Diurnal and Phenological Modulation of Canopy Temperature in Wheat Breeding Under Mediterranean Conditions

Abstract

Canopy temperature (CT) is widely used to assess crop water and heat status, but it is often recorded at a single hour, implicitly treating CT as a stable trait. Here, we show that canopy cooling is a dynamic phenotype whose expression depends on time of day, phenological stage, and environment. First, we monitored 184 spring wheat (Triticum aestivum L.) genotypes in two Mediterranean environments (fully irrigated vs. rainfed, contrasting atmospheric demand) using UAV-based thermal imaging. CT was measured six times per day (10:30–17:30 h) at four reproductive stages (anthesis, milk-grain, milk-dough, and dough), enabling quantification of diurnal plasticity, seasonal shifts, and environmental effects on canopy cooling. Second, repeated-measures mixed models confirmed that Location, Stage, and Time of day, and all interactions, were highly significant (p < 0.001). Variance-component analyses showed a strong genetic signal within each Stage × Environment combination, with 87.6–97.7% of total variance attributable to genotypic effects pooled across hours. Third, the optimal phenotyping window was context dependent: under rainfed conditions, genotypic discrimination consistently peaked around mid-afternoon (~15:00 h), whereas under irrigation, the optimal window shifted with stage (13:30–15:00 h). Genotype rankings were also markedly less stable across hours under rainfed conditions, indicating substantial within-day re-ranking as atmospheric demand increased. Finally, thermal exposure analyses showed that exceeding a physiologically relevant threshold (CT > 32 °C) depended strongly on time of day and stage; maximum CT captured short heat events missed by daily means. Clustering and alluvial analyses revealed frequent reclassification across stages, with only a small subset remaining consistently cooler, particularly under stress. Random regression of CT on vapor pressure deficit (VPD) indicated that CT–VPD sensitivity was largely environment-dependent and showed weak cross-environment correspondence (Spearman ρ = −0.166). Overall, single-time-point CT phenotyping provides an incomplete view of thermal status, underscoring the need for multi-temporal protocols and context-specific measurement windows for breeding and physiological interpretation under drought and heat.

1. Introduction

Canopy temperature (CT) has long served as an important indicator of plant–atmosphere interactions, reflecting the balance between absorbed radiation, transpirational cooling, and atmospheric demand. Early studies established CT as a practical method for assessing plant water status under field conditions, thereby linking canopy energy balance to environmental constraints [1]. With advances in infrared thermometry and thermographic techniques, CT measurements are now widely used in crop physiology and phenotyping studies, especially in cereals grown under water- and heat-stressed environments [2,3]. Recently, UAV-based thermal imaging has enabled plot-scale, repeated CT measurements across large breeding trials [4,5], supporting a shift from snapshot assessments to time-resolved phenotyping under field conditions [6,7,8].

In most field trials, CT is measured at a single time point, typically near solar zenith. This approach is operationally efficient and has been effectively used to compare genotypes or treatments under specific environmental conditions [9,10]. However, such measurements implicitly assume that CT remains a stable and representative trait of the genotype, capable of summarizing its canopy cooling plasticity throughout the day and supporting consistent phenotypic comparisons [11].

Since water flow throughout the plant varies in response to fluctuations in radiation, ambient temperature, vapor pressure deficit, and soil water content [12], CT is inherently dynamic and expected to change throughout the day. In this context, recent thermal phenotyping studies emphasize that CT should be seen as a dynamic canopy-level signal that shifts with environmental fluctuations and measurement timing, rather than a fixed indicator of “cool” versus “warm” canopies [13,14,15]. In fact, CT can fluctuate in magnitude and even in its relative ranking, depending on the measurement time and environmental conditions [16,17]. However, most phenotyping protocols still depend on a single daily measurement, with limited attention to how canopy temperature varies throughout the day and across different developmental stages.

Over the past decade, substantial evidence has shown that plant species exhibit a wide range of genotypic variability in transpiration rates (TR) in response to increased vapor pressure deficit (VPD) during the day; including cereals like wheat (Triticum aestivum L.) [18,19,20,21] on which two response types have been identified: a linear response, in which TR remains steady as VPD increases, and another, in which TR decreases after surpassing a threshold VPD [18]. This suggests that in environments where plants face terminal drought stress, adopting a more conservative (segmented) strategy could help limit water loss as VPD increases [18,20]. Similarly, genotypes from the same species display these two response types to increasing VPD: some do not reduce TR, while others maintain TR up to a specific VPD threshold before decreasing TR, as observed in wheat [21], soybeans [22,23], corn [24], cowpea [25], sorghum [26], and chickpeas [27,28].

Crucially, since transpirational cooling is a main factor influencing canopy temperature, genotype-specific TR–VPD response patterns are expected to lead to distinct diurnal CT trajectories in the field; especially during periods of rapidly increasing VPD [29,30]. This establishes a direct physiological link between the TR–VPD trend and UAV-based diurnal CT phenotyping; repeated thermal measurements throughout the day can pinpoint when genotypic differences in cooling emerge. They can also show under which atmospheric conditions those differences emerge and whether genotype rank order based on CT remains stable or shifts over time within breeding panels [7,13].

In Mediterranean environments, where temperatures and atmospheric demand often spike sharply in the afternoon, CT can reach critical levels during the reproductive period [31]. Temperatures around or exceeding 32 °C have been linked to critical responses in wheat reproductive tissues, especially during anthesis and early grain development [32,33]. Importantly, genotypes may vary not only in whether this thermal threshold is reached but also in the consistency with which canopy cooling is maintained throughout the day and across the season [13,34]. As environmental conditions become more challenging, fewer genotypes may be able to maintain lower canopy temperatures, indicating that phenotypic assessments under non-stress conditions do not necessarily reflect cooling capacity under stress [7,13].

Despite the recognized dynamic nature of CT, understanding how canopy cooling manifests across the diurnal cycle and varies during the reproductive season in active wheat breeding panels remains limited. Specifically, the extent to which diurnal plasticity, phenological changes, and environmental factors influence genotype re-ranking is still poorly documented under realistic field conditions.

The goal of this study was to determine whether single-time-point CT measurements accurately represent genotype-specific canopy cooling trends across different times of day, phenological stages, and contrasting Mediterranean environments. By collecting repeated UAV-based CT measurements at multiple time points across four reproductive stages under both fully irrigated and rainfed conditions. We aim to identify patterns of daily and seasonal plasticity in canopy cooling and to quantify the degree of genotype re-ranking.

2. Materials and Methods

2.1. Plant Material and Experimental Setup

A set of 184 advanced lines and cultivars of spring bread wheat (Triticum aestivum L.), exhibiting similar phenology and previously characterized variability under rainfed conditions, was selected from a larger panel of 384 genotypes [35]. They were tested in two different environments within Chile’s Mediterranean region in 2022. Cauquenes (35°58′ S, 72°17′ W; 177 m.a.s.l.), known for higher VPD and rainfed conditions (WS), represented the drought-prone area. Santa Rosa (Chillán) (36°32′ S, 71°55′ W; 217 m.a.s.l.), with irrigation and lower VPD, was the fully irrigated (FI) environment. Rainfall during the experiment was 468 mm in Cauquenes and 523 mm in Chillán.

The experimental design used an alpha lattice with two replicates. Plots consisted of five rows, each 2 meters long, with 0.2 m spacing between rows. The sowing rate was 20 g m⁻², and sowing took place on July 18. Fertilization involved 260 kg ha⁻¹ of ammonium phosphate (46% P₂O₅ and 18% N), 90 kg ha⁻¹ of potassium chloride (60% K₂O), 200 kg ha⁻¹ of Sulpomag (22% K₂O, 18% MgO, and 22% S), 10 kg ha⁻¹ of boronatrocalcite (11% B), and 3 kg ha⁻¹ of zinc sulfate (35% Zn). An additional 153 kg ha⁻¹ of urea was applied during tillering. Weed control included pre-emergence treatment (Flufenacet + Flurtamone + Diflufenican; 96 g active ingredient) and post-emergence treatment (MCPA + Metsulfuron-methyl; 525 g and 5 g of active ingredients, respectively). The cultivars were disease-tolerant, and no fungicides were used. For the fully irrigated (FI) setup, furrow irrigation was performed with three 50 mm irrigations at the flag leaf stage (Z37), heading (Z50), and mid-grain filling (Z70). The rainfed condition (WS) depended on natural rainfall until heading, after which a plastic shelter prevented further rain during grain filling.

2.2. Canopy Temperature

For each phenological stage (anthesis, milk-grain, milk-dough, and dough) and water regime (fully irrigated, FI; water-stressed, WS), canopy temperature (CT) was analyzed for its diurnal pattern: six times per day (10:30, 12:00, 13:30, 15:00, 16:30, and 17:30 h), and one day per phenological stage.

Measurements were taken using a DJI 200 V2 UAV (Dajiang Innovation Technology Co. Ltd., Shenzhen, China) equipped with a dual-camera Zenmuse XT2 (FLIR, Wilsonville, OR, USA), under the following criteria: (i) flight altitude: 30 m; (ii) image overlap: 85% (both horizontal and vertical); (iii) flight speed: 2.3 m/s; (iv) clear, sunny weather with wind speeds below 15 km/h; and (v) the camera triggered every 1.5 s by an intervalometer. The cameras were turned on 20 minutes before each flight to allow their internal temperature to stabilize [36].

During all flights, emissivity will be set to 1; for image processing, it will be reset to 0.96 for vegetation [2,37] and to 1 for targets. An empirical linear radiometric calibration of the temperature obtained from the UAV was performed using the methodology proposed by Gómez-Candón et al. [38]. During each flight, the temperature of various land targets (white and gray fabrics and a container with water) was measured using a FLUKE 62 MAX IR thermometer (Fluke Corporation, Everett, WA, USA). Additionally, nine square ground control points (PCT: black-and-white, 0.5 × 0.5 m) were installed and georeferenced using a Differential Global Positioning System (DGPS) (GeoXT Handheld Trimble, Houston, TX, USA). This information, along with data from the UAV GPS, was later used to improve the accuracy of the orthomosaics.

Before creating the thermal orthomosaics for all plots (Figure 1), the digital numbers (DN) of each image were converted into temperatures in degrees Celsius using a code developed in FIJI (Fiji is Just ImageJ; https://fiji.sc/). An orthomosaic was then generated for each flight using Agisoft PhotoScan Professional software 2.0 (Agisoft LLC, St. Petersburg, Russia). This software stitches images together with a structure-from-motion (SfM) algorithm by matching common points between images [39]. The workflow in Agisoft PhotoScan Professional followed the same steps as those described by Perich et al. [14]. Finally, the orthomosaics were exported in their original format for further analysis.

Finally, the images were analyzed using an interactive filter to remove all non-plant material and backgrounds, with custom MATLAB R2025a code [40]. Using the MosaicTools plug-in (Fiji is Just ImageJ; https://fiji.sc/), the genotype plot and canopy temperature were identified with a filter based on automatic classification by the Otsu algorithm [41]. This algorithm assumes that the images contain two classes of pixels (soil and vegetation) and automatically finds the optimal threshold between them [42]. Pixels classified as “vegetation” were used to calculate the mean CT per plot and replicate (i.e., vegetation–surface temperature at canopy level). A custom batch processing macro in FIJI performed this task.

2.3. Environmental Data and Thermal Threshold Definition

Hourly meteorological data, including air temperature, relative humidity, and derived vapor pressure deficit (VPD), were collected from weather stations near the experimental fields. These data helped characterize atmospheric demand during each measurement period and supported the interpretation of CT dynamics.

A canopy temperature threshold of 32 °C was chosen as a biologically relevant reference point based on previous studies showing negative effects of high temperature on wheat reproductive processes [32,33]. Instead of treating this threshold as simply binary, the analysis focused on the timing, frequency, and duration of how individual genotypes exceeded this value during diurnal cycles and various growth stages.

2.4. Data Analyses

The CT measurements were treated as repeated observations within the same day, enabling characterization of daily CT plasticity at the genotype level. Diurnal CT distributions were summarized with box-and-whisker plots. To characterize changes in the rate of canopy temperature increase during the day, piecewise linear regressions were fitted to the diurnal CT trajectories for each environment and phenological stage. Slopes were calculated for consecutive time intervals (10:30–12:00, 12:00–13:30, 13:30–15:00, 15:00–16:30, and 16:30–17:30 h). Differences in slope magnitude among intervals were used to identify periods of accelerated canopy warming and maximum genotypic differentiation.

To statistically examine the contextual expression of CT across different environments, developmental stages, and times of day, we used linear mixed models with repeated measures. A comprehensive model was applied to all observations, incorporating fixed effects of Location (Cauquenes, WS; Chillán, FI), Stage (anthesis, milk-grain, milk-dough, and dough), and Time of day (hours, treated as a continuous covariate), along with all two- and three-way interactions. Random factors included genotype (ID), genotype × location (ID: Location), spatial block effects (replicate × row), and plot-level correlation among repeated flights through a random slope of Time at the plot level. Models were fitted using REML with the lme4 and lmerTest packages in R; fixed effects were tested with a Type III ANOVA using Satterthwaite degrees of freedom.

To quantify the magnitude of genetic signal within each Stage × Location combination, we fitted separate mixed models in each stratum, including random effects of genotype (ID) and genotype × hour (ID: Hour; six levels), plus block effects. Variance components were used to calculate the proportion of total variance attributable to genotypic differences across hours. For intra-day rank stability, genotype-by-hour BLUPs were extracted, and Spearman rank correlations were computed between all pairs of measurement hours. Finally, genotype sensitivity to atmospheric demand was assessed through random regression of CT on concurrent vapor pressure deficit (VPD), fitting Location × Stage × VPD fixed effects and genotype- and genotype × location-specific random intercepts and slopes.

To describe genotype-specific exposure to high-temperature conditions, we measured the proportion of the panel exceeding a canopy temperature of 32 °C using both mean and maximum CT values. This approach allowed comparisons among genotypes and phenological stages while accounting for differences between average daily thermal conditions and short-term extreme events. Additionally, cumulative CT exceedance frequencies were calculated separately for the morning, midday, and afternoon periods (10:30–13:30, 13:30–15:00, and 15:00–17:30 h, respectively).

To analyze patterns in the multidimensional space defined by diurnal CT responses, hierarchical clustering (HC) was performed separately for each environment × phenological stage combination. For each case, diurnal CT profiles with six time points were used as feature vectors to represent genotype-specific cooling trajectories. Before clustering, CT profiles were standardized to ensure comparability across time points. Hierarchical clustering was conducted using Euclidean distance and Ward’s minimum variance method (Ward.D2 linkage). The optimal number of clusters (k) was identified by maximizing the mean Silhouette width across k values from 2 to 6 [43]. This approach allowed the detection of coherent groups based on similarities in diurnal canopy temperature dynamics rather than on absolute CT values.

To compare canopy thermal behavior across different environments while considering variations in ambient conditions, canopy (c) − air (a) temperature differences were calculated. In line with standard practice, canopy temperature depression (CTD = Ta − Tc) was defined as the difference, with positive values indicating a cooler canopy than Ta [16,44].

Finally, an alluvial diagram was used to visualize changes in genotype classification between high- and low-maximum CT classes across phenological stages and water regimes. All statistical analyses and graphical outputs were generated using R (v3.3.3). Throughout the analysis, CT was treated as a dynamic response variable, and emphasis was placed on describing how genotype rankings varied across the diurnal cycle rather than on identifying single-time-point metrics.

3. Results

3.1. Diurnal and Seasonal Modulation of Canopy Cooling Across Environments

Across all phenological stages, canopy temperature (CT) showed a clear daily pattern, with lower values in the morning and rising toward the afternoon (Figure 2A–D). However, the degree of daily warming and variability among genotypes varied between environments and across stages. Under rainfed conditions (WS), CT increased more sharply in the afternoon, while under full irrigation (FI), the daily CT patterns were relatively more moderate.

A repeated-measures mixed model confirmed that CT was jointly modulated by environment, phenological stage, and time of day: all main effects and their two- and three-way interactions were highly significant (p < 0.001), supporting the interpretation of canopy temperature as a dynamic phenotype whose diurnal trajectory changes across stages and water regimes.

Although the biggest differences between environments were observed during later phenological stages (i.e., milk–dough and dough), some level of divergence was already evident during anthesis and the milk–grain stage, though within a narrower range. During anthesis, CT under WS exceeded FI only at 15:00 h, while during the milk-grain stage, this divergence was delayed until after 16:30 h. In contrast, during the more advanced stages under WS, differences between environments appeared earlier in the day (from 10:30 h onward) and peaked in the afternoon.

Slope analyses of diurnal CT trajectories revealed that both the timing and the magnitude of maximum CT change vary with phenological stage and environment (Figure 2E–H), closely corresponding to concurrent shifts in atmospheric demand (Figure 2I–L). Under WS conditions, the time of greatest genotypic difference consistently occurred around 15:00 h across stages, while under FI, this timing was more variable, ranging from 13:30 to 15:00 h. The increasing separation among genotypes during later stages suggests that canopy cooling capacity is strongly influenced by the interaction of the time of day and seasonal development.

Variance-component analyses within each Stage × Location combination showed a strong genetic signal in CT. The proportion of total variance attributable to genotypic differences expressed across hours ranged from 87.6% to 97.7% across strata. Under WS (Cauquenes), this proportion was consistently high (≥97.1%) and peaked at the milk-dough stage (genotype × hour variance component, σ²G × Hour = 23.66; proportion = 97.4%), indicating maximal discrimination among genotypes during this period. Under FI (Chillán), the proportion remained high but was comparatively lower (87.6–93.6%), with the smallest value at milk-dough (87.6%) (

Table 1. Variance components and proportion of genotypic variance for canopy temperature (CT) within each phenological stage × location combination (184 genotypes; 6 daily measurement times). Estimates were obtained from stratum-specific mixed models including genotype (ID) and genotype × hour (ID:Hour) random effects. σ²G × Hour: variance of genotype × hour effects; σ²Residual: residual variance; propG: proportion of total variance attributable to the genotypic component pooled across hours. n = 184 spring wheat Triticum aestivum L. genotypes.

Stage	Location	σ2G × Hour	σ2Residual	Prop. σ2G	Interpretation
Anthesis	Cauquenes (WS)	10.42	0.312	97.1%	Highest WS
Anthesis	Chillán (FI)	12.58	0.862	93.6%	Highest FI
Milk	Cauquenes (WS)	15.96	0.402	97.5%	Peak WS
Milk	Chillán (FI)	14.66	1.441	91.0%	High FI
Milk-Dough	Cauquenes (WS)	23.66	0.640	97.4%	Absolute peak
Milk-Dough	Chillán (FI)	6.19	0.876	87.6%	Lowest FI
Dough	Cauquenes (WS)	12.81	0.297	97.7%	High WS
Dough	Chillán (FI)	9.23	1.285	87.8%	Moderate FI

).

Despite the strong genetic signal, genotype rankings were not equally stable across hours. Spearman rank correlations between measurement times revealed higher rank stability under FI (typically ρ ≈ 0.27–0.79, with adjacent hours often >0.70) than under WS, where correlations were frequently low and occasionally near zero or negative (e.g., milk-grain stage: 10:30 vs. 17:30 h, ρ = −0.09). These results indicate substantial within-day re-ranking under water stress, reinforcing that single-time-point CT phenotyping is more prone to time-of-day effects when atmospheric demand increases sharply through the afternoon (Figure S1).

3.2. Thermal Exposure Patterns and Sensitivity to a Critical Canopy Temperature Threshold

Frequency distributions of mean and maximum CT values showed differences among phenological stages and environments (Figure 3). Mean CT values rarely exceeded 32 °C during early reproductive stages but did so at later stages. Conversely, maximum CT values exceeded 32 °C at all stages, especially under rainfed conditions.

Maximum CT distributions showed greater variability than mean CT distributions, indicating that short, intense thermal events are common and are not well captured by average values (Figure 3E–H). As phenological development advances from anthesis to the dough stage, the proportion of genotypes exceeding the 32 °C threshold increases, especially under WS conditions. These differences between mean and maximum CT values emphasize the importance of accounting for extreme thermal exposure when analyzing canopy cooling trends.

Cumulative frequency analyses showed that exceeding the 32 °C threshold mainly depends on the time of day (Figure 4). Morning periods had minimal exceedances across different environments and stages, while midday and afternoon periods experienced the highest number of thermal exposure events. These temporal patterns varied among phenological stages, indicating that genotype exposure to high canopy temperatures changes throughout the daily cycle and across the season.

For example, during the morning of advanced phenological stages under WS, the threshold was not surpassed until around 12:00 (Figure 4C,D), while during midday and in the afternoon, all genotypes under WS exceeded this threshold. These results demonstrate that both the timing and length of thermal exposure depend on the stage and environment.

3.3. Dynamic Grouping and Cross-Environment Consistency of Canopy Temperature

Hierarchical clustering of diurnal CT profiles revealed distinct groups based on cooling capacity within each environment × phenological stage combination (Figure 5 and Figure 6). Silhouette analyses consistently supported the presence of two main clusters across all cases (Figure 5), with group membership determined by differences in diurnal cooling patterns rather than the CT values themselves.

Within each environment, both clusters displayed similar diurnal patterns; however, the separation between groups differed between FI and WS conditions (Figure 6). Overall, FI conditions showed greater variation across cooling capacity groups than WS. As phenological development progressed, the size and composition of these groups changed, with the group characterized by lower CT gradually decreasing in size under WS, while a larger proportion of genotypes kept lower CT classifications under FI.

Analysis of CTD (Ta − Tc) revealed weak or no correlations between environments across various phenological stages and times of day (Figure 7). Coefficients of determination remained low, mostly under 1%, indicating limited consistency in canopy cooling capacity across different environmental conditions. When examining canopy–air temperature differences, genotypes under WS consistently showed negative values across stages, whereas under FI this response varied with time of day.

Genotypic sensitivity to atmospheric demand was further assessed by random regression of CT on concurrent VPD. Variation in CT–VPD slopes was negligible at the genotype level (σ²slope(ID) = 7.26 × 10⁻¹⁰), but substantial at the genotype × environment level (σ²slope(ID × Location) = 1.49 × 10⁻⁵), indicating that VPD sensitivity is strongly environment-dependent. Consistently, genotype-specific CT–VPD slopes showed a weak negative correlation between FI and WS (Spearman’s ρ = −0.166), indicating crossover responses across water regimes (

Table 2. Variance components for genotypic sensitivity of canopy temperature (CT) to vapour pressure deficit (VPD) estimated by random regression. Reported values correspond to the variance of VPD slopes at the genotype level (ID) and at the genotype × location level (ID × Location). ρ indicates the Spearman correlation of genotype-specific VPD slopes between environments (FI vs. WS).

Random Effect	Variance (σ2)—VPD Slope	Std. Dev.	Interpretation
Genotype (ID)	7.26 × 10−10	2.69 × 10−5	Negligible—slope homogeneous across genotypes globally
Genotype × Environment (Gen_Loc)	1.49 × 10−5	3.86 × 10−3	Meaningful—genotypic slope depends on environment

; Figures S1 and S2). n = 184 spring wheat Triticum aestivum L. genotypes.

Alluvial analysis further revealed important reclassification of genotypes between low- and high-canopy-temperature groups at different phenological stages (Figure 8). Only a small set of genotypes consistently maintained a lower CT classification throughout the season, and this set was even smaller under rainfed conditions. In contrast, most genotypes shifted between cooling capacity groups as development progressed, highlighting strong phenological and environmental effects on CT. A limited number of genotypes remained at the extremes of the CT distribution across stages, with more of them under FI than under WS.

4. Discussion

4.1. Canopy Cooling as a Dynamic Phenotype Shaped by Diurnal and Phenological Modulation

This study demonstrates that canopy cooling in wheat is inherently dynamic and cannot be accurately captured with a single measurement. In various settings, canopy temperature consistently followed a daily pattern, with lower readings in the morning and higher ones in the afternoon. However, the extent of this daily variation and the differences between genotypes varied across developmental stages and environments.

This behavior is consistent with the physical and physiological basis of canopy temperature as an emergent property of plant–atmosphere interactions. CT reflects the combined effects of stomatal regulation, canopy structure, radiation load, and atmospheric demand, all of which vary throughout the day [1,2]. Therefore, a measurement taken at a single point in time only captures a temporary thermal state rather than the full expression of a genotype’s canopy cooling capacity [13,14,15].

The results also show that phenological progression enhances genotypic differences in canopy cooling, particularly during later reproductive stages when atmospheric demand rises [29,30]. Similar effects of phenology on canopy temperature have been observed in wheat under field conditions, where differences among genotypes become more evident during stages with higher heat and atmospheric stresses [16,17]. These findings collectively support the idea that canopy cooling should be seen as a temporally organized trait, influenced by both daily and seasonal changes.

4.2. Temporal Heterogeneity of Thermal Exposure and Limited Cross-Environment Consistency

Temperatures at or above 32 °C are widely recognized as physiologically detrimental for wheat, especially during reproductive development when heat exposure can cause lasting tissue damage [32,33]. In this study, CT exceeding this threshold was not consistently observed across genotypes, phenological stages, or times of day. Instead, the timing and duration of heat exposure varied greatly, with extreme thermal events mostly occurring during midday and afternoon periods.

Notably, maximum CT captured short-duration thermal peaks that average values did not reflect, highlighting the limitations of averaged metrics for describing thermal exposure. These results demonstrate that sensitivity to high CT cannot be assumed solely because a threshold is exceeded; instead, it must also consider when during the day the exposure occurs and for how long [7,13]. This temporal variability complicates the interpretation of the canopy cooling trend along the day.

Consistent with this view, the links between canopy cooling capacity across settings were weak or absent at different phenological stages and times of day. The limited consistency across environments highlights the strong influence of atmospheric demand and water availability on canopy temperature, as previously demonstrated in different field conditions [12,45]. This interpretation is consistent with the lower within-day rank stability under WS (Figure S1) and the environment-dependent CT–VPD sensitivities (Table 2; Figure S2). These findings indicate that canopy cooling trends observed in one environment should not be assumed to occur in another, especially when they differ considerably; as in this study, water availability and VPD vary along the day.

This interpretation is consistent with mixed-model results showing pronounced within-day re-ranking under WS and largely environment-dependent CT–VPD sensitivities, reinforcing that thermal phenotypes inferred from a single hour or a single environment may not extrapolate across contrasting water regimes [16,17].

4.3. Dynamic Reclassification of Genotypes and Implications for Phenotypic Interpretation

The alluvial analysis offers a detailed view of how CT changes throughout the reproductive season. Only a few genotypes consistently kept a lower canopy temperature classification across all phenological stages, and this group was even smaller under rainfed conditions. In contrast, most genotypes shifted between different cooling capacity groups as development progressed, highlighting the relevant impact of phenological and environmental factors on canopy cooling capacity.

This dynamic reclassification underscores a key limitation of phenotypic assessments based on a single measurement. The apparent identification of “cooler” genotypes in favorable or early-season conditions does not necessarily mean they will retain stable CT later in the season or in more challenging environments. The fact that fewer genotypes exhibit consistent cooling under unfavorable conditions further highlights the context sensitivity of thermal phenotypes.

5. Conclusions

This study demonstrates that CT in wheat behaves as a dynamic trait, shaped by the interactions among time of day, phenological stage, and environment. UAV-based thermal imaging performed repeatedly from morning to late afternoon across four reproductive stages revealed a strong genotypic signal within each Stage × Environment combination, yet genotype rankings were not stable across hours, particularly under rainfed conditions.

These results show that relying on a single daily CT observation can misrepresent genotype-specific cooling capacity because within-day re-ranking increases as atmospheric demand rises. Under water stress, maximum genotypic differentiation generally occurred around mid-afternoon (~15:00 h), whereas under full irrigation, the most informative measurement window varied with phenological stage. In addition, maximum CT and threshold-based metrics captured short, physiologically relevant heat events (e.g., CT > 32 °C) that were not evident from daily means, highlighting the value of time-resolved thermal phenotyping for interpreting heat exposure during sensitive reproductive periods.

The calibration workflow and multi-temporal analysis framework proposed here can support multiple research communities. Wheat breeders can use diurnal CT trajectories to select genotypes that maintain cooler canopies more consistently across stages and to define phenotyping windows that maximize genetic discrimination under target drought/heat scenarios. Future work should extend this methodology across years and a wider range of environments, integrate CT dynamics with yield and water-use data, and couple thermal phenotypes with genomic and multi-trait selection frameworks to accelerate breeding for heat- and drought-resilient wheat.