in the contribution margin dedicated to alley cropping could potentially lead to a 18% increase in its adoption probability.The responsiveness is even more pronounced for short rotation coppice.A 10-euro change in the contribution margin translates into an average change of 2.2% in the probability of choosing this land use alternative.Interestingly, a very similar pattern emerges when examining the impact of agri-environmental payments.This finding suggests that farmers, on average, exhibit a certain degree of indifference towards the source of additional financial incentives.
Moreover, an increase in the contribution margin variability leads to a reduction in the probability of adopting the associated land use alternative, while the probability of the others increase.Regarding the minimum useful lifetime, an increase in one wood-based alternative leads to an increase in both the directly affected land use and its woodbased counterpart, while the probability of crop farming decreases.This can be explained by the fact that a modification in the lifespan of woody perennials impacts both agroforestry alternatives in equal directions, as both systems are (partly) based on woody perennials.Therefore, if the lifetime of woody perennials changes in a way that makes them more appealing to farmers in either short-rotation coppice or alley cropping, the other system will also benefit from this change, while the status quo will become relatively less attractive.Finally, eligibility as an ecological priority area boosts probability of adoption.The magnitude of this boost varies from approximately 2.2% for short rotation coppice to 3.4% for alley cropping.
This paper refrains from analyzing the coefficient estimates of the weather variables individually because they are likely to suffer from some degree of multi-collinearity, which is not a problem per se but makes ceteris paribus statements very difficult.This issue is discussed further in Sections 4.3 and 5. preferences for short rotation coppices.This might suggest little heterogeneity in how individuals perceive or value SRC.While alternative explanations such as model misspecification or measurement error remain a possibility, our analysis provides no evidence to suggest this.As mentioned above, there are several ways to empirically specify the lag structure of the weather history data to reflect longer-term weather patterns and short-term weather variations.Therefore, we tested a series of alternative weather variable specifications and reestimated the correlated RPL model and compared the model fit with our selected model (lags 1-3 and lags 4-10) (Supplementary Material S.10).It can be seen that the preferred model fits the data best followed by models with 1/2-20 and 1-3/4-25 lag structures.

Weather simulations
To examine farmers' agroforestry adoption in the context of more extreme and adverse weather patterns, which are predicted to occur more frequently and last longer, we simulated a 2018-like (and 2003like) extreme weather event and observed the deviations of land-use probabilities from the average 30-year baseline weather, considering multiple socioeconomic scenarios.
Fig. 6 summarizes the results for a 2018-like weather shock that lasts between one and five years for the predefined scenarios from Section 3.5.Baseline probabilities show that crop farming is the most favored option (ranging from 63% to 79%) across all regular-case scenarios, reflecting farmers' reservations about wood-based alternatives (with average baseline probabilities ranging from 6% to 22%).Even when conditions for these alternatives improve, alley cropping is less likely to be adopted than crop farming (21% vs. 52% in the mid-case and 26% vs. 42% in the best-case scenario).Short rotation coppice has a baseline probability of 27% in the mid-case and 31% in the best-case scenario.
For a one-year weather shock, we find that farms' land-use average adoption probabilities remain comparatively close to their baseline levels in the long term after an adjustment period.In all scenarios, alley cropping becomes more likely to be adopted in the long run.However, in the regular-case scenarios, a one-year shock does not make farmers favor alley cropping (or short rotation coppice) over crop farming.There is a noticeable pattern: The transition path exhibits a consistent structure across all scenarios.In the initial years after a shock, farmers show a preference for status quo crop farming, as indicated by a rising adoption probability before dropping below its initial level over time.Furthermore, Fig. 6 shows that these trends become more pronounced as the weather shock persists for two to five years.Similar patterns emerge, with an increased likelihood of crop farming adoption during and shortly after an extreme weather year, alongside a decrease in alley cropping probability followed by a recovery phase.Nevertheless, following a more pronounced extreme weather event (in terms of duration), the average adaptation path of farms becomes considerably more pronounced.However, in the regular case scenario, the probability of adopting alley cropping is only comparable to crop farming after a 4-5 year shock, but it never reaches the baseline level of crop farming.For example, the probability of crop cultivation in the regular case scenarios approaches more than 90% after a prolonged weather shock before dropping significantly below the baseline.Lastly, when considering a five-year weather shock scenario, alley cropping eventually becomes the preferred land-use option in all cases.Notably, the regular-case scenario, with a shorter minimum useful lifetime, suggests that reducing this lifetime could drive the shift to alley cropping even more effectively than policy support.Short-rotation coppice seems to be less significant in farmers' long-term adaptation behavior (despite favorable conditions).
To ensure the validity of our findings, we also conducted simulations of a 2003-like weather shock (Supplementary Material S.11).While the overall trends remained consistent, slight variations were observed.Notably, in a scenario of a 2003-like (extended) weather event, alley cropping more consistently emerged as the favored land-use choice.As mentioned before, the chosen approach to modeling weather expectations (Ramsey et al., 2021;Wimmer et al., 2024), while conceptually sound, may be susceptible to inherent structural multicollinearity, that is, short-and long-term weather variables can be highly correlated (akin to polynomial regressions).This correlation can pose a challenge to the model's stability, potentially leading to unstable estimates that are sensitive to small changes in the data.To demonstrate this concern, we employed a leave-one-out cross-validation (LOOCV) approach (James et al., 2021), systematically removing one respondent at a time from the estimation (i.e., changing the data slightly) and comparing the simulation results of each run.The results of this check for the regular case scenario are presented in the Supplementary Material (S.12).While the overall patterns remain consistent on average across LOOCV runs, a substantial degree of variability persists, especially regarding long-run results, emphasizing the sensitivity of our findings to minor data perturbations.In light of this uncertainty, we advocate for cautious interpretation of the study results.
In summary, the analysis reveals that farmers initially favor crop farming during one-year weather shocks, and gradually shift to alley cropping in the long run.When weather shocks last for two to five years, alley cropping becomes the preferred choice.Farmers initially prefer the status quo of crop farming after a shock, with the adoption probability increasing before falling below its initial level over time; meanwhile, alley cropping becomes increasingly favored in the long run.

Discussion
Both natural conditions and the business environment (the economic, regulatory, and institutional conditions) influence farmers' land use decisions (Mosquera-Losada et al., 2023;Cai et al., 2022;Wimmer et al., 2024).The estimation results, as well as the simulations demonstrate that both extreme weather events (natural conditions) and land-use attributes (business environment) significantly impact the adoption of agroforestry practices.Effective dissemination of climateresilient agroforestry practices therefore requires the consideration of both natural conditions and the business environment.
When examining farmers' dynamic responses to extreme weather years (Section 4.3), interesting adoption pathways emerged.These patterns can be broadly categorized into three phases: the absorption phase (immediately following a shock, with land-use probabilities diverging from the baseline), the recovery phase (when probabilities return to their initial levels), and the adaptation phase (when probabilities shift towards a new equilibrium).These phases illustrate crucial resilience capacities within agricultural systems (OECD, 2020;Meuwissen et al., 2019).
C. Stetter and J. Sauer During the absorption phase, the probability of maintaining the status quo in crop farming increases after a weather shock similar to the 2018 ( 2003) event.This suggests that, on average, farmers are more likely to adhere to their existing practices immediately after a shock, as crop farming consistently has the highest probability in most scenarios.This may seem counter-intuitive, as one might expect farmers to shift to more weather resilient land uses such as alley cropping and short rotation coppice (Ogunbode et al., 2019;Wilson et al., 2020).However, short-term decision factors are often rigid, with fixed production structures that limit the immediate capacity of farmers to make changes (Girard et al., 2021).Additional barriers may include behavioral factors such as perceived risks and assessments of the costs and benefits associated with adopting more weather resilient land uses (Dessart et al., 2019).As a result, farmers may be inclined to make adjustments within their familiar land use system, namely arable crop farming.This trend may increase with the duration of the weather shock.
The duration of the recovery phase varies depending on the scenario and ranges from one to five years.Irrespective of the scenario, it is evident that the sampled farms recover their land-use probabilities after a weather shock (also observed in Béné et al., 2012;OECD, 2020).This phase can be interpreted as a period during which the effects of extreme weather have subsided, allowing farmers to reconsider their initial land use and prepare for adaptive actions.
In the adaptive phase, mixed effects are found regarding farmers' adaptive capacity (Smit and Wandel, 2006;Engle, 2011).Alley cropping and short rotation coppice both offer comparative advantages over crop farming, apart from their relative excellence in terms of climate robustness.However, if there is no monetary incentive or technological improvement, farms remain reluctant to transform and adopt these options.Although there was a certain degree of heterogeneity across the two shocks, this trend was quite stable in our analyses.Nevertheless, farmers seem to recognize the relative excellence of the alley cropping system, as the probability of adopting such a system after a weather shock increases in the long run, regardless of the scenario.In particular, agroforestry becomes the preferred land use option in the case of a very long-lasting extreme weather period (i.e., five years).
Next, our results have important implications for policy-makers.First, agri-environmental payments increase farmers' probability of adopting wood-based and agroforestry land-use systems and farmers seem to be indifferent as to where the financial bonus comes from.Therefore, they can be an important lever to promote the cultivation of these climate-robust systems.The current Common Agricultural Policy of the EU suggests that agroforestry systems are financially supported by way of the newly created eco-scheme framework (Deutscher Bundestag, 2021).In Germany, farmers receive e60 per hectare of tree cover.This means in an agroforestry system with 10% tree cover, e6 per ha (Landwirtschaftskammer Niedersachsen, 2022).This would mean an average increase in adoption probability of less than one percentage point and is therefore unlikely to have an noticeable impact on farmers' land use decisions.
Another policy-relevant driver of agroforestry adoption is the minimum useful lifetime of the wood-based land use options.Farmers appear to place a high value to their entrepreneurial flexibility (see also Musshoff, 2012).Rosenqvist and Dawson (2005), Avohou et al. (2011) and Londo et al. (2001) showed that the useful lifetime of woodbased land uses is very important for their economic viability.To better incentivize land-use change, legislators could create a framework that encourages the development of coppices with reduced minimum useful lifetime but without reduced economic benefits.One way to do this could be the promotion of novel breeding methods, which have shown high innovation potential across several domains (Qaim, 2020).
Moreover, our analysis adds to a small but growing body of studies assessing the link between climate variability and land-use change (Girard et al., 2021).While most of these previous studies focus on established land use types and crops (Ramsey et al., 2021;Salazar-Espinoza et al., 2015;He and Chen, 2022), the approach of this study allows the ex ante assessment of the potential of novel, non-established land-use types that may play an important role in the future.What is more, the integration of a choice experiment into the simulation approach allows the evaluation of different scenarios, thus providing a more holistic view on the link between extreme weather and land-use responses.
Finally, this study has some limitations that bear mentioning.For example, we use cross-sectional weather data for the estimation of the econometric models.This means that farmers' preferences were measured only at one point in time (October 2020).This might be problematic under the assumption that preferences vary temporally (neglecting weather changes that the model accounts for).However, several studies suggest that preferences are likely to be stable at least in the short to medium term (see e.g.Dasgupta et al., 2017;Doiron and Yoo, 2017;Andersen et al., 2008).
Another limitation relates to the direct interpretation of the estimated weather coefficients in the RPL model.It is unlikely that any of the weather indicators change in isolation, i.e. ceteris paribus statements are not valid.This is why we refrained from direct interpretation of the estimated weather coefficients and instead focused on the weather simulations, which alleviates this problem to some extent (Ramsey et al., 2021).However, the presence of this structural collinearity can lead to unstable parameter estimates.To overcome this challenge, future research could investigate alternative modeling techniques that eliminate this issue.A promising way forward could involve a grid search method to identify individuals' weather expectation formation patterns.This approach would entail evaluating various combinations of positive and negative discounting factors for each weather variable.While computationally demanding, this method could offer an accurate representation of the decision-making process underlying weather expectation formation, while at the same time eliminating structural collinearity as a concern.
Last, hypothetical bias might be a concern in the presented choice experiment setup.The issue was addressed using cheap talk.There are varying results in the literature regarding the effectiveness of cheap talk at eliminating hypothetical bias (Penn and Hu, 2019;Liu and Tian, 2021;Murphy et al., 2005).Because the evidence on cheap talk is inconclusive, it is uncertain what effect it has in terms of increasing validity (Mariel et al., 2021).Therefore, the possibility of hypothetical bias should be considered when interpreting the results of this study.

Summary and concluding remarks
Climate change poses exceptional challenges to farm businesses, and the rising number of extreme weather events call for action in terms of climate change adaptation and mitigation.The cultivation of agroforestry and wood-based land-use systems could play a key role in making agriculture more resilient to climate change.This study analyzed farmers' dynamic adoption decisions regarding such systems under consideration of extreme weather periods.To this end, random utility theory was integrated with the concept of adaptive weather expectations.Methodologically, a DCE was conducted with farmers in Bavaria, Germany and combined with local weather data.For this analysis, a correlated random parameter model was used, which served as a basis for regional weather simulations.
The results of this study indicate that farms are generally reluctant to adopt agroforestry and short rotation coppice compared to crop farming.However, they are more likely to adopt these options in the medium to long term following an extreme weather event.Furthermore, characteristic weather response pathways were found.These pathways can be divided into three phases that reflect important resilience capacities: absorption, recovery and adaptation.In addition, these results show that policy makers can effectively promote the adoption of agroforestry through agri-environmental payments and the promotion of technological progress.Several robustness checks were conducted to assess the plausibility of our model.Overall, our results show that farms may be increasingly likely to switch to agroforestry and wood-based systems in response to regional weather extremes.
Finally, we want to outline potential pathways for future research.Firstly, it would be worthwhile to assess the statistical uncertainty of the simulations.This could, for instance, be done either by means of a (computationally very expensive) nonparametric bootstrap procedure or by switching to a (hierarchical) Bayesian estimation framework.It would also be interesting to evaluate the appropriateness of this study's approach for other climate change adaptation strategies beyond land-use.Furthermore, future research could consider directly asking farmers about their predictions of future climate conditions and the uncertainties associated with them.This could be another valuable avenue for understanding their decision-making process regarding agroforestry adoption.This approach could also serve as a validation tool for the proposed expectation formation process.Finally, we would appreciate similar studies in different regions around the world to get a better overall understanding of the causal links between climate change and land use.

Fig. 1 .
Fig. 1.Conceptual framework of farmers' decision-making process under consideration of several types of uncertainty regarding their decision stemming from their local weather history.Source: Own depiction based on McFadden (1973), Singh et al. (2020), and Ramsey et al. (2021).

Fig. 2 .
Fig. 2. The case study region Bavaria is a federal state of Germany and lies in Central Europe.The map on the right depicts the spatial distribution of the study participants.Points are slightly jittered to prevent overlapping and enhance the visibility of the data distribution.

Fig. 3 .
Fig. 3. Summary of the weather variables used for the estimation of the baseline model with lag structure 1-3 and 4-10.

Fig. 4 .
Fig. 4. Illustration of the composition of the weather variables as they enter the simulation scenarios and replace the original weather variables used for the RPL estimation.The replacement procedure is demonstrated of a one-year, two-year and three-year shock scenario.Longer-term shocks change accordingly.

Fig. 5 .
Fig. 5. Average marginal effect of a ceteris paribus change in the land use attributes on farmers' adoption probability expressed as percentage change.

Fig. 6 .
Fig. 6.Simulated probabilities from a 2018-like extreme weather event lasting one year.

Table 1
Description of attributes and levels.
b Attribute level that only applies for the status quo alternative.

Table 2
Description of the weather indicators as they enter the 2018-like shock simulations.

Table 3
Description of the simulation scenarios and corresponding attribute values.

Table 4
Sample description and comparison with the population mean.