Key messages

  • The use of meteorological data is a common source of errors in pest risk modelling.

  • We propose the more frequent use of mechanistic microclimate models.

  • These models estimate the actual conditions to which pest organisms are exposed.

  • They help us understand pest responses to climate and identify causes of pest model failure.

  • Routine integration of microclimate modelling into pest risk modelling systems is now feasible

Introduction

Plant pests reduce agricultural productivity and food security with annual losses reaching c.18% of global crop production (Oerke 2006; Savary et al. 2019; Fones et al. 2020). Significant economic costs are also imposed on the forestry sector where pest outbreaks reduce ecosystem function, resilience and biodiversity (Godfray et al. 2010; Bradshaw et al. 2016). The routine application of pesticides and other measures to control pests incurs additional economic, environmental and human health costs (Pretty et al. 2000; Pimentel and Burgess 2014; Bourguet and Guillemaud 2016). These impacts of plant pests are forecast to change as the distributions of many pest species shift poleward, and invasive species more readily establish, under warmer climatic conditions (Bebber et al. 2013, 2014).

Quantitative pest risk models form a key component of pest management; the ability to forecast and map the risk of pest outbreaks is essential for the deployment of effective monitoring and control methods (Rock and Shaffer 1983; Tonnang et al. 2017; Coop et al. 2020; Hemming and Macneill 2020). Various indicators of pest risk can be modelled, informed by pest biology and the mitigation options available. Quantitative measures of pest risk may be derived from models of pest survival, growth or development. For an insect pest, model outputs may be a predicted date of completing a key developmental stage, or indicators of population growth such as voltinism or the likely exposure to causes of pest mortality. For a fungal pest or pathogen growth, model outputs may describe risk in terms of the favourability of conditions for pest growth, reproduction or dispersal. Models may express the response of a ‘typical’ or ‘average’ individual, a set of individuals or that of a pest population.

Climate is fundamental for determining the conditions in which ectothermic pest species flourish, imposing limits on their distribution and governing the biological processes that determine individual fitness and population dynamics (Uvarov 1931; Taylor 1981; Gillooly et al. 2001; Damos and Savopoulou-Soultani 2011; Clarke 2017). Climate-related variables are therefore key predictors in nearly all pest risk modelling. Long-term pest risk projections are commonly based on current or future climate suitability to endemic or invasive pests (Meyerson and Reaser 2002; Venette et al. 2010; Magarey et al. 2011; Hlásny et al. 2021), whereas short-term projections, that improve the efficacy and lower the economic and environmental costs of pest control measures, are typically driven by seasonal weather and short-range forecasts (Brunner et al. 1982; Damos and Savopoulou-Soultani 2010; Crimmins et al. 2020; Barker et al. 2020). Models used to inform the management of insect pests are often driven solely by temperature (Taylor 1981; Gillooly et al. 2001; Damos and Savopoulou-Soultani 2011), while models of slug pests (Choi et al. 2004; Wilson et al. 2015) or of fungal pathogens and oomycetes, are commonly expressed as functions of temperature and moisture (Jacome et al. 1991; Huber and Gillespie 1992; Magarey and Sutton 2007).

Although statistical models have been employed in pest risk analyses, for example to extrapolate from existing distributions of pest species to predict their potential establishment and range in new regions or under different climate scenarios (Lantschner et al. 2019), mechanistic or semi-mechanistic approaches are generally preferred (Sutherst and Maywald 1985; Savary et al. 2019) due to the acknowledged limitations of purely statistical approaches (Sutherst 2014; Lantschner et al. 2019; Srivastava et al. 2021). The use of mechanistic models facilitates a ‘biophysical’ approach to pest risk modelling which has particular benefit when forecasting how pests respond to novel conditions, whether as a result of climate change, pest or host migration. Under such conditions, the relationships between variables described by purely statistical models are less likely to endure than those derived from explicit biophysical processes and interactions.

A key aspect of a biophysical approach is an ability to reliably measure or estimate the abiotic conditions under which organisms live. Unfortunately, the importance of climate-derived variables in pest risk modelling does not denote the suitability of standard meteorological data as model inputs. Indeed, the use of meteorological data has been identified as one of the most common causes of failure in pest forecasting (Magarey and Isard 2017) whether in the form of data from individual weather stations (Trnka et al. 2007; Elliott et al. 2011; Jacquemin et al. 2014; Babu et al. 2014), in-field weather loggers (Chuang et al. 2014) or as gridded climate datasets derived from interpolated meteorological station data (Jarvis and Baker 2001; Hijmans et al. 2005; Magarey et al. 2011; Kriticos et al. 2012; Fick and Hijmans 2017; Lantschner et al. 2019; Crimmins et al. 2020; Barker et al. 2020).

The problem is that neither the microclimates experienced by pest organisms nor their body temperatures are thermally coupled to standard meteorological data derived from weather stations (Baker 1980; Ferro et al. 1979; Ruesink 1976; Samietz et al. 2007; Wellington 1950). This divergence between meteorological data and the conditions experienced by organisms is not simply due to a mismatch between the spatial resolution of meteorological data and the size of organisms (sensu Potter et al. 2013). Weather stations, measuring ambient air temperature at 2 m above ground, are designed and located to minimise the effects of local topography and vegetation on energy fluxes between the biosphere and atmosphere (WMO 2018). Yet it is these same energy fluxes that determine the microclimate conditions experienced by organisms (Daly 2006; WMO 2018; Bramer et al. 2018). Furthermore, the complex interactions mediating the relationship between meteorological and microclimate conditions cannot be reduced to a simple correlative relationship (Monteith and Unsworth 2013; Briscoe et al. 2023).

In this paper, we argue that the continued use of standard meteorological data in pest risk analyses ignores advances in mechanistic microclimate modelling over the past fifty years (Waggoner and Reifsnyder 1968; Goudriaan 1977; Deardorff 1978; Ogée et al. 2003; Kearney and Porter 2017; Maclean and Klinges 2021). These models are based on well-established physical processes (Richardson 1922; Penman 1948; Monteith and Szeicz 1962) and are today able to model microclimate conditions, at fine temporal and spatial resolutions, across entire landscapes (e.g. Trew et al. 2022). We propose their application can help resolve key shortcomings in existing biophysical approaches to pest risk modelling.

High-resolution microclimate and other environmental data forms a key part of new approaches to agricultural and forest management, such as ‘precision agriculture’ (Robert 2002). These approaches use data generated by ‘in-field’ sensor technologies and remote sensing to tailor management inputs and activities, including pest control, to conditions within individual fields or plots (Schultze et al. 2021). The deployment and maintenance of an extensive network of sensors, however, remains expensive and labour-intensive (Skocir et al. 2021), especially outside of highly controlled or homogeneous environments. The integration of mechanistic microclimate models has the potential of lowering the cost and extending the scope of precision agriculture approaches.

In the first part of this paper, we consider how pest thermal responses are measured and represented in pest risk models. In doing so, we revisit long-standing debates on how the type of mathematical model, conditions of fitting and use of meteorological data in model application can all introduce significant bias to model results. Secondly, we consider how microclimate conditions diverge from local weather conditions and the implications for pest risk modelling. Thirdly, we outline the theory of mechanistic microclimate modelling and how recent technological advances are transforming access to and use of these models. In the final part, we outline potential applications, and remaining challenges, of using the outputs of mechanistic microclimate models as inputs to pest risk models. We argue that the integration of microclimate models provides a powerful heuristic tool to improve our understanding of pest ecology and the reliability of pest risk modelling systems.

Although we focus primarily on modelling the effects of temperature on insect pests, our arguments apply to the modelling of other climate-related conditions relevant to pest and pathogen risk, such as incident radiation, humidity, soil moisture and leaf wetness.

Modelling thermal response

Thermal response (or ‘performance’) curves, ‘TPCs’, (Angilletta 2009) describe the relationship between temperature and a biological function or process. TPCs are typically derived from empirical observations made at different constant temperatures under controlled conditions (Quinn 2017; Shi et al. 2017). The concept commonly underlies models of pest life cycles and phenology, voltinism and population dynamics.

When measured over a sufficiently wide range of temperatures, ectothermic organisms typically display a skewed, unimodal response to temperature (Vasseur et al. 2014). The most likely cause of this response are the thermodynamics of enzyme reaction rates and protein stability. The response can be described by the fitting of various types of nonlinear statistical models to empirical observations (Kontodimas et al. 2004; Damos and Savopoulou-Soultani 2011; Rebaudo and Rabhi 2018). Pest modelling typically applies TPCs to actual field conditions using rate summation, where the response (such as developmental completion) is measured by accumulating response fractions per unit time (Liu et al. 1995). A key assumption of the approach is that the thermal response at any given instant is independent of the wider thermal regime.

Despite the empirical evidence and theory supporting a nonlinear response to temperature as the norm, in practice the majority of invertebrate pest risk models apply a simplified linear response (Quinn 2017; Shi et al. 2017; Rebaudo and Rabhi 2018). In part, this reflects the greater effort and resources required to fit a nonlinear response across a temperature range. The use of linear responses is also justified by assuming organisms are rarely exposed to extreme temperatures (Campbell et al. 1974). Among the most common insect pest risk models are those that use units of heat accumulation, such as degree days, to predict developmental rates. In such cases, the x-axis intercept of the thermal response curve defines a lower limit for the calculation of heat accumulation and the reciprocal of the slope defines a thermal constant that designates the amount of heat accumulation required, for example, to complete a developmental stage or life cycle (see Fig. 1).

Fig. 1
figure 1

Demonstrating the effects of different temperature data on linear and nonlinear thermal response models. In a, response curves of linear (green) and Brière 2 nonlinear (violet) models of codling moth (Cydia pomonella) larval development (Aghdam et al. 2009) that are applied to the frequency distributions of three contrasting hypothetical temperature datasets (derived from normal distributions with different means and standard deviations). The range of temperature values where both model responses are approximately linear is also shown. In b–d, we show model responses to each temperature dataset and indicate both the mean response to the temperature distribution (dashed lines) and the response to overall mean temperature of the distribution (dash-dot lines) with colours corresponding to the model type as in (a). In (b), temperatures exceed the shared zone of linearity in model response and hence for the nonlinear model the mean response differs from the response to mean temperature, whereas the corresponding values for the linear model are the same. In (c), temperatures fall below the shared zone of linearity and therefore the mean response differs from the response to mean temperature for both linear and nonlinear models. In (d), temperatures vary predominantly within the shared zone of linearity and hence mean response and response to mean temperature are very similar for both response models. The differences between mean response and response to mean temperature of the different models demonstrate how both the choice of model type and the use of aggregated temperature data can introduce bias in the predictions of thermal pest response. Note response and temperature scales are identical for figures (b–d) although minimum and maximum values vary

Full size image

In principle, the type of model used to describe the thermal response, and the conditions under which it is fitted, should determine the temperature data used for model application. In practice, however, this logic is often inverted. Readily available weather station measurements, in the form of daily averages or degree days, have been widely used as inputs to pest response models (Babu et al. 2014; Magarey and Isard 2017; Crimmins et al. 2020; Barker et al. 2020). By using such data, the likelihood increases of values falling within the range of linear response (Fig. 1) and can suggest the suitability of a simplified, linear model of thermal response. However, the approach also introduces several well-documented sources of error that can affect the reliability of not only linear, but also nonlinear, response models.

Firstly, the statistical method used to calculate degree days (or daily averages) and the extrapolation of a linear response to higher temperatures, or use of subjective cut-offs, can all affect model predictions, as has been acknowledged and discussed for several decades (Allen 1976; Higley et al. 1986; McMaster and Wilhelm 1997; Bonhomme 2000).

Of greater significance is the implicit presumption when using spatially or temporally aggregated temperature data that pest response to mean temperature is the same as the mean pest response to fluctuating temperatures. It is a presumption that does not prevail when the underlying response is nonlinear and temperatures fluctuate significantly—irrespective of whether the TPC is simplified to a linear response or not. The implications of this mathematical property of nonlinear functions, referred to as Jensen’s inequality (Jensen 1906; Ruel and Ayres 1999), have been widely discussed in pest modelling literature by reference to rate summation (Tanigoshi et al. 1976) and the Kaufmann effect (Bryant et al. 1999; Ikemoto and Egami 2013). The key implication is that changes to temperature variance can be as important as changes to mean temperature in determining a biological response (Ruel and Ayres 1999; Vasseur et al. 2014; Bütikofer et al. 2020). Referring to Fig. 1, temperature fluctuation around Tlow will result in a higher accumulated response than that predicted from the mean temperature, while variation around the optimal temperature will result in a lower response. The magnitude of the divergence will depend upon the degree of nonlinearity and skewness of the response curve, as well as temperature mean and variation.

Finally, and most importantly, when models are fitted to observations made under controlled and constant conditions, ambient air temperature corresponds to pest microclimate and body temperature. No such correspondence exists when applied to field conditions where the use of meteorological temperatures, measured within standardised environments (e.g. Stevenson screen at 2 m above ground level), exclude many of the factors, such as direct sunlight and wind, that affect pest microclimates and their body temperatures.

In summary, the application of models using rate summation from TPCs requires not only addressing the nonlinearity of thermal responses and how they are studied, but also the spatiotemporal variation of temperature to which organisms are exposed in the field (Maiorano et al. 2012; von Schmalensee et al. 2021). Recent studies (von Schmalensee et al. 2021) have demonstrated how the use of nonlinear TPCs driven by microclimate measurements can provide much more accurate predictions of pest biology than using linear degree day models or nonlinear TPCs driven by weather station data. These same studies suggest reliability is more affected by microclimate divergence than the parameterisation of TPCs under constant conditions. The divergence of pest microclimates from meteorological conditions is therefore a key element and challenge to the reliable application of pest modelling.

The importance of pest microclimates

The concept of microclimate does not simply refer to the spatial or temporal resolution of climate observations but incorporates the effects of topography and vegetation on energy fluxes and hence on local air and surface conditions (see Table 1). These effects vary ‘horizontally’ across a landscape and ‘vertically’ relative to a canopy, trunk, leaf or ground surface. A range of empirical and theoretical studies have shown how differences in microclimate conditions can affect the growth, development, mortality and diversity of pest organisms (Baker 1980; Pincebourde and Woods 2012; Saudreau et al. 2013; Faye et al. 2017; Pincebourde and Casas 2019).

Table 1 Primary abiotic and biotic properties, and the associated mechanisms, which determine temperatures and other conditions measured at different climatic scales
Full size table

The divergence between microclimate and meteorological conditions varies according to the type of microhabitat, prevailing weather conditions, the spatiotemporal scale and the type of metric. Locations near the top of vegetative canopies typically experience greater extremes than ambient air temperature during the day. Greater heat accumulation at the top of the canopy occurs because radiative heating of leaf surfaces generally exceeds cooling except under conditions of low sunlight. Locations lower within the canopy are buffered from air temperature extremes experienced above the canopy (De Frenne et al. 2019, 2021; Haesen et al. 2021). This relationship between microclimate and ambient air temperatures varies under different conditions. Summer temperatures beneath a dense canopy are generally lower than above-canopy air temperatures due to canopy shading, but reduced air movement within the canopy can cause warmer relative temperatures under certain wind conditions.

The divergence between meteorological data and microclimate conditions varies not only relative to a canopy or ground surface, but also spatially across a landscape. Many of these landscape effects are often termed ‘mesoclimatic’ (Maclean et al. 2019) and include the effects of elevation, topographical wind sheltering, cold-air drainage and coastal exposure. At high spatial resolutions, the effects of slope and aspect on the exposure to direct solar radiation can significantly affect surface and beneath-canopy temperatures. Regions characterised by more variable topography will often result in more variable microclimate conditions. In alpine environments, for example, diurnal and seasonal averages may diverge from nearby weather station data by up to 20 °C (Scherrer and Körner 2011).

Microclimate conditions therefore not only play a role in determining current geographical ranges, but also mediate biotic responses to climate change and thereby the propensity of current ranges to enlarge or diminish under future conditions (Maclean and Early 2023). The importance of microclimate conditions may be most significant at the margins of a pest range where it is more subject to climate-related limiting factors. Microclimates can permit species to persist under otherwise unfavourable macroclimatic conditions, or conversely, cause local extinctions despite a favourable climate (Suggitt et al. 2018; Ma and Ma 2022). Depending on the type of habitat and weather, a microclimate may reduce or increase exposure to the extreme temperatures that can be a direct cause of pest mortality (Alford et al. 2018; Pincebourde and Casas 2019). Winter minima temperatures, for example, are often associated with cold-air pooling and temperature inversions (Cooke and Roland 2003) that are influenced by fine-scale topographical features and rarely captured by standard weather station measurements (Schultze et al. 2021). Conversely, under dense canopy, minimum temperatures are generally higher than in open areas due to the effect of vegetation (De Frenne et al. 2021).

The importance of microclimate conditions to the reliable application of pest risk and crop growth models is well-established (Cellier et al. 1993; Bonhomme 2000) but not often captured. Where pest risk modelling has explicitly addressed the importance of microclimate, it has often done so by estimating local mesoclimate or microclimates from meteorological measurements by statistical downscaling methods using covariates of those properties, such as elevation or aspect, thought to affect local conditions (Royer et al. 1989; Marques da Silva et al. 2015; Rebaudo et al. 2016; Zellweger et al. 2019; Ogris et al. 2019). These estimated local temperatures are then used to drive thermal response-type models parameterised under constant, controlled conditions. An alternative approach uses models parameterised from observed pest distributions or behaviour in the field and relates these directly to local meteorological measurements.

In certain cases, different suites of models develop according to whether they are derived primarily from controlled experiments or from field observations. For the bark beetle, Ips typographus, one set of models uses estimates of bark temperature to drive nonlinear response models parameterised under controlled conditions (Baier et al. 2007; Ogris et al. 2019). Another set of models uses meteorological data to drive a degree day response parameterised from field observations, to map pest ranges under current and future conditions (Jönsson et al. 2007, 2011; Bentz et al. 2019).

Although the second approach does avoid discrepancy between the temperature data used in model parameterisation and model application, it does not address the issue of microclimates.

Whereas the first approach explicitly downscales meteorological data to local conditions, the second approach subsumes the processes relating weather to microclimate into the model of pest response to meteorological conditions. Both approaches suffer the same limitation; namely that the relationships they describe are valid within highly prescribed spatiotemporal limits. These limits are imposed by the many and complex physical processes that relate macroclimate to microclimate conditions. These processes vary in relative importance according to topographical, vegetative and weather conditions. Models depend on local parameterisation (where this is possible) and are unreliable when applied to novel combinations of geography and climate.

The adoption of a biophysical approach, that includes direct measurement and/or mechanistic modelling of the abiotic conditions to which pest organisms are exposed, offers a way to improve the reliability of pest risk modelling. Such an approach can explicitly address (i) the physical mechanisms determining the relationship between meteorological conditions and those experienced by pest organisms, (ii) how these processes vary spatially and temporally across the landscape, and (iii) how microclimate conditions relate to the body temperature of pest organisms. The development of more performative and accessible mechanistic microclimate models over recent years, offers the potential of their routine incorporation in integrated pest risk modelling and mapping.

Mechanistic microclimate modelling

Physical theory and model methods

The mechanisms that determine microclimate conditions reflect well-understood physical processes (Bramer et al. 2018) expressed by equations for the exchange of energy and mass between different components of the environment and with organisms within that environment (Campbell and Norman 1998; Briscoe et al. 2023).

The mechanistic modelling of above-canopy microclimates has its origin in the work on weather forecasting by Richardson (1922), who demonstrated the basic laws of turbulent mixing in the surface layer of the atmosphere. Theory and methods were subsequently developed in the agricultural and forest sciences and have been well-established since the 1950s (Penman 1948; Monin and Obukhov 1954; MacHattie and McCormack 1961; Monteith and Szeicz 1962; Allen et al. 1976; Goudriaan 1977). Local conditions relative to a ground or vegetative surface are modelled using the physical processes that govern energy exchange (sensible, latent and radiative heat fluxes). By including a term for the rate of heat storage (by the ground or by vegetation), the energy fluxes can be assumed to reach steady state allowing equations to be rearranged to solve for temperature iteratively or using for example the Penman–Monteith equation (e.g. Jarvis & Stewart 1979).

The effects of terrain, vegetation and soil properties are incorporated by calculating their effects on components of the energy budget. The incorporation of mesoclimate processes, such as lapse rates, coastal effects and wind sheltering effects of terrain, must be informed by whether the weather data used as model inputs have partly or fully accounted for such effects. Local field measurements may capture many effects while the gridded products interpolated from weather stations often include correction for elevation, coastal and urban heat effects (Hollis et al. 2019) to the spatial resolution of the final product.

Beneath a canopy, vegetative properties and canopy structure affect radiation transmission, vapour exchange and wind profiles, which in turn determine heat exchange, air and surface temperatures. Modelling microclimates beneath canopies, and within the soil, is typically undertaken by estimating heat and vapour fluxes between a series of homogeneous horizontal layers (Maclean and Klinges 2021). Establishing temperature profiles requires a theory that quantifies how heat is exchanged between these levels in the canopy. Doing so using the same principles of heat transfer that are used in above-canopy models would imply that vertical heat exchange occurs from warm to cool parts of the canopy in a predictable way determined by average wind speed (i.e. a gradient diffusion process). However, accumulated evidence (reviewed by Finnigan & Raupach 1987, and Raupach 1988) shows that this is not a valid assumption as frequent and pronounced counter gradient fluxes can be observed (Denmead and Bradley 1985, 1987). An alternative theory for modelling microclimate within canopies has been developed by adopting a Lagrangian (fluid-following) viewpoint (e.g. Raupach 1989). Here, the dispersion pattern of heat is assumed to be determined by the motion of the air that exchanges heat from multiple leaves within the canopy. Of dominant importance is the effect of the persistence of the turbulence relative to the travel time of the heat. At any given point in the canopy, heat arrives from multiple sources with varying travel times, but by assuming the canopy to be horizontally homogeneous, it is possible to find a computationally tractable solution to derive the temperature profile. These principles form the basis of several more recent below-canopy microclimate models (e.g. Bonan et al. 2021; Ogée et al. 2003) meaning that it is possible to estimate the microclimate temperature of almost any canopy environment.

Mechanistic microclimate models are typically used to return time series at specified heights or depths relative to the ground and/or canopy surface. Time increments can vary from seconds to several days, and outputs are usually for single locations or gridded points across a landscape. The outputs variables will typically include not only air or surface temperatures but also humidity, wind speeds and solar radiation which in turn allow integration with the biophysical modelling of pest organism body temperatures if required.

Model inputs and interfaces

Awareness of mechanistic microclimate models has permeated slowly across disciplines within the biological sciences. Early use was largely limited to the study of relatively homogeneous environments, such as crop fields, for which vegetation, topographical and other parameters could be measured in situ and/or constant values applied. Only in recent years have we begun to realise many potential applications of these models (Briscoe et al. 2023).

Microclimate models are typically run at high temporal and spatial resolutions, making model execution computationally intensive. The array of data inputs (see Fig. 2) and significant processing resources required for microclimate modelling presented major barriers to the application of these models only a decade ago (Sutherst 2014). Today, many of the same technologies that are transforming agricultural practices—including numerical weather forecasting, geographical positioning systems, in-field sensors, remote sensing and unmanned aerial vehicles—also facilitate the routine and extensive use of microclimate models.

Fig. 2
figure 2

Schematic of how microclimate models can be integrated within a biophysical approach to pest risk modelling. Schema shows key data inputs (pale blue) and outputs (blue), essential and optional modelling steps (solid and dotted green rectangles respectively), stages where pest ecology and functional traits may be incorporated into modelling process (see Table 2 and main text for further detail). Typical variables and formats used by microclimate models are provided (constant values , arrays , 2D spatial or 3D spatiotemporal grids). Microclimate model box indicates major aspects of the model, key radiative (shortwave direct, diffuse and longwave net) and heat fluxes (primarily convective and latent heat fluxes within canopy, and conductive through soil layers) and indicative profiles in wind speed and air temperature relative to canopy. Pest body temperature typically determined by surface and air temperatures of microhabitat but radiative and convective heat fluxes may also be important

Full size image

Key to microclimate modelling is the climate forcing data used to drive models. Model re-analyses data, such as ECMWF Re-Analysis 5 (Hersbach et al. 2020) and NCEP re-analysis dataset (Saha et al. 2014), provide a physically consistent, hourly description of global atmospheric characteristics, including cloud cover and incoming radiation. These datasets assimilate empirical weather station observations with short-range, weather forecasting models to generate low spatial, but high temporal, resolution climate data. Remote sensing products provide high spatial resolution topographical and land surface data, including properties such as albedo (Bonafoni and Sekertekin 2020).

An additional requirement of many microclimate models are descriptors of vegetation structure, such as plant area index and canopy height, to permit the quantification of microclimate processes such as radiation transfer through a canopy. For some applications, constant values may suffice, or dynamic canopy or crop development models can be applied to simulate seasonal changes in vegetative properties. Such models may be driven by the same climate inputs already described.

In other cases, the direct or remote measurement of actual canopy characteristics may be required, either as direct inputs into microclimate models or as drivers of crop growth or canopy development models (Toda and Richardson 2018; Kasampalis et al. 2018). Novel data products derived from sensors such as light detection and ranging (LIDAR), digital aerial photogrammetry, thermal radiometry and synthetic aperture radar (Jensen et al. 2021; Steele-Dunne et al. 2017), including those mounted on the International Space Station (Fisher et al. 2020; Dubayah et al. 2020), demonstrate the ability to quantify the vertical structure and properties of plant canopies across a landscape. The availability of global datasets of climate, land surface and increasingly canopy structure, combined with faster processing and programmatic access, today allows the near real-time modelling of microclimates across landscapes at high spatial and temporal resolution (Zellweger et al. 2019; Kearney et al. 2020; Duffy et al. 2021; Klinges et al. 2022).

Programmatic access to the inputs of microclimate models (Kemp et al. 2012; Kearney et al. 2020; Klinges et al. 2022) also facilitates the development of on-line portals that generate microclimate conditions (e.g. http://bioforecasts.science.unimelb.edu.au/app_direct/soil/), improving access to microclimate modelling for non-specialist users. Existing portals generate time series for single point locations, but could be extended to provide more extensive microclimate modelling for specific crop or forest systems. Faster, high-level programming languages (Bezanson et al. 2017), grid-modelling frameworks, access to cloud computing services (Yang et al. 2017) and statistical emulations of mechanistic microclimate models using Gaussian processes (Baker et al. 2022; Gómez-Dans et al. 2016) all offer the opportunity for more rapid modelling over larger geographical and temporal extents.

Recent years have also seen the publication of freely available regional, continental and global datasets of microclimate measurements. These include hourly gridded temperature estimates of above and below ground microclimates under different shade conditions or substrate types (Kearney et al. 2014; Kearney 2019) and a compilation of soil and near-surface temperature measurements from across the globe (Lembrechts et al. 2020).

The range of potential data inputs, algorithmic methods and precomputed microclimate datasets is such that guidance concerning their suitability for different end-uses is now required (Meyer et al. 2023). Technological advances have also greatly facilitated the validation of microclimate model outputs, whether by direct measurement of remotely sensed surface temperature (Marques da Silva et al. 2015; Zellweger et al. 2019) or the availability of reliable and cheap temperature and humidity micro-loggers (Lembrechts et al. 2020; Maclean et al. 2021).

Applying microclimate models to pest risk assessment and management

The greatest benefit of mechanistic microclimate models will be realised when they form part of a broader biophysical approach to pest risk modelling. Potential applications of the approach include the short-term forecasting of pest risk, typically to inform the timing and targeting of pest monitoring or treatments, as well as longer-term risk mapping of potential shifts in geographical range of invasive or migratory species and/or as a response to climate change (Kearney and Porter 2009). Microclimate conditions under future climate scenarios (Maclean 2020; Briscoe et al. 2023) can be modelled using probabilistic estimates of future projected climate from climate/earth system models (Smith et al. 2017) or by using the outputs of regional climate models. The outputs can be used to estimate pest risk under future climate scenarios in an analogous approach to that used with crop growth models to forecast the suitability of novel crops (Gardner et al. 2021).

There remain, however, many challenges to realising the potential of mechanistic microclimate models as a component of pest risk modelling and analysis. The integration of microclimate models in pest risk modelling requires a sufficient understanding of pest ecology to provide information on the microhabitats occupied by the organism at key stages in its life cycle. Existing microclimate models may need to be extended to capture the conditions of specific habitats. For example, modelling the conditions experienced by wood-boring beetles, a group of pests of significant ecological and economic concern, would need to capture heat fluxes within the host tree trunk. Field observations of whether pests display a preference between available microhabitats, such as height within a canopy, or the sunny or shady sides of a tree trunk (Gent et al. 2017), can also inform the choice of model inputs and parameters.

In general, the body temperature of small ectotherms with a high surface area to volume ratio remains closely coupled to surrounding air and/or surface temperatures, particularly where wind speeds are non-negligible. Pest physiology and behavioural responses will nonetheless mediate how microclimate conditions affect their body temperatures. Where necessary, heat fluxes between an organism and its surface and air microclimates can be integrated into microclimate models to capture, for example, convective and radiative heat transfers (Pincebourde and Woods 2020; Kearney et al. 2021).

Microclimate modelling also renders explicit the selection of biologically relevant spatial and temporal scales used to inform risk modelling. This choice of scale must be informed by the type of pest response and microhabitat that are modelled. Whereas models of insect development or phenology may be driven by daily data, measures of pest exposure to extreme temperatures may require modelling at higher temporal resolutions. Similarly, a lower temporal resolution may prove adequate for modelling buffered, soil microclimates, than for fluctuating microclimates like leaf surfaces. The choice of spatial resolution when modelling microclimates across a landscape will also need to consider variation in the processes and underlying properties determinant of microclimate conditions. A higher spatial resolution may be required to capture the effects of mountainous or hilly terrain.

The integration of mechanistic microclimate modelling must reflect the end-use of pest risk modelling, and tolerance to different kinds of error. If mapping a measure of pest risk across a landscape grid, then it may be appropriate to model the most and least favourable microclimates within individual grid cells. For example, where a pest can occupy habitats throughout a canopy, modelling microclimates at the top and bottom of the canopy can provide a plausible range of risk. Appropriate weighting might be applied to reflect the propensity of different habitats available to a pest or its typical distribution within a canopy. In cases where pest risk mapping does not need to reflect actual landcover, the adoption of canopy parameters typical of a habitat may be appropriate. Conversely, in cases where it is important to reflect actual landcover characteristics, remote sensing observations or in-field sensors can provide estimates of actual vegetative properties.

In all cases, the computational demands of microclimate modelling will also need to be weighed against potential benefits. Even with more efficient algorithms and exploitation of cloud computing, the mapping of multiple microclimates across extensive spatiotemporal extents requires significant resources in terms of time and computational resource.

Microclimate modelling as a heuristic tool

Progress in the development of more reliable pest risk models has been hampered by a dearth of published information about the occurrences and causes of model failure (Magarey and Isard 2017), augmented by the use of model systems which conflate different sources of error. This includes the use of meteorological data as a descriptor of pest abiotic conditions. As a result, it has proved difficult to resolve important questions about the constancy of thermal response models derived from controlled environment studies, or the relative importance of pest adaptation or acclimatisation to abiotic conditions (von Schmalensee et al. 2021).

We believe the use of mechanistic microclimate models offers a valuable heuristic tool to help discriminate between potential causes of model failure. The validity of thermal response curves continues to be debated when used to predict pest response under fluctuating field conditions (Niehaus et al. 2012; Colinet et al. 2015; Khelifa et al. 2019; Ma et al. 2021). It is known that the thermal response of ectotherms can vary between different organisms and populations of the same species, even if the extent of such variation remains uncertain (Chuine and Régnière 2017). The speed and duration of temperature fluctuations can affect insect response to high or low temperatures through thermal stress accumulation, temperature-induced or time-dependent acclimation (Sinclair et al. 2016; Sears et al. 2019). Emerald ash borer pre-pupae, for example, have greater heat tolerance when shifted slowly to high temperatures (Sobek et al. 2011). Under certain conditions, some invertebrates may even display semi-permanent endothermy, maintaining higher body temperature than their environment (Dinets 2022).

The use of meteorological data as a measure of the conditions to which pests are exposed conflates potential biological explanations of why a model may fail to describe responses in the field, with physical processes mediating effects on microclimate and with mathematical effects such as Jensen’s inequality. Mechanistic microclimate models provide a means to disentangle these different effects and assess different biological explanations of the discrepancies between observed and predicted pest response. Their use complements carefully designed empirical studies and re-analyses of published data which suggest thermal response models are reliable subject to consideration of nonlinearity and spatiotemporal heterogeneity in microclimate conditions (von Schmalensee et al. 2021).

A biophysical approach also permits better understanding of how behavioural and physiological adaptations mediate pest response to climate, whether by selecting from available microclimates, modifying their microclimate conditions or altering the effect of their microclimate on body temperatures (Table 2). In such cases, the thermal response model remains valid if driven by suitable data. With an adequate knowledge of pest biology, we can adapt mechanistic microclimate models to capture many of these effects.

Table 2 Pest adaptations influencing their response to local climate conditions, underlying biological processes and implications for biophysical microclimate—pest risk model systems
Full size table

For example, if it is known that a pest can actively move between different available microclimates, we might assume it will select the microclimate where conditions are closest to their optimal response. Modelling microclimate conditions at contrasting locations within a host plant canopy, at an appropriate temporal resolution that reflects the mobility of the organism and microclimate variability, can provide a range of available microclimate conditions from which the most optimal conditions could be selected. Adaptive responses are likely to reduce exposure to temperatures lying outside the zone of linear response (see Fig. 1), but it cannot be inferred that the modelling of microclimate conditions is less important as this also depends on the coupling between microclimate and meteorological conditions.

Modifications of pest microclimates through feeding and/or the creation of specialist structures such as leaf mines may require the integration of additional models (Pincebourde and Casas 2006). Integrated biophysical models offer a means of capturing how plant parameters are changed by pest feeding activity (Pincebourde and Casas 2019) and their resultant impact on microclimate conditions.

The use of microclimate modelling with dynamic plant canopy and pest developmental models also offers the possibility of better understanding how the correspondence between plant and pest phenology varies, an important aspect of pest risk, or of informing control methods based on the management of canopy structure (Barradas and Fanjul 1986; Huber and Gillespie 1992; Pangga et al. 2011; Caffarra et al. 2012; Saudreau et al. 2013). A better understanding of the spatiotemporal distribution of abiotic conditions to which pests are exposed can help improve our understanding of the limiting factors that act upon pest populations. Where available microclimate conditions approach optimal conditions, biological limiting factors may exert greater influence (Łaszczyca et al. 2021).

The integration of microclimate, plant canopy and pest body temperature models has been undertaken for over a decade (Saudreau et al. 2013) and such biophysical model systems have contributed to our understanding of pest biology and control. However, until recently it has been impractical to apply such an approach to risk modelling across entire landscapes and multi-year timeframes. We believe the new generation of microclimate models removes many of these barriers to their use and can make a significant contribution towards more reliable pest risk modelling.

Conclusion

Recent advances in remote sensing and computing resources have greatly expanded the capacity of microclimate modelling to estimate conditions across wide spatiotemporal extents at high resolutions. Although there remain challenges to the modelling of processes determining canopy microclimates, the fundamental principles of mechanistic microclimate modelling have been resolved for several decades.

Mechanistic microclimate models bring significant benefits to pest risk modelling where the widespread use of meteorological data as model inputs conflates various physical, experimental, mathematical and biological mechanisms that can undermine the reliability of risk forecasting. By approximating the conditions experienced by pest organisms and their body temperatures, microclimate estimates, (i) more closely correspond to the data used in the development and fitting of thermal response models under controlled environment conditions; (ii) provide a better estimate of pest exposure to extreme temperatures; (iii) capture spatiotemporal variations and interdependences of landscape, vegetative and climate characteristics that determine the conditions to which pests are exposed; and (iv) help discriminate between possible causes of model failure.

To fully exploit the benefits of microclimate modelling, nonlinear pest models that better reflect thermal responses at extreme temperature need to be adopted. Application to pest risk management also requires an adequate understanding of pest habitat and ecology, including how ecological, behavioural and physiological adaptations mediate the relationships between body temperatures and weather conditions. We believe the adoption of mechanistic microclimate modelling is a key element for building a more robust biophysical approach to pest risk management, informed by pest ecology, the physical determinants of microclimates and the practical requirements of end users.

The application of mechanistic microclimate models helps refocus attention on pest ecology, in particular the availability and occupancy of microhabitats that determine the abiotic conditions to which pests are exposed, and those behavioural or physiological adaptations that mediate the relationship with pest body temperature.