Thermal Imaging and Infrared Sensing in Plant Ecophysiology

Abstract

This chapter outlines the potential of thermal sensing as a tool for plant ecophysiological studies and provides a summary of the key biophysical equations involved in the use of thermal sensing for the study of plant water relations. Particular emphasis is placed on the precautions that need to be adopted for high precision applications. The use of reference ‘mimic’ surfaces for improved estimation of stomatal conductance and evapotranspiration is outlined, and some of the precautions necessary for accurate work are described. Not only are recent applications reviewed, but some additional opportunities for use of the technique are described.

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

1 Introduction

1.1 Why Thermal Sensing?

Leaf temperature is important for plants both because of its effects on the rates of critical physiological processes and because of the damaging effects of temperature extremes. Notwithstanding these phenomena, remote estimation of leaf temperature also provides a powerful tool for the study of plant water relations and some other phenomena such as freezing. Most ecological and physiological applications of thermal remote sensing are based on the fact that the latent heat of vaporization required to evaporate from plant leaves is a major part of the leaf or canopy energy balance such that increasing evaporation or transpiration through the stomata leads to a cooling of the surface. The degree of cooling can therefore be used as an indicator of transpiration rate and stomatal opening, and hence as a measure of plant response to environmental stresses such as drought. A major practical problem in the application of this result in agriculture and in natural ecosystems is that, in addition to stomatal closure, many plant factors such as leaf orientation or reflectance and environmental factors such as radiation, wind speed and humidity also impact on leaf or canopy temperature over a rapid timescale. Practical application of this result for ecophysiological studies therefore requires approaches to allow for the various physiological and environmental determinants of canopy temperature (Jones 1999a, b, 2004), as will be discussed in detail below.

Thermal imaging also has applications in fields other than plant water relations. For example the high latent heat of fusion of water means that a large amount of heat is released when plant tissue freezes and various workers have used thermal imaging to follow the (rapid) progress of freezing (Fuller and Wisniewski 1998; Hamed et al. 2000; Carter et al. 2001; Stier et al. 2003). Similarly, thermal dynamics can provide useful information at scales from the functioning of ecosystems (Bendoricchio and Jørgensen 1997; Aerts et al. 2004) through soil moisture content estimation (Verstraeten et al. 2006) to the aerodynamic properties of single leaves (Jones 2014).

The principles underlying the control of leaf temperature were formalised by the early 1960s (Raschke 1956, 1960) and, with the development of infrared thermometers, were first used for the study of plant water relations by Tanner and Fuchs (Tanner 1963; Fuchs and Tanner 1966). Extensive subsequent studies by Jackson, Idso and colleagues (1977, 1981; Idso 1982) promoted thermal sensing as a tool for irrigation scheduling.

The greatest uptake of thermal sensing, however, came with the widespread availability of thermal imaging and especially, from about the year 2000, uncooled micro-bolometer sensors that operate in the long-wave (c.9–14 μm) thermal infrared (Jones 2004). For thermal sensing in the field it is essential to use only imagers that operate in the long-wave thermal region rather than shorter wavelengths as the latter are much more subject to interference by reflected sunlight (Leigh et al. 2006). Imagery has many advantages over the use of point sensors, especially the possibility of multiple replication within an image, and the ability to ensure that the temperature is that of the material of interest (e.g. a leaf) without any contamination by background.

Thermal imagery can be used in several ways for ecophysiological purposes. The simplest is in a purely qualitative mode where visual comparison of temperatures of different canopies can be used to identify plants, or areas of vegetation, under particular stress (where those areas having particularly high temperatures indicate stomatal closure). The second mode is semi-quantitative, and is widely used in plant phenotyping and involves ranking plots or genotypes for temperature (and hence stomatal closure). The most powerful approach, however, involves quantitative application to estimation of stomatal conductance, evaporation rate or aerodynamic properties based on application of energy balance and measurement of the necessary meteorological variables.

1.2 Basic Energy Balance Calculations

Application of thermal sensing as an ecophysiological tool is largely based on an understanding of leaf or canopy energy balance. Useful summaries of the energy balance equations and their application to thermal imagery may be found in a number of texts (Monteith and Unsworth 2008; Jones and Vaughan 2010; Jones 2014) and reviews (Jones 2004; Maes and Steppe 2012).

For a plant leaf (or equivalently for a canopy) at any instant, the net energy fluxes into and out of the system must equal the rate of energy storage

$$ {\mathbf{R}}_{\mathrm{n}}-\mathbf{C}-\lambda \mathbf{E}=\mathbf{M}+\mathbf{S} $$

(8.1)

where R _n, is the net heat gain from radiation, C is the net ‘sensible’ heat loss, λE is the net latent heat loss, M is the net heat stored in biochemical reactions (usually negligible) and S is the net physical storage. In the steady state net storage is zero, so the net radiation input is balanced by evaporative and sensible heat losses. All fluxes are expressed per unit leaf area as flux densities in W m⁻².

Leaf temperature affects all these fluxes, with sensible heat loss given by

$$ \mathbf{C}={\mathrm{g}}_{\mathrm{H}}\left(\rho {\mathrm{c}}_{\mathrm{p}}\right)\left({\mathrm{T}}_{\mathrm{s}}-{\mathrm{T}}_{\mathrm{a}}\right) $$

(8.2)

where g_H is the leaf conductance to heat transfer, ρ is the density of air, c_p is the specific heat of air, T_s is the leaf temperature and T_a is the air temperature (Jones 2014).

Similarly one can write the evaporative heat loss as

$$ \lambda \mathbf{E}={\mathrm{g}}_{\mathrm{W}}\left(\rho {\mathrm{c}}_{\mathrm{p}}/\gamma \right)\left(\mathrm{D}+s\left({\mathrm{T}}_{\mathrm{s}}-{\mathrm{T}}_{\mathrm{a}}\right)\right) $$

(8.3)

where g_W is the total conductance to water vapour, γ is the psychrometer constant, D is water vapour pressure difference between the inside of the leaf and the free air and s is the slope of the saturation vapour pressure-temperature curve.

Substituting these equations into the steady state energy balance and rearranging gives the leaf to air temperature difference as the following function of environmental and physiological variables:

$$ {\mathrm{T}}_{\mathrm{s}}-{\mathrm{T}}_{\mathrm{a}}=\left(\gamma {\mathbf{R}}_{\mathrm{n}}-{\mathrm{g}}_{\mathrm{W}}\rho {\mathrm{c}}_{\mathrm{p}}\mathrm{D}\right)/\left(\rho {\mathrm{c}}_{\mathrm{p}}\left({\mathrm{g}}_{\mathrm{H}}\gamma +{\mathrm{g}}_{\mathrm{W}}s\right)\right) $$

(8.4)

It is convenient to eliminate the need for a direct measurement of the net radiation actually received by the leaf by using the concept of net isothermal radiation (R _ni), defined as the net radiation that would be received if the leaf were at air temperature (Jones 2014), and rewriting Eq. (8.4) as

$$ {\mathrm{T}}_{\mathrm{s}}-{\mathrm{T}}_{\mathrm{a}}=\left(\gamma {\mathbf{R}}_{\mathrm{ni}}-{\mathrm{g}}_{\mathrm{W}}\rho {\mathrm{c}}_{\mathrm{p}}\mathrm{D}\right)/\left(\rho {\mathrm{c}}_{\mathrm{p}}\left({\mathrm{g}}_{\mathrm{HR}}\gamma +{\mathrm{g}}_{\mathrm{W}}s\right)\right) $$

(8.5)

where g_HR is the parallel resistance to sensible heat and radiation.

2 Applications

2.1 Stress Indices and Normalisation of Leaf Temperature

A number of approaches have been used over the years to develop semi-quantitative approaches to normalisation of leaf or canopy temperature to account for variation in environmental conditions. One of the main drivers has been the wish to develop indicators of plant water-deficit stress as indicated by stomatal closure. The simplest approach is to normalise data with reference to canopy air temperature, with early studies defining stress as the canopy-air temperature difference around mid-day (Jackson et al. 1977) or as the temperature difference between the canopy of interest and a comparable well-irrigated canopy (Gardner et al. 1981). More powerful was the introduction by Idso and colleagues (Jackson et al. 1981; Idso 1982) of a crop water stress index (CWSI; see Fig. 8.1) that takes account of variation in atmospheric humidity and makes use of the temperature of a notional non-transpiring canopy as well as the temperature of a well watered canopy transpiring at its potential rate (T_nws) using

$$ \mathrm{CWSI}=\left({\mathrm{T}}_{\mathrm{s}}-{\mathrm{T}}_{\mathrm{nws}}\right)/\left({\mathrm{T}}_{\mathrm{dry}}-{\mathrm{T}}_{\mathrm{nws}}\right) $$

(8.6)

Fig. 8.1

Illustration of the calculation of CWSI. The non-water stressed baseline temperatures are obtained empirically for a given crop/site combination. The upper limit represents the temperature of a non-transpiring surface in the same environment. For any measured canopy temperature, CWSI is calculated as x/y. The two broad arrows indicate potential errors in temperature measurement and show that a given measurement error has a much greater relative impact on errors in CWSI in humid environments than in more arid situations

This approach has been very widely used since the 1980s for irrigation scheduling, but it suffers from a number of disadvantages that limit its application in ecological systems and in more humid or temperate climates. In particular it requires a standard empirical non-water stressed base line derived for a comparable crop in a similar environment, but does not take account of any potential variation in irradiance or wind speed, though extensions that overcome some of these limitations have been suggested (Keener and Kircher 1983). Another problem with this formulation is that it is very non-linearly related to evaporation rate or stomatal conductance, which are usually of most interest in plant water relations studies.

The use of actual physical references measured simultaneously by thermography is a particularly powerful approach to improvement of this approach that reduces the effects of short-term environmental fluctuations. The use of reference surfaces to mimic dry or wet leaves (Brough et al. 1986; Qiu et al. 1996; Jones et al. 1997) allows one to eliminate the need for one or more of these environmental variables. Using such references having aerodynamic and radiative properties as similar as possible to real leaves, Jones (1999b) defined a stress index (SI_CWSI) analogous to Idso’s CWSI, but using a wet surface as reference (with infinite surface conductance), rather than a well-watered vegetation surface. Importantly, this paper also proposed a ‘conductance index’ (SI_gl) that is proportional to stomatal conductance and can be expressed as

$$ {\mathrm{SI}}_{\mathrm{gl}}=\left({\mathrm{T}}_{\mathrm{dry}}-{\mathrm{T}}_{\mathrm{s}}\right)/\left({\mathrm{T}}_{\mathrm{s}}-{\mathrm{T}}_{\mathrm{wet}}\right) $$

(8.7)

where T_wet is the temperature of a wet freely evaporating mimic leaf.

2.2 Use of Temperature Variance As a Stress Index

An alternative approach to the use of thermal data for the study of differences in stomatal conductance is based on the variability of temperature between different leaves in the canopy. This approach was proposed by Fuchs (1990) and is based on the idea that temperature variance increases as stomata close, because the importance of the radiative component of the energy balance increases as stomata close. This idea was further developed by Bryant and Moran (1999), who derived a histogram-derived crop water stress index (HCWSI) based on the skewness or kurtosis of the temperature frequency distribution. In principle the variance-based approaches will be most sensitive for sunlit leaves which maximise the temperature variation as compared with shaded leaves (Jones 2004). Unfortunately attempts to use these approaches have thus far had only variable success (Grant et al. 2007, 2016).

2.3 Stomatal Conductance and Evaporation

More quantitative data can be obtained by use of the full energy balance equation; for example Eq. (8.5) can be rearranged (Leinonen et al. 2006; Guilioni et al. 2008) to allow one to estimate the total conductance to water vapour according to

$$ {\mathrm{g}}_{\mathrm{W}}=\gamma \left(\left({\mathrm{R}}_{\mathrm{ni}}/\rho {c}_{\mathrm{p}}\right)-{\mathrm{g}}_{\mathrm{HR}}\left({\mathrm{T}}_{\mathrm{s}}-{\mathrm{T}}_{\mathrm{a}}\right)\right)/\left(s\left({\mathrm{T}}_{\mathrm{s}}-{\mathrm{T}}_{\mathrm{a}}\right)+\mathrm{D}\right) $$

(8.8)

Unfortunately, in addition to the requirement for an accurate estimate of leaf temperature use of this equation for estimation of conductance, requires information on air temperature, net radiation, humidity and boundary layer conductance which are not all easy to estimate at the level of the leaf. Useful reductions in environmental data requirements can be obtained by using temperatures of the ‘mimic’ reference surfaces discussed above. For example, measurement of the temperature of a dry reference surface that mimics the radiative and aerodynamic properties of the transpiring leaves (T_dry) allows one to eliminate R_ni from Eq. (8.8), giving (Leinonen et al. 2006),

$$ {\mathrm{g}}_{\mathrm{W}}={\mathrm{g}}_{\mathrm{HR}}\gamma \left({\mathrm{T}}_{\mathrm{dry}}-{\mathrm{T}}_{\mathrm{s}}\right)/\left(s\left({\mathrm{T}}_{\mathrm{s}}-{\mathrm{T}}_{\mathrm{a}}\right)+\mathrm{D}\right) $$

(8.9)

If one adds the temperature (T_wet) of a wet ‘mimic’ surface with no surface resistance to water loss, one can simplify this further (Guilioni et al. 2008; Leinonen et al. 2006). The precise formulation depends on whether the reference is wetted on one or both sides and whether the stomatal conductance is the same or different for the two leaf surfaces (Guilioni et al. 2008), however, for an amphistomatous (isolateral) leaf with a wet reference wetted on both sides this reduces to an estimate of the stomatal conductance (g_s) as

$$ {\mathrm{g}}_{\mathrm{s}}=\left(\left({\mathrm{T}}_{\mathrm{dry}}-{\mathrm{T}}_{\mathrm{s}}\right)/\left({\mathrm{T}}_{\mathrm{s}}-{\mathrm{T}}_{\mathrm{wet}}\right)\right)\ \left[2\gamma {\mathrm{g}}_{\mathrm{HR}}/\left(\left(\gamma {\mathrm{g}}_{\mathrm{HR}}/{\mathrm{g}}_{\mathrm{aW}}\right)+2s\right)\right] $$

(8.10)

where g_aW is the boundary layer conductance to water vapour for one side of the leaf. This equation is of the same form as Eq. 8.7, where the term in square brackets is a multiplier that depends primarily on the boundary layer conductance. Advantages of the use of reference surfaces include the fact that where all temperatures are measured by the same infrared thermometer/camera, errors in absolute temperature calibration can be eliminated.

In practice it is difficult to design a wet reference surface that remains wet continuously in hot, dry environments, though various designs have been proposed (Maes et al. 2016). Therefore most systems are based on the use of only dry references combined with humidity measurements (Eq. 8.9) with only a small reduction in accuracy (Leinonen et al. 2006).

2.4 Evaporation

Although it is the most difficult component of the surface energy balance to obtain from thermal sensing, rates of evapotranspiration (ET) can be conveniently estimated from remotely-sensed canopy temperature if appropriate assumptions are made (for a detailed discussion see Jones and Vaughan 2010). The best approach depends on the type of data available. The usual approach for satellite thermal imagery is to estimate regional ET as the residual term in the canopy energy balance using one-source energy balance algorithms such as SEBAL (Bastiaanssen et al. 1998a, b) and METRIC (Allen et al. 2007) or the more sophisticated two-source algorithms that treat vegetated and non-vegetated areas separately (Kustas 1990; Kustas and Anderson 2009).

Other more approximate approaches can be valuable in many situations. One of these is to make use of the fact that the upper limit of ET is set by the available energy (R_n) and that for large well vegetated areas ET is often well approximated by the ‘equilibrium evaporation rate’ (McNaughton and Jarvis 1983) given by

$$ \mathbf{ET}=1.26s\left({\mathbf{R}}_{\mathrm{n}}-\mathbf{G}\right)/\left(\lambda \left(s+\gamma \right)\right) $$

(8.11)

in which λ is the latent heat of evaporation, G is heat storage in the soil and the constant 1.26 is known as the Priestley-Taylor constant. Where ground cover is not continuous it is necessary to modify the calculation by multiplying the result by a ‘evaporative fraction’ that takes account of the fraction of the surface that is evaporating at the potential rate (Sobrino et al. 2007). This can be achieved by making use of the linear relationship between ET and surface temperature and combining this with remotely sensed information on canopy cover.

Temperature gives a direct estimate of evaporation rate, while combination with information on the fractional canopy cover (f_veg, often estimated from the Normalised Difference Vegetation Index (NDVI) as obtained from multispectral sensing (Jones and Vaughan 2010)) allows one to further derive information on the degree of stomatal closure (Fig. 8.2).

Fig. 8.2

Illustration of the general relationship between surface temperature and f_veg showing the trend to increasing Evaporation as canopy temperature decreases (T_a − T_s) increases. The solid circles show temperatures decrease as f_veg increases for dry soil and well watered plants, while the open triangles show the corresponding temperatures expected for water stressed vegetation showing increasing stomatal closure, increasing evaporation above that expected for transpiring vegetation alone. The open circles indicate samples with areas of wet soil or areas of open water. The line described by λ ET = s (R _n − G)/(s + γ) indicates the temperatures that correspond to the potential rate of ET as determined by incoming energy

2.5 Dynamics

The dynamics of temperature changes can provide important ecophysiological information, whether at the scale of single leaves or plants, or at the plant community or even regional scales. At the single leaf scale it can be used as an alternative method for estimation of aerodynamic resistances.

The rate of change of temperature (dT/dt) depends on the tissue heat capacity (ρ*c _p*𝓁*) according to (Jones 2014)

$$\mathrm{d}\mathrm{T}/\mathrm{d}\mathrm{t}=\mathbf{S}/\left({\rho }^{\ast }{c_{\mathrm{p}}}^{\ast }{\mathcal{𝓁}}^{\ast }\right)$$

(8.12)

where the ρ* and c _p* are the density and heat capacity of the tissue material, and 𝓁* is the thickness. S is the rate of heat storage ignoring any M (=R _n − C − λ E). Following a step change in the environment that affects the rate of heat storage (such as a change in radiation), leaf temperature changes according to the following

$$ \mathrm{T}={\mathrm{T}}_2-\left({\mathrm{T}}_2-{\mathrm{T}}_1\right)\ \exp \left(-\mathrm{t}/\tau \right) $$

(8.13)

where T₁ is the initial temperature and T₂ is the final equilibrium temperature and τ is the time for c.63.2% of the total change to occur (known as the time constant). The time constant is given by

$$\tau =r^{\ast }{c_{\mathrm{p}}}^{\ast }{\mathcal{𝓁}}^{\ast }/\left(r{c}_{\mathrm{p}}\left(\left(1/{\mathrm{r}}_{\mathrm{H}\mathrm{R}}\right)+\left(s/\left(\gamma \left({\mathrm{r}}_{\mathrm{aW}}+{\mathrm{r}}_{\mathcal{𝓁}W}\right)\right)\right)\right)\right)$$

(8.14)

which for a non-transpiring surface where r_𝓁W is infinite, can be simplified and rearranged to give the aerodynamic resistance as

$${\mathrm{r}}_{\mathrm{aW}}=\tau \rho c_{\mathrm{p}}/\left(\left({\rho }^{\ast }{c_{\mathrm{p}}}^{\ast }{\mathcal{𝓁}}^{\ast }\right)-\tau \rho c_{\mathrm{p}}/{\mathrm{r}}_{\mathrm{g}}\right)$$

(8.15)

Any suitable curve-fitting method can be used to estimate τ (and hence r_aW) from the time course of temperature change after an environmental perturbation. For non-transpiring leaves (for example covered in petroleum jelly) this provides a potentially useful method for estimation of aerodynamic resistances, though it does not appear to have been used previously, probably because it requires accurate estimates of the heat capacity per unit area of tissue.

For more massive bodies such as fruits, and at larger scales for soils or regions, where temperature does not rapidly equilibrate throughout the body, the time course of surface temperature change is more complex. In this case the rate of change of surface temperature depends on the thermal conductance from the surface down into the depths and on the heat capacity of the material. A high thermal conductivity leads to surface temperature changing relatively slowly for a given energy input giving a material with a high thermal inertia (a strong resistance to temperature change).

The thermal dynamics of fruit surfaces can provide a useful tool for the detection of internal bruising or other damage. For example, Baranowski and colleagues (2008) have demonstrated that the thermal inertial was substantially greater for apple fruits with water core than for healthy fruits and that bruising can be readily detected (Baranowski et al. 2009). This difference results from the higher tissue water content and hence higher thermal conductivity in the damaged fruits.

The high conductivity of water lies behind the widespread application of the use of thermal inertia, as estimated from diurnal changes in surface temperature obtained from satellite imagery such as from MODIS which has two overpasses per day, as an indicator of soil moisture content (Verstraeten et al. 2006; Veroustraete et al. 2012; Zhang and Zhou 2016). Soils with higher water content have significantly greater thermal inertia together with a delay in the phase maximum.

Thermal inertia, otherwise known as thermal buffer capacity, has been proposed as an overall indicator of ecosystem integrity, summarising biotic and abiotic controls over energy flows in ecosystems (Aerts et al. 2004). In practice such a measure is, however, dominated by differences in vegetation, with forests and high soil moisture favouring high thermal inertia, while dry desert or grassland will have the lowest thermal inertia (most rapid temperature changes).

2.6 Boundary Layer Conductance

A number of approaches are available for estimation of the boundary layer conductance required for application of the methods described above. The value of g _H can be estimated at a leaf scale from aerodynamic theory using the relationships between wind speed and leaf size from

$$ {g}_{\mathrm{H}}=6.62\ {\left(u/l\right)}^{0.5} $$

(8.16)

where u (m s⁻¹) is the wind speed and l (m) is the characteristic dimension of the leaf (Jones 2014). Alternatively, at a canopy scale, one can use wind profile theory (Monteith and Unsworth 2008; Jones 2014) to estimate g _aH as

$$ {g}_{\mathrm{H}}={\kappa}^2{u}_z/{\left[\ln \left(\left(z-d\right)/{z}_{\mathrm{o}}\right)\right]}^2 $$

(8.17)

where κ is von Karmann’s constant (=0.41), u _z is the wind speed at canopy height z, z _o is the roughness length (often assumed equal to 0.64*z), and d is the zero plane displacement (often assumed equal to 0.13*z).

A third approach that is well adapted to the leaf scale and for continuous monitoring in the field would be to calculate g _H from the temperatures of heated and unheated replica leaves mounted in the canopy (Brenner and Jarvis 1995).

A number of other, thermally-based approaches are at least theoretically possible. One of these is to use the cooling dynamics of heated leaves (as described above), while another would be to use thermal imagery with leaf ‘mimic’ reference surfaces wetted on 0, 1 or 2 surfaces.

2.7 Some Other Qualitative Applications

There has been a recent increase in the range of applications of thermal imaging in other eco-physiological studies. For example, thermal sensing can be a powerful tool for the study of plant disease, as many diseases lead to alterations in leaf temperature that can be detected by thermography, with the precise temperature responses and their sequence depending on the species/disease combination (Chaerle et al. 1999; Chaerle and Van Der Straeten 2001; Chaerle et al. 2004). Although mostly applied in a proximal or near-field situation for early detection of infections, there is evidence that useful information can be obtained from satellite or airborne thermal imagery (Lindenthal et al. 2005; Stoll et al. 2008).

Another area of interest has been in the study of thermogenic respiration in flowers (Skubatz et al. 1991; Bermadinger-Stebentheiner and Stabentheiner 1995; Lamprecht et al. 2002) as well as the role of solar heating in floral physiology, especially in arctic and alpine plants (Lamprecht et al. 2002, 2006; Dietrich and Korner 2014). Thermal imaging has also been used in forestry to investigate tree species diversity, with typical canopy temperatures being shown to vary between coniferous and broad-leaved forests (Leuzinger and Körner 2007).

3 Some Practical Aspects of Thermal Imaging

3.1 Absolute Accuracy

The absolute accuracy of thermal measurements is only critical for applications where absolute temperatures are required (e.g. for estimation of ET or stomatal conductance) and is not particularly critical for relative studies as in crop phenotyping. Most imagers currently available have thermal resolution fully adequate for plant water relations studies (i.e. <0.1 K), but their absolute accuracy is often no better than ±1 or 2 °C, so care is needed when absolute rather than relative values are of interest. This poor absolute accuracy is one of the strongest justifications for the use of reference surfaces in calculations because in such situations only temperature differences (which are generally available at higher precision) are required.

The accuracy of surface temperature measurement using thermography is critically dependent on the surface emissivity chosen and on settings for the ‘environmental temperature’ and for the air temperature, humidity and distance between object and sensor. The temperature of a surface (T_s) is normally estimated by inversion of the following equation

$$ \mathrm{R}\cong \tau\ \varepsilon\ \sigma {{\mathrm{T}}_{\mathrm{s}}}^4 $$

(8.18)

where R is the total radiant flux density (W m⁻²) received at unit area of sensor surface, τ is the longwave transmission by the atmosphere between the object and the sensor, ε is the surface emissivity and σ is the Stefan-Boltzmann constant (5.6697 × 10⁻⁸ W m⁻² K⁻⁴). Software in the sensor usually converts the received radiance to the radiance expected for the long wave band (correcting for the spectral sensitivity of the sensor) to produce an estimate of T_s. However, as shown in Fig. 8.3, the total longwave radiance arriving at the sensor is the sum of the thermal radiation emitted by the object (given by Eq. 8.18), the longwave radiation emitted by the atmosphere intervening between the object and the sensor, and the environmental longwave radiation reflected by the object towards the sensor, together with any of the emitted radiation absorbed by the intervening atmosphere.

Fig. 8.3

Illustration of the radiation streams impinging on a thermal sensor, all of which need to be allowed for to obtain accurate temperature estimates. These corrections are commonly available in the thermal imaging software provided with the thermal cameras. The required corrections are greater, the more remote the sensor is from the object

For close-range sensing (less than about 5–10 m) the atmosphere is normally assumed to have a negligible effect on the at sensor radiance, though at high humidities it can be necessary to make some correction. The object emissivity also affects the temperature estimate, through its effect on both the emitted and reflected environmental radiation streams. The environmental radiation impinging on the object can have quite substantial effects on the apparent object temperature, especially for surfaces with emissivities substantially different from 1; for example, dry soils may have emissivities substantially below 0.9. The importance of emissivity to the correct estimation of temperature is illustrated by the fact that a 1% error in emissivity can give rise to an error in the estimated temperature of more than 0.5 K (Jones and Vaughan 2010), suggesting that particular care is needed when comparing different types of surface in a single image.

3.2 Illumination of Object

The temperature of any surface is critically dependent on its exposure to the incoming solar radiation (Eq. 8.4). As pointed out by Jones et al. (2009) it follows therefore that leaves on the shaded side of trees will be cooler than those on the sunlit side. Similarly, even within a homogeneous canopy, the temperature of different leaves will vary over quite a wide range (Fuchs 1990) dependent on their incident radiation (which is a function of leaf orientation and any shading). A continuing discussion point is whether it is better to orient a thermal sensor with the sun behind the observer (to maximise the proportion of sunlit leaves), or whether better discrimination between plants is obtained when orienting the sensor towards the sun so that more of the observed leaves are shaded. The former has the advantages that temperatures are higher so that there is a greater temperature response for given differences in conductance, while conductances themselves tend also to be higher than for shaded leaves. The latter approach of studying shaded leaves has the advantage that temperature is much less sensitive to leaf orientation, but the lower temperatures and conductance can reduce sensitivity (Jones et al. 2002). Other studies, however, have suggested that thermal measurements on the top of canopies rather than the sides can produce more reliable estimates of conductance indices than viewing from the side (Grant et al. 2016), supporting the value of imagery from UAVs.

3.3 Design of Reference Surfaces

In order to give useful estimates of dry or wet reference temperatures for use in the equations outlined above, it is critical that the reference surface has similar radiative and aerodynamic properties to the surface of interest (e.g. a plant canopy). The choice of reference surfaces varies from the use of reference crops grown under similar conditions (Idso 1982) through simple temperature approximations (Irmak et al. 2000) to a range of physical references (Jones et al. 2017). For the dry reference it has frequently been proposed (Irmak et al. 2000; Ben-Gal et al. 2009; Meron et al. 2010) that approximating the dry reference as T_a + 5 °C is adequate in Mediterranean climates. Unfortunately as pointed out by Jones et al. (2017) there can in practice be substantial deviations from this simple approximation, especially in low radiation or high wind speed environments, leading to large errors in the derived estimates of conductance or ET. Similarly, large errors can arise if the spectral absorptance differs from the actual leaves (Jones et al. 2009) or where the size or thermal properties of the reference do not mimic the real leaves.

Most reference surfaces have been constructed as flat leaf replicas made of green filter paper or similar material (Maes and Steppe 2012), but such surfaces cannot effectively replicate the range of illumination experienced by typical leaves in a plant canopy. In order to better represent the temperature of the range of leaves in a canopy, especially as solar elevation changes during the day, Jones et al. (2017) proposed use of the average temperature of a hemispherical surface (constructed of a half table-tennis ball; Fig. 8.4) on the basis that for a typical canopy with leaves oriented according to a spherical distribution (Monteith and Unsworth 2008) this would provide a closer approximation of the effective mean leaf temperature than would a flat surface.

Fig. 8.4

Illustration of the ArduCrop dry reference sensor, showing the field of view of the downward-looking thermal infrared sensor and the dry reference hemisphere whose temperature is detected by an upward pointing infrared sensor. Power to drive the sensor and its connection to a local Wireless Sensor Network is provided by the solar panel shown. (see Jones et al. 2017, for further details)

Because the aerodynamic properties of the hemispherical reference may not exactly mimic those of leaves in a canopy, and also because of potential differences in solar absorptance from actual leaves, calculated estimates of conductance or ET may require modification by the use of a constant correction factor as outlined by Jones et al. (2017). A particular advantage of this type of reference is that it allows continuous monitoring of stomatal conductance and ET, allowing long-term and diurnal studies of water relations.

4 Conclusions

Thermal infrared sensing is becoming an increasingly powerful tool for the study of plant water relations and a range of other ecophysiological processes in both natural and agricultural ecosystems. Thermal imaging, whether from in-field sensors or from airborne sensors mounted on drones, aircraft or even satellites, has particular advantages over the use of simple infrared thermometers with their single field of view because of the ability to make use of information about the variance of temperature of any object. A further advantage of imagery is that it can also be readily combined with multispectral or hyperspectral imagery to enhance the information about the object, for example through separating plant material from background and allowing the extraction of temperature of the plant material without confounding by background (Leinonen and Jones 2004). Nevertheless, arrays of simple cheap infrared thermometers, especially when combined with dry reference sensors (Jones et al. 2017), can be valuable for long term remote monitoring of evaporation and stomatal conductance. The recent advances in automated image analysis have the potential to greatly increase the power of thermal imagery for the study of plant water relations as it is becoming possible to avoid or reduce what is often a labour-intensive step in the application of thermography (Fuentes et al. 2012).