The field experiment was conducted during the rabi season of 2021–22 at the Main Agricultural Research Station, University of Agricultural Sciences, Dharwad (15°29′N, 74°59′E; 678 m above mean sea level), located in the Northern Transitional Zone (Zone-8) of Karnataka, India. The soil at the experimental site was medium-deep black clay, slightly alkaline (pH 7.8), with an electrical conductivity of 0.32 dS m⁻¹, low organic carbon (0.46%), low available nitrogen (278 kg N ha⁻¹), medium available phosphorus (35 kg P₂O₅ ha⁻¹), and high potassium (393 kg K₂O ha⁻¹). Textural analysis by the international pipette method (Piper, 1966) indicated 8.30% coarse sand, 14.58% fine sand, 20.80% silt, and 56.32% clay, classifying the soil as clay. Bulk density was 1.32 g cm⁻³ (Black, 1967). The soil had a field capacity of 30.52% at 0–15 cm and 31.68% at 15–30 cm depth (Dastane, 1972), and a permanent wilting point of 18.02% and 18.78% at the respective depths (Veihmeyer and Hendrickson, 1949).

During the crop-growing period, a total of 57.2 mm of rainfall was recorded across three events, all of which occurred after the dough stage (R4) and thus did not interfere with treatment imposition (Figure 1). All irrigation treatments were implemented as planned without deviations. Although the experiment was conducted in a single season (2021–22 rabi), the methodology remains robust and reproducible. The results are further interpreted in the context of recent UAV-based multispectral sensing and precision irrigation studies (Kapari et al., 2024; Kapari et al., 2025; Liu et al., 2023), ensuring their continued relevance to contemporary agricultural practices.

The experiment was arranged in a randomized complete block design (RCBD) with nine irrigation treatments and three replications. The maize hybrid NK-6240 was selected as it is widely cultivated in the region and recommended for rabi maize production. It is also known for its high yield potential and adaptability to local agro-climatic conditions. Sowing was carried out at a spacing of 60 × 20 cm, corresponding to a plant population of approximately 83,333 plants ha⁻¹, as per regional recommendations to ensure optimum growth and yield performance under irrigated conditions. Fertilizers were applied at 150:65:65 N:P₂O₅: K₂O kg ha⁻¹, with 50% of nitrogen and full dose of phosphorus and potassium applied as basal, and remaining 50% of nitrogen top-dressed at 35 days after sowing (DAS). UAV-based aerial imagery was also acquired to document the experimental layout and canopy development across irrigation treatments, as illustrated in Figure 2.

Irrigation was scheduled according to the treatment plan: (T₁) at all growth stages (VE, V4, V8, VT, R1, R3, R4), (T₂) skipping irrigation at tasseling (VT), (T₃) skipping irrigation at silking (R1), (T₄) skipping irrigation at milky stage (R3), (T₅) skipping irrigation at VT and R1, (T₆) skipping irrigation at VT and R3, (T₇) skipping irrigation at R1 and R3, (T₈) skipping irrigation at all critical stages (VT, R1, R3), and (T₉) irrigation only upto knee-high stage (V8). Irrigation was provided to each plot using a 2-inch siphon pipe at a discharge rate of 2.5 L s⁻¹. To prevent lateral water movement and ensure treatment integrity, earthen bunds (30 cm height) were constructed around each plot, and buffer channels (0.5 m wide) were maintained between adjacent plots. Stage-wise water applied and total water for each treatment is presented in Table 1.

Multispectral images of maize were acquired at the dough stage (R4) using a DJI Phantom 4 quadcopter UAV (SZ DJI Technology Co., Ltd., Shenzhen, China). The UAV was equipped with a GPS-assisted navigation system, four brushless motors, and an intelligent LiPo battery, enabling a flight duration of approximately 25–28 min. For imaging, it was integrated with a MicaSense RedEdge multispectral camera (Mica Sense Inc., Seattle, WA, United States). The sensor captures five spectral bands—blue (475 nm), green (560 nm), red (668 nm), red-edge (717 nm), and near-infrared (840 nm)—with a bandwidth of 10–20 nm. It also features a global shutter, 12-bit radiometric resolution, and a cosine-corrected irradiance sensor to account for varying light conditions, ensuring consistent reflectance measurement.

Prior to the UAV flight, preparations included battery charging and inspection, mission planning for flight path and altitude, GPS lock verification, and safety checks for wind and weather conditions. Radiometric calibration was performed using the manufacturer’s reference panel, and the sensor was mounted in a nadir orientation to minimize angular effects. Detasseling was carried out to avoid tassel interference with canopy reflectance. The UAV flight was executed at an altitude of 45 m, with images captured at one-second intervals; the entire field was covered in approximately 4–5 min, and flights were completed before 11:00 a.m. to minimize variation in solar illumination. To ensure dataset comparability, GreenSeeker measurements (NDVI) were collected immediately after the UAV flight, with a time gap of about 10 min.

The raw multispectral images were then processed using Pix4Dmapper software (version 4.8.0; Pix4D, 2023). The workflow included geotagging of individual images using flight metadata, followed by stitching and mosaicking to generate high-resolution orthomosaic maps. Radiometric correction was performed in Pix4Dmapper by applying sensor-specific adjustments (vignetting, irradiance, and exposure) and by using reflectance calibration panel images acquired before and after each flight. This ensured that the resulting orthomosaics represented consistent surface reflectance suitable for vegetation index calculation and subsequent analysis.

NDVI measurements were recorded with a GreenSeeker during the tasseling, silking, milky, and dough (R4) stages. For each plot, ten readings were collected at evenly spaced positions across the central rows, and the mean value was used for analysis. Measurements were consistently taken between 09:30 and 11:30 a.m. on clear days to minimize variability and ensure repeatability. Although NDVI was recorded at multiple reproductive stages, only the R4 stage values are presented in the results to enable direct comparability with UAV-derived NDVI collected at the same stage.

UAV imagery was processed in Pix4DMapper to generate orthomosaic maps for each spectral band. Vegetation indices—including NDVI, RDVI, SAVI, OSAVI, and TCARI—were subsequently calculated from reflectance values using ArcGIS, following established remote sensing protocols (Zheng et al., 2021).

Kernel yield was determined by harvesting all cobs from the net plot area at full maturity. The harvested cobs were dried to a constant weight and shelled using a maize sheller. The kernels were then cleaned and weighed. The kernel weight was recorded on a per-plot basis and extrapolated to hectare basis, with the final yield expressed in kilograms per hectare (kg ha⁻¹).

Data from the field experiment was analysed using analysis of variance (ANOVA) under a randomized complete block design (RCBD) to evaluate treatment effects at a significance level of p = 0.05. Treatment means were compared using Duncan’s Multiple Range Test (DMRT). Variability across replications was expressed as standard deviation (SD), reported with the means in the tables. In addition, pooled standard error of mean (SEm) values was calculated and are presented in the bottom rows of the tables for reference. Correlation analysis was performed using all plot-level data (27 observations; nine treatments × three replications) to capture the full variability. Correlation heat maps were used to depict the strength and direction of associations among vegetation indices, CWSI, and kernel yield.

To complement the inferential analysis, supplementary visualizations were generated to aid interpretation. Boxplots were used to illustrate the distribution, variability, and treatment effects for kernel yield and vegetation indices, highlighting patterns and potential outliers. For clarity, significance groupings from DMRT were presented only in tables, while boxplots were reserved for visualizing distributions without additional lettering.

All analyses were performed in R software (version 4.2.2). The ggplot2 package (v3.4.0) was used to generate boxplots and other visualizations, while the corrplot package (v0.92) was employed for correlation matrices. This workflow ensured both statistical rigor and clear visualization of treatment effects and inter-variable relationships.

Crop growth and yield are strongly influenced by prevailing weather conditions. During the crop growth period, a total of 57.2 mm of rainfall was received across three events. Importantly, all rainfall occurred after the dough (R4) stage, so it did not interfere with the planned irrigation treatments (Figure 1). At the early vegetative stages, daytime temperatures averaged 29 °C, while night temperatures dropped to ∼13 °C. Elevated relative humidity during this period created favorable conditions for infestation by fall armyworm (Spodoptera frugiperda). The outbreak was effectively controlled through timely insecticide applications, preventing further spread. These weather conditions, particularly the absence of rainfall until R4, provided a controlled environment for studying water stress. Similar controlled dry periods have been shown to enhance the precision of irrigation trials and stress detection (Zhu and Burney, 2022).

NDVI values differed significantly among irrigation treatments (Figures 3 and 4). Although GreenSeeker NDVI was measured at multiple reproductive stages, only the R4 stage values are presented here to ensure direct comparability with UAV-derived NDVI acquired at the same stage. The highest MSS-based NDVI (0.76) and GreenSeeker-based NDVI (0.78) were recorded under full irrigation (T₁). These values were statistically comparable (p > 0.05) to those obtained when irrigation was skipped at tasseling (T₂: MSS 0.71; GreenSeeker 0.72), silking (T₃: MSS 0.70; GreenSeeker 0.71), or milky stage (T₄: MSS 0.72; GreenSeeker 0.73). This indicates that skipping irrigation at a single critical stage did not significantly reduce canopy vigor, suggesting compensatory responses under mild stress, as also reported by Xu et al. (2022).

In contrast, skipping irrigation at two consecutive stages resulted in sharp NDVI reductions (p ≤ 0.05). Irrigation skipped at tasseling and silking (T₅: MSS 0.45; GreenSeeker 0.48), tasseling and milky (T₆: MSS 0.51; GreenSeeker 0.53), and silking and milky stages (T₇: MSS 0.49; GreenSeeker 0.51) produced values 30%–40% lower than T₁. These reductions are also evident in the NDVI values (Figures 3, 4), where water-stressed treatments show distinct declines compared with fully irrigated plots. Such declines reflect impaired canopy development, chlorophyll loss, and reduced photosynthetic efficiency, consistent with earlier findings that NDVI decreases under drought due to senescence and diminished leaf area (Zhang et al., 2019; Liu et al., 2022).

The lowest NDVI values were recorded when irrigation was applied only up to knee-high (T₈: MSS 0.38; GreenSeeker 0.40) or withheld at all critical stages (T₉: MSS 0.37; GreenSeeker 0.39), corresponding to a ∼50% reduction relative to T₁ (p ≤ 0.05). These severe declines highlight the sensitivity of NDVI to prolonged water deficit, as limited vegetative growth and canopy senescence reduce near-infrared reflectance and increase red reflectance.

Although NDVI remained statistically comparable when irrigation was skipped at only one stage (T₂–T₄), yield reductions still occurred, reflecting the varying sensitivity of reproductive stages. Water stress at silking restricts pollination and fertilization (Li et al., 2024), tasseling stress lowers pollen viability and silk receptivity (Klocke et al., 2019), and milky-stage stress reduces kernel filling (Irmak et al., 2019). Thus, even when NDVI values were unaffected statistically, reproductive processes were disrupted.

Overall, NDVI proved to be a sensitive indicator of water status and canopy health in maize. The boxplots clearly demonstrate that severe or prolonged water deficits substantially reduce NDVI, emphasizing the importance of continuous irrigation during both vegetative and reproductive phases to sustain crop productivity.

RDVI values differed significantly among irrigation treatments (Figure 5). The highest value was observed under full irrigation (T₁: 0.75), followed by treatments where irrigation was skipped at tasseling (T₂: 0.67), silking (T₃: 0.64), or milky stage (T₄: 0.70). These values were statistically comparable to each other (p > 0.05), indicating that skipping irrigation at only one critical stage did not significantly reduce canopy vigor. This suggests that maize can tolerate brief water stress without major loss of canopy greenness, consistent with the findings of Song et al. (2019).

In contrast, RDVI declined significantly when irrigation was skipped at two consecutive stages (p ≤ 0.05). Treatments T₅ (tasseling + silking: 0.38), T₆ (tasseling + milky: 0.49), and T₇ (silking + milky: 0.50) recorded 30%–45% lower RDVI compared with T₁. These patterns are also evident in the RDVI values (Figure 5), where stressed treatments show clear reductions in canopy vigor compared with fully irrigated plots. Such declines reflect restricted canopy expansion, reduced leaf area, and limited light interception, which lower near-infrared reflectance relative to red reflectance, resulting in decreased RDVI values (Gao, 1996; Delegido et al., 2023).

The lowest RDVI values were observed when irrigation was applied only up to the knee-high stage (T₈: 0.29) or withheld throughout critical stages (T₉: 0.26), corresponding to a ∼65% decline relative to T₁ (p ≤ 0.05). These treatments highlight the severe impact of early termination or complete withdrawal of irrigation, leading to canopy senescence, reduced photosynthetic activity, and impaired crop growth.

Overall, these findings demonstrate that RDVI is a robust indicator of water stress in maize, particularly under prolonged or severe deficits. Its strong sensitivity to vegetation vigor and chlorophyll status aligns with earlier reports (Zhang et al., 2019), confirming its utility for monitoring crop water status under varying irrigation regimes.

SAVI values differed significantly among irrigation treatments (Figure 6). The highest values were recorded under full irrigation (T₁: 0.84), which were statistically comparable (p > 0.05) to treatments where irrigation was skipped at tasseling (T₂: 0.80), silking (T₃: 0.81), or milky stage (T₄: 0.83). This indicates that skipping irrigation at only one critical stage did not significantly affect canopy vigor, suggesting that maize can compensate for short-term water stress by maintaining leaf area and greenness. Similar compensatory responses were reported by Zhang et al. (2019) and Teodoro et al. (2024).

In contrast, skipping irrigation at two consecutive stages significantly reduced SAVI (p ≤ 0.05). Treatments T₅ (tasseling + silking: 0.52), T₆ (tasseling + milky: 0.56), and T₇ (silking + milky: 0.58) showed 30%–40% lower SAVI compared with T₁. These reductions reflect decreased chlorophyll content, limited photosynthetic activity, and impaired canopy development, with the most pronounced declines under T₅, highlighting the sensitivity of tasseling and silking to water deficits. These patterns are also evident in the SAVI values (Figure 6), where treatments subjected to consecutive irrigation skips show visibly reduced canopy vigor compared with fully irrigated treatments. Such responses are consistent with earlier reports linking reproductive-stage stress to reduced kernel set and yield (Irmak et al., 2019).

The lowest SAVI values were observed when irrigation was applied only up to the knee-high stage (T₈: 0.37) or withheld at all critical stages (T₉: 0.36), representing a ∼55% decline relative to T₁ (p ≤ 0.05). These results illustrate the severe impact of prolonged water stress on canopy cover, as restricted vegetative growth and reduced leaf expansion limit light interception and NIR reflectance while enhancing red reflectance. This explains the sharp decline in SAVI under extreme stress. Overall, these findings underscore the sensitivity of SAVI to water availability and its reliability in distinguishing between moderate and severe stress conditions in maize.

OSAVI values varied significantly among irrigation treatments (Figure 7). The highest value was recorded under full irrigation (T₁: 0.72), which was statistically comparable (p > 0.05) to treatments where irrigation was skipped at tasseling (T₂: 0.67), silking (T₃: 0.68), or milky stage (T₄: 0.71). This indicates that skipping irrigation at only one critical stage did not significantly affect canopy vigor, suggesting a compensatory response by maize under mild water stress. Similar tolerance has been reported by Zhang et al. (2019).

In contrast, withholding irrigation at two consecutive stages significantly reduced OSAVI. Treatments T₅ (tasseling + silking: 0.53), T₆ (tasseling + milky: 0.56), and T₇ (silking + milky: 0.55) exhibited 25%–30% lower OSAVI compared with T₁ (p ≤ 0.05). These reductions are also evident in the OSAVI values (Figure 7), where stressed treatments show visibly reduced canopy vigor compared with fully irrigated plots. Such declines reflect impaired canopy structure, reduced chlorophyll content, and limited photosynthetic activity. Since OSAVI minimizes soil background effects, its reductions here primarily capture vegetation stress rather than soil noise, in line with findings of Colović et al. (2022).

The lowest OSAVI values were observed in treatments irrigated only up to the knee-high stage (T₈: 0.29) or where irrigation was withheld at all critical stages (T₉: 0.28), corresponding to nearly 60% reduction compared to T₁ (p ≤ 0.05). Such sharp declines highlight the severity of early and prolonged water stress, which restricts vegetative growth, accelerates canopy senescence, and reduces leaf water content. UAV-based studies further confirm strong correlations between OSAVI and crop water status, demonstrating its reliability for precision irrigation scheduling and high-throughput stress detection in maize (Mndela et al., 2023).

TCARI values differed significantly among irrigation treatments (Figure 8). The lowest values were recorded under full irrigation (T₁: 0.11), indicating high chlorophyll preservation and healthy vegetation. Similar values were observed when irrigation was skipped only at the milky stage (T₄: 0.13), and these treatments were statistically comparable to T₁ (p > 0.05). In contrast, irrigation applied only up to the knee-high stage (T₈: 0.32) and complete irrigation withdrawal across all stages (T₉: 0.34) resulted in the highest TCARI values, representing ∼65–70% higher readings relative to T₁ (p ≤ 0.05), indicating severe chlorophyll degradation.

Dual-stage irrigation skips also produced significantly higher TCARI values compared with T₁. Treatments T₅ (tasseling + silking: 0.29), T₆ (tasseling + milky: 0.26), and T₇ (silking + milky: 0.27) recorded more than double the values of fully irrigated treatments, reflecting accelerated chlorophyll breakdown under combined reproductive-stage stress (p ≤ 0.05). These patterns are also evident in the TCARI values (Figure 8), where stressed treatments show elevated values corresponding to severe pigment loss.

These variations in TCARI reflect the crop’s physiological adaptation to water availability, particularly in maintaining chlorophyll integrity and canopy health. Since TCARI is inversely related to chlorophyll content (Haboudane et al., 2004; Zarco-Tejada et al., 2003), the higher values observed under T₈ and T₉ confirm the loss of chlorophyll pigments and accelerated senescence. Early vegetative irrigation, while beneficial for root establishment, proved insufficient to sustain chlorophyll stability through flowering and grain filling, which are highly sensitive to water deficits (Anjum et al., 2011; Tardieu, 2012).

By contrast, the lower TCARI values in T₁ and T₄ indicate effective preservation of chlorophyll and photosynthetic activity under well-watered or moderately stressed conditions, supporting delayed senescence and prolonged biomass accumulation (Ali et al., 2019; Fang et al., 2020). Treatments involving dual irrigation skips (T₅–T₇) recorded the highest stress responses, particularly when tasseling and silking coincided with water deficit. These stages represent peak metabolic and reproductive activity. Under stress, nutrient mobility, chloroplast stability, and hormonal balance are reduced, leading to rapid canopy senescence and increased reflectance in the red band (Zhang et al., 2019; Avni et al., 2020).

Overall, the observed trends confirm that strategic irrigation scheduling especially ensuring water availability during tasseling and silking is critical for maintaining chlorophyll content and canopy health, as reflected in lower TCARI values (González-Dugo et al., 2014; Liu et al., 2023; Zhou et al., 2021).

The spatial variability of these vegetation indices is further illustrated in Figure 9, where panels 9a–9e show the spatial patterns of NDVI, RDVI, SAVI, OSAVI, and TCARI, respectively. These maps provide a clear visual comparison of water-stressed and irrigated treatments and complement the boxplot analyses.

Kernel yield varied significantly across irrigation regimes (Figure 11). The highest yield was recorded under full irrigation (T₁: 7,040.7 kg ha⁻¹), which was significantly higher (p ≤ 0.05) than all other treatments. Skipping irrigation at tasseling (T₂: 5,744.7 kg ha⁻¹) or silking (T₃: 5,874.7 kg ha⁻¹) reduced yield by 18%–20% compared with T₁, while skipping irrigation at the milky stage (T₄: 6,411.0 kg ha⁻¹) caused a smaller decline (9%).

More severe yield reductions were observed under dual-stage irrigation skips. Yield under tasseling + silking (T₅: 3,681.3 kg ha⁻¹) was 47.7% lower than T₁, while tasseling + milky (T₆: 4,271.3 kg ha⁻¹) and silking + milky (T₇: 4,016.0 kg ha⁻¹) reduced yields by 39.3% and 42.9%, respectively. The lowest yields were recorded when irrigation was applied only up to the knee-high stage (T₈: 2,729.7 kg ha⁻¹) or withheld during all reproductive stages (T₉: 2,628.3 kg ha⁻¹), corresponding to 61%–63% yield losses compared with T₁.

These results highlight the importance of irrigation during reproductive phases. Water deficit at tasseling and silking (T₅–T₇) drastically reduced yields, consistent with earlier findings that stress during these stages’ limits pollen viability, silk receptivity, and kernel initiation (Daryanto et al., 2016; Song et al., 2019).

The extremely low yields in T₈ and T₉ confirm that early irrigation alone cannot sustain water requirements during reproductive stages, leading to severe assimilate shortages for grain filling. Similar conclusions were drawn by Fereres and Soriano (2007), Pandey et al. (2000), and Farooq et al. (2017), who reported that early stress restricts canopy development, biomass accumulation, and yield potential.

Interestingly, yields from T₈ and T₉ were statistically comparable to those in multi-stage skips, indicating that both prolonged early stress and severe late stress can equally limit kernel production. Stress during kernel filling reduces kernel number and weight, as also reported by Zhang et al. (2019). Although maize tolerates mild vegetative stress, reproductive-stage water deficits directly reduce photosynthetic activity and assimilate partitioning to kernels, resulting in major productivity losses.

Overall, kernel yield trends confirm that irrigation at tasseling, silking, and milky stages is indispensable for securing high maize yield, while skipped irrigation during these critical phases results in irreversible losses.

Mean CWSI, derived from stage-specific non-water-stressed baselines (LL) with UL values remaining nearly constant across stages, showed strong correlations with both vegetation indices and grain yield (Figure 12). CWSI was negatively correlated with NDVI (r = −0.98), RDVI (r = −0.99), SAVI (r = −0.99), and OSAVI (r = −0.88), while TCARI exhibited a strong positive correlation (r = 0.96). As canopy stress intensified, vegetation indices declined markedly, reflecting reduced chlorophyll content, weaker canopy development, and diminished photosynthetic activity.

Grain yield was also closely associated with these stress indicators. Yield declined sharply with increasing CWSI (r = −0.99, p < 0.001), whereas it showed strong positive correlations with NDVI (r = 0.99), RDVI (r = 0.98), SAVI (r = 0.99), and OSAVI (r = 0.90). In contrast, TCARI correlated negatively with yield (r = −0.97), consistent with its sensitivity to stress-induced pigment degradation and reduced photosynthetic capacity.

Overall, the integration of thermal (CWSI) and spectral indicators (NDVI, RDVI, SAVI, OSAVI, TCARI) provided a comprehensive picture of crop stress and yield response. While spectral indices captured canopy chlorophyll status and biomass accumulation, CWSI reflected stomatal regulation and transpiration cooling. Recent studies (Hatfield and Prueger, 2010; Zhang et al., 2019; Yetik and Candogan, 2023; Liu et al., 2023) similarly highlight that combining spectral and thermal data enhances the accuracy of stress detection and yield prediction. Moreover, indices incorporating soil-adjustment parameters, such as SAVI (L = 0.5) and OSAVI (C = 0.16), consistently performed better, as these formulations reduce soil background effects, in line with recent advances emphasizing parameter calibration for robustness across environments (Song et al., 2024; Liang et al., 2024).

This study introduces key advances in maize water stress detection. First, UAV multispectral indices at the R4 stage were integrated with stage-specific ground-based CWSI measurements (VT, R1, R3, and R4). This directly linked canopy spectral responses with physiological stress and yield, offering more reliable detection than using spectral or thermal indicators alone.

UAV imaging was timed at R4, when irrigation treatments ended and stress effects were fully expressed. This captured cumulative stress impacts while reducing the need for multi-date campaigns. CWSI was calibrated locally using well-irrigated plots to derive stage-specific LL baselines. Although both LL and UL were estimated, UL varied minimally (<0.1 °C), consistent with Zhang et al. (2019). Thus, LL provided the main source of variability, improving the robustness of stress quantification compared with theoretical models.

The experimental design, with nine irrigation treatments and systematic stage omissions, created a rigorous stress gradient. This enabled a thorough evaluation of index sensitivity and yield response. The combined use of UAV multispectral indices and calibrated thermal indices demonstrates strong potential for precision irrigation, improving water use efficiency and stabilizing yields under limited water supply.

Present study was conducted under tropical wet–dry conditions at the Main Agricultural Research Station, University of Agricultural Sciences, Dharwad, with site-specific calibration of CWSI baselines and UAV imaging at R4. While the results provide robust local insights, broader applicability requires validation in diverse agro-ecological settings.

Recent studies suggest strategies for scaling UAV- and index-based stress detection. Cheng et al. (2023) emphasized standardized calibration protocols for cross-regional comparability, while Zhang et al. (2019) highlighted domain-adaptation approaches to reduce site-specific bias. Together, these methods can enhance the robustness and transferability of stress detection frameworks.

Integrated approach combining UAV multispectral indices with stage-specific CWSI—shows strong potential for adaptation in other environments, provided local calibration is performed. Since UL was nearly constant under local conditions, future studies should test whether it is applicable across different climates. If confirmed, LL-based calibration offers a simplified but reliable framework for cross-regional applications.