1. Introduction
There is great interest in using near-target remote sensing techniques such as time-sequential thermography (TST) in precision agriculture, ecology [
1] and phenomics [
2]. Thermography techniques have to address the thermal condition of the object of interest and the thermal and humidity conditions of the intervening atmosphere. Near-surface atmospheric temperature is influenced by synoptic weather patterns and their interaction with local topography at the smaller scale, which together determines the nature of the air turbulence that envelops the plant and controls the rate of water vapor and heat exchanges. On the other hand, plants are more than passive objects and employ stomata to sense the surrounding environment and respond rapidly to abiotic stresses, such as the air temperature. Their response is typically through stomatal conductance to water vapor and/or transpiration, which are critical physiological controls. The plant’s surface temperature, or its brightness temperature as sensed by a thermal infrared camera, is the result of the interaction of the air temperature and the plant’s physiological response. Thus, to understand the plant’s microclimate through thermography (or the environment that embodies the plant to a few orders of magnitude in spatial scale relative to the plant’s volume), it is important to understand the brightness temperature signal (measured by a infrared camera) as a function of near-surface meteorological parameters controlling the energy exchanges happening across the plant-environment envelope.
Land surface temperatures are influenced by surface energy balance [
3] especially when horizontal advection processes are negligible. Surface temperature varies as a consequence of partitioning of net-all wave radiation (Q*, or the balance between solar and infrared radiation input and output to the surface) into the subsurface conduction of heat (Q
G) and changes in sensible (Q
H) and latent heat exchange (Q
E) with the overlying atmosphere. On short time scales (less than an hour), radiative input is relatively constant, unless clouds interfere or overlying plant canopy causes rapid changes (flickering) in solar radiation [
4]. Higher frequency (seconds to minutes) surface temperature fluctuations are a response to the turbulent sensible and latent heat fluxes. Turbulence, caused by eddy motion, is expected to control temperature fluctuations on the same length and time scales as the atmospheric eddy motions.
The brightness temperatures of objects within the surface layer were not typically considered in the atmospheric community as a proxy for near-surface turbulence, but as infrared cameras become cheaper and are able to record data at high spatial and temporal resolutions, it is now feasible to study turbulence through the acquisition of brightness temperature. One of the earlier studies to investigate the coupling between coherent turbulent structures and surface temperature over an agricultural field (maize canopy) employed a directional infrared thermometer (sampling at 10 Hz) in identifying ramp structures in the surface temperature signal of the canopy with significant correlation with fluctuations in the air temperature above the canopy [
5]. Coherent structures were identified as temperature ramps in the surface and air temperature time series, with the magnitude of surface temperature ramps being significantly smaller than the air temperature ramps. Surface temperature ramps are caused by turbulent eddies mixing warmer (or cooler) air with cool (warm) air from aloft. A similar study was conducted over grass [
3] and also found direct relationships between surface brightness temperatures and independently measured surface-layer turbulence parameters.
Application of time-sequential thermography (TST) to calculate urban sensible heat fluxes (from a building) was first demonstrated by Hoyano et al. [
6], and was further developed conceptually by Voogt [
7] as a method for viewing the “footprint” of the coherent flow structures, and it was later emphasized by Christen et al. [
8] that brightness temperature fluctuations are largely controlled by atmospheric turbulence while the level of fluctuation becomes modulated by surface properties, especially its thermal admittance. The application of TST to detect large temperature fluctuations in the unstable surface layer to understand the turbulence structure has shown great promise in field experiments [
9], and was successful in deriving surface wind velocities over simple grass areas [
10] via the principle of turbulent eddy interaction with surface brightness temperatures.
As forward-looking infrared cameras become more affordable, TST will become an attractive method to measure the energy and moisture exchanges between the surface and overlying atmosphere. This research utilizes spatial brightness temperature data from infrared cameras looking onto a vineyard canopy. The canopy is also instrumented with an eddy covariance system measuring in situ turbulent and radiation fluxes and near-ground thermistor-based temperature sensors. The brightness temperature fluctuations (sampled sequentially over a nighttime period at a high frequency) are then used to interpret the spatial variability of the turbulent nature of the site using a combination of in situ metrological measurements and a pattern recognition algorithm (or self-organizing maps, SOM) applied to the acquired brightness temperature data. The SOM approach allows for clustering self-similar images into groups that could then be analyzed according to their unique meteorological context. This research highlights the significance and relevance of the methodology in terms of relating the brightness temperature variability to atmospheric turbulence, which also highlights the local meteorology. This approach is not only limited to vineyard applications and could be applied and assessed over various other crop types or surfaces.
4. Discussion
The results comparing the brightness temperature measured by the Optris infrared camera and in situ air temperatures show a very good match for the canopy height air temperature measurement and a warm bias for the air temperature measurement at the near-ground level (
Figure 3c). The FLIR camera results show a systematic cold bias, with larger temperature oscillations when compared with the Optris, which could be explained by the automatic focusing method employed by the FLIR camera, which tends to periodically auto-sharpen the image, but also could be explained by the need for a camera calibration.
Figure 3d shows a correlation diagram between the brightness temperature and the in situ air temperature measurement between the smaller area-averaged regions (1) and (2) and the larger field of view region (3). The results show a good linear correlation and a low value of root mean square error; they also show the cold bias offset previously revealed by the FLIR camera. This results also shows that the one-to-one relationship between the brightness temperature and the in situ air temperature is still preserved while spatially up-scaling the image over a homogenous terrain.
The brightness temperature signal of the Optris camera when compared to the direct measurements of turbulent heat flux (shown in red in
Figure 4b) follows the air temperature trend and responds to the cooling and warming period suggested by the heat flux advection. The brightness temperature oscillation range also scales to the ranges shown by the air temperature record. The FLIR brightness temperature trend (shown in blue in
Figure 4b) generally follows the initial cooling and then warming cycle but tends to overestimate the range with around five relatively large peaks. These peaks are linked to the automatic focusing of this specific infrared camera. Between 5 p.m. and 6:30 p.m. the brightness and air temperature trend did not exhibit either a positive or negative trend in comparison with the negative (cooling) or warming (positive) trends outside this period. This period also reflects a period of quiescence and little to no turbulent heat flux as shown in
Figure 4a.
The unsupervised clustering carried out by the SOM technique in
Figure 5 was successful in distinguishing nine clusters that have a meteorological context when compared to in situ measurements. The color in
Figure 5 represents the brightness temperature perturbation (red being a warming trend and blue a cooling trend). The SOM shows distinct features that are relatively different among nodes. For example, node 1 shows a field-wide warming trend, while node 9 a field-wide cooling trend. Nodes 4 and 6 show the same extremes but with lower magnitudes, while nodes 3 and 7 show a north to south cooling or warming gradient and appear to represent an opposite brightness temperature gradient. In order to link the SOM patterns to a meteorological context we have composited two different variables (A and B in
Figure 5 and Equations (1) and (2)) from the un-clustered data behind the construction of each of the node patterns. The mean sky brightness temperature perturbations, or A (varying between −0.17 and 0.17), correlate well with the brightness temperature trends. Positive values of quantity A (such as in nodes 1 and 4 for example) indicate a warming sky brightness temperature that relates to nocturnal cloud cover, which reradiates long-wave radiation back onto the surface, creating a homogenous spatial brightness temperature positive trend. The opposite applies for clear sky conditions (for example nodes 6 and 9). This result does not apply for nodes 2, 3, 5, 7, and 8 which show an order of magnitude lower mean sky brightness temperature perturbation and higher values of turbulent heat flux, which clearly creates localized and distinct warming and cooling trends within the brightness temperature spatial pattern. The particular patterns shown by nodes 3 and 7 are intriguing, as one could hypothesize that these opposite patterns are related to turbulent advection or mixing events that are happening within this local topographic catchment, especially as these patterns only exist during high turbulent heat flux periods (see quantity B for these nodes in comparison to quantity A for the other nodes).
5. Conclusions
We have demonstrated the use of forward-looking infrared cameras measuring the surface brightness temperature over a vineyard in the spectrum bandwidth of 7.5 to 14 μm at a relatively high temporal rate of 10 s for the application of vineyard-scale micrometeorology. Our results show that this technique, when applied for interpreting the micrometeorology as a function of cloud cover and within-canopy turbulence, could become a useful tool for up-scaling point measurements to spatially wide footprints. The temporal surface brightness signal over a few hours of the stable nighttime boundary layer intermittently interrupted by periods of turbulent heat advection was shown to be related to the atmospheric surface-layer dynamics observed by the eddy-covariance measurements, and reflects the temporal evolution of above-canopy wind variability.
The analysis also introduced the SOM of the spatio-temporal brightness temperature data to reduce the dimensionality of this large dataset, but more importantly to highlight the physical dynamics of nighttime surface brightness temperature over a complex canopy measured by an infrared camera. The resultant spatial clusters were self-organized and compared to the meteorological context they reflected, and the spatial brightness temperature pattern was explained in terms of the presence or absence of nighttime cloud cover and down-welling of long-wave radiation and the canopy turbulence heat flux. Time sequential thermography as demonstrated in this research provides positive evidence behind the application of thermal infrared cameras in the domain of micrometeorology. The results of this experiment could then be used in accordance with the surface renewal theory (which assumes that surface-atmospheric turbulence exchanges are driven by ramp-like structures within the temperature time series), which will eventually allow for a spatial pixel-based derivation of sensible and latent heat flux which are essential for the canopy’s water balance during daytime periods [
19,
20,
21,
22].
There are a couple of limitations to this study that need to be considered when it is applied for more complex terrain. The first limitation comes from the potential effect of air temperature and humidity fluctuations along the camera’s line of sight on the interpretation of surface brightness fluctuations. This effect is usually addressed by simple one-dimensional radiative transfer modeling, which delineates the role of infrared signal attenuation. Both of these effects have been previously found to be less than 10% (for atmospheric temperature) and less than 3% (for atmospheric humidity) [
8] for an urban setting and results may vary for other applications. The other limitation is the variable image pixel resolution as a function of depth. So the pixels furthest away from the camera have a different pixel resolution than pixels close to the camera. This could be only fixed with ortho-rectification when a high-resolution digital elevation map (DEM) for the site is available. A DEM was not available for this study, and given that the focus of this study was not to study atmospheric turbulence as a function of length scale, we considered that not affecting our major conclusions in this study.