Remote sensing–derived vegetation indices for assessing water-use efficiency and associated physiological traits in grass species adaption to drought stress: a review

The increasing frequency and severity of droughts, combined with rising water demand and declining water resources, exacerbate water scarcity and threaten plant growth and productivity. Consequently, developing strategies to enhance plant tolerance to drought stress by improving water use efficiency (WUE) is essential. WUE refers to the amount of carbon assimilated during photosynthesis or used to produce biomass or grains per unit of water consumed by the crop^[1,2]. Several plant traits influence WUE, including biomass, leaf area or plant density, photosynthetic rate (A), evapotranspiration (ET), transpiration (T), and stomatal conductance (g_s)^[3−5]. Evaluation and quantification of these WUE-related components are necessary to assess WUE and to develop water-use-efficient plant varieties.

Most current methodologies for assessing WUE are either destructive or labor-intensive, including measurements such as the A/T and the biomass/ET, and estimations based on carbon isotope discrimination (CID; Δ¹³C) analysis. Alternatively, vegetation indices derived from remote sensing offer significant potential for high-throughput and non-destructive evaluation of WUE. Spectral technologies evaluate plant vegetation by analyzing how much light is reflected and absorbed by plants in different wavelengths, which are closely linked to plant growth, physiological traits, and environmental stresses^[6−14]. For instance, visible light (450−750 nm) reflectance relates to pigments such as chlorophyll, carotene, and anthocyanin in leaves^[15]. Pigment-related indices include Photochemical Reflectance Index (PRI), Chlorophyll red-edge Index (Clred-edge), Structure-Insensitive Pigment Index (SIPI), and Modified Chlorophyll Absorption in Reflectance Index (MCARI). Visible and NIR have been successfully used in the estimation of chlorophyll and water content in leaves exposed to drought stress^[16]. Near-infrared red (NIR, 750−1,000 nm) reflectance relates to cell structure such as protein, starch, fatty acid, the thick leaf cuticle, mesophyll exposure ratio in plants^[15,17], and is captured by structural indices such as Normalized Difference Vegetation Index (NDVI), Simple Ratio (SR), Enhanced Vegetation Index (EVI), and Soil-Adjusted Vegetation Index (SAVI). Short-wave infrared (SWIR, 1,000−2,500 nm) contains absorption peaks for water, sugar, and other plant chemicals, in which water has a high absorption rate at 1,450, 1,900, and 2,500 nm^[15,18]. SWIR has advanced in non-destructively differentiating genotypic variations and phenotyping for relative water content (RWC), A, g_s, and T in plants exposed to drought^[19]. Water-related indices include Water Band Index (WBI), Shortwave Infrared Water Index (SWWI), Normalized Difference Moisture Index (NDMI), and Normalized Difference Water Index (NDWI). When spectral technologies are integrated with Unmanned Aerial Vehicle (UAV), they make remote sensing a powerful tool for data acquisition in complex physiological traits such as WUE-related traits.

To capture spectral reflectance of plants, the most widely used cameras or sensors mounted on UAVs include Red-green-blue (RGB), multispectral, and hyperspectral imaging systems. These sensors are primarily different in spectral resolution, the level of interpretable physiological traits from the data, and cost, making each suited to different application scenarios. The sensor type requirements for VIs associated with WUE-related traits are listed in Table 1. RGB camera is the most basic form, capturing three color bands (red, green, and blue). With its low-cost and widespread accessibility, RGB imagery is widely used for a quick and cheap check of plants, especially for canopy color, structure, visible biotic and abiotic stress symptoms, but lacks invisible spectral information. Hyperspectral imaging has the highest spectral resolution, capturing hundreds of narrow contiguous bands to provide highly specific information about plant growth, physiological status, and stress responses. Hyperspectral imaging offers high spectral resolution to assist physiological traits estimation but generates large datasets, requires complex processing, and is high-cost. Multispectral cameras use several specific bands (four to ten bands) like visible and near-infrared, enabling the calculation of more vegetation indices. Multispectral systems are more practical for large-scale breeding programs, providing high-throughput phenotyping with a balance between operation efficiency, data content, and affordable cost.

This review overviews and analyzes methodologies for assessing WUE and VIs related to WUE, and further discusses potential indices associated with carbon-isotope-discrimination-based WUE in grass species, including annual cereal crops and perennial grasses cultivated as turf or forage.

WUE at the plant community level is typically calculated as the ratio of total plant biomass to total community ET, where a higher WUE indicates more biomass produced per unit of water consumed^[62]. The total amount of plant biomass that has been produced by the community is often the dry weight of the plants. The ET rate is the total amount of water lost from the system to the atmosphere. It is the sum of the transpiration rate from the leaves of plants and the evaporation rate from the soil. The ET rate can be measured using lysimeters in field plots or pot-grown plants in controlled environments, such as growth-chamber and greenhouse experiments. The biomass/ET based WUE has been used widely in various crops, including perennial grasses and annual crops, such as maize (Zea mays L.), winter wheat (Triticum aestivum L.), and winter rye (Secale cereale L.)^[5,63].

At the leaf level, intrinsic or instantaneous WUE is calculated from the ratio of A and T or g_s, expressed as μmol CO₂ per mmol H₂O. The leaf-level WUE can be quantified directly using gas exchange systems which measure the rates of A and T or g_s, allowing for real-time calculation of A/T or A/g_s in individual leaves of field plants or pot-grown plants^[64,65].

Leaf WUE can also be estimated by measuring Δ¹³C. Both atmospheric ¹²CO₂ and ¹³CO₂ enter the stomata together, but Rubisco has a higher affinity to the lighter and smaller ¹²CO₂ due to their different kinetics and diffusion rates, resulting in a change in the composition of ¹³C and ¹²C (δ¹³C) in leaf tissues compared to the constant δ¹³C in the air^[66]. However, when stomatal closure occurs under drought, limited availability of CO₂ makes Rubisco take up more unfavorable ¹³CO₂ when ¹²CO₂ runs out in the stomata, resulting in a higher δ¹³C in the plant. Therefore, when the stomata close more, there is a higher Δ¹³C, which refers to a greater discrimination of δ¹³C between the plant and the air. These plants were considered more water-use efficient as their stomata close more under abiotic stress. The Δ¹³C is negatively correlated with instantaneous WUE^[67]. WUE can be effectively estimated with Δ¹³C, and genetic variations in Δ¹³C were found among grass species^[68]. The Δ¹³C is highly heritable and has high broad-sense heritability (over 0.91) in various plant species, such as wheat^[69]. Significant QTLs associated with Δ¹³C were identified on chromosomes 5 and 6 in creeping bentgrass (Agrostis stolonifera) under drought^[70]. The Δ¹³C value provides a long-term, integrated measure of the ratio of internal leaf CO₂ concentration to ambient CO₂ concentration by reflecting the plant's physiological responses over the period of tissue formation, offering a practical alternative to instantaneous WUE measurements by laborious gas exchange systems.

Destructive methods for measuring WUE, such as plant community-based biomass assessments, present challenges, especially for crops like turfgrass, which are not designed for clipping yield harvesting. Additionally, the accuracy of these measurements is often limited by non-uniform canopy growth^[71,72]. Instantaneous WUE estimation provides more direct and precise physiological data; however, it is time-consuming and labor-intensive^[7]. The Δ¹³C serves as a more convenient surrogate for estimating integrated WUE^[70,73]. However, this method involves destructive sampling, and the analysis of δ¹³C can be costly. Therefore, a non-destructive approach that can assess WUE and monitor its dynamic changes is highly desirable.

Remote sensing imagery, which captures the spectral reflectance of plant canopies, offers a powerful approach for evaluating plant growth, physiological traits, and responses to environmental stresses^[6−13], suggesting a significant potential for high-throughput evaluation of WUE. Various VIs derived from remote sensing and spectral imaging are used to evaluate plant or canopy phenotypic traits or physiological status (Table 1). Under drought stress, plants exhibit varying reductions in leaf water status, chlorophyll content, A, ET, and leaf area index (LAI), all of which can be detected by UAV remote sensing^[74,75]. This approach can identify drought-related physiological traits long before visible symptoms appear by analyzing changes in spectral reflectance related to water content and stress^{[10,21,75,76]}. Those VIs specifically related to major physiological traits that control WUE are listed in Table 1 and depicted in Fig. 1. Several VIs exhibited associations with multiple components of WUE; these multifunctional indices are summarized in Table 2.

Figure 1. 

Network visualization of vegetation indices (VIs) and water-use efficiency (WUE) components. WUE is the central node, surrounded by its physiological traits: carbon isotope discrimination (Δ¹³C, in green), chlorophyll (in brown), photosynthesis (A, in red), stomatal conductance (g_s, in teal), leaf area index (LAI, in blue), and water content (in purple). Vegetation indices are arranged in the outer ring and connected to specific physiological traits by lines. Lines connect each VIs to its corresponding WUE component. VIs associated with multiple WUE components are displayed in the color of the additional traits they relate to and are summarized separately in Table 2 to highlight their versatility.

Vegetation index related to LAI

LAI affects WUE through transpiration and photosynthetic capacity^[77]. As LAI increases, the surface area for transpiration increases. However, higher LAI also increases the potential for photosynthesis and carbon uptake, which can improve WUE, but only up to an optimal point where water loss and water availability are balanced^[78]. LAI has a direct but non-linear relationship with WUE^[79]. NDVI, Difference Vegetation Index (DVI), EVI, and SR were highly and significantly correlated with LAI in different plant species, as reported in winter wheat^[80,81] (Fig. 1; Table 1). Ratio Vegetation Index (RVI), NDVI, Green Normalized Difference Vegetation Index (GNDVI), Modified Soil Adjusted Vegetation Index (MSAVI), SAVI, Optimized Soil Adjusted Vegetation Index (OSAVI), and EVI efficiently described LAI in different growth stages of cotton (Gossypium hirsutum) under irrigation, with coefficients of determination of R² ranging from 0.94 to 0.5^[82]. LAI was strongly positively associated to EVI (r > 0.8), OSAVI (0.6 < r < 0.8), GNDVI (r > 0.8), Triangular Vegetation Index (TVI, 0.6 < r < 0.8), and modified Triangular Vegetation Index (MTVI, 0.6 < r < 0.8), but strongly negatively correlated with Normalized Difference Moisture Index (NDMI, r < −0.6) and Normalized Difference Water Index (NDWI, r < −0.8) in wheat exposed to drought stress^[83]. Those aforementioned studies indicate that VIs, including NDVI, EVI, DVI, SR, RVI, GNDVI, MSAVI, TVI, SAVI, OSAVI, and MTVI, show strong correlations with LAI across various crops and growth stages, while moisture-related indices such as NDMI and NDWI may show negative relationships under drought conditions, highlighting the effectiveness of spectral indices in estimating LAI under both irrigated and water-stressed environments.

Vegetation index related to chlorophyll

Chlorophyll absorbs sunlight to provide energy for photosynthesis, where carbohydrates are produced in leaves. Spectral reflectance at the red edge provides valuable information about plant chlorophyll concentration^[84]. Chlorophyll content has been estimated by Chlorophyll Vegetation Index (CVI) in wheat under drought^[27]. Bell et al.^[55] found that GNDVI and NDVI were strongly correlated to chlorophyll (R² = 0.7;0.75) and nitrogen concentration (R² = 0.76; 0.81) in bermudagrass (Cynodon dactylon) and creeping bentgrass (Fig. 1; Table 1). Some vegetation indices, such as Normalized Difference Index (NDI) NDI_{602, 598, 600}, NDI _{644, 630, 652}, NDI_{648, 662, 624}, were significantly correlated with gravimetric WUE, with the R² ranges in 0.19−0.25, also showed significant correlation with chlorophyll (R²: 0.18−0.58) across different levels of drought stress^[54]. Overall, indices such as NDVI, CVI, GNDVI, and specific NDI variants have proven reliable for estimating chlorophyll, nitrogen concentration, and WUE across different crops and drought conditions, highlighting their value for precision monitoring of plant health and stress responses.

Vegetation index related to photosynthesis

Photosynthesis provides sugar and energy for biomass accumulation and is related to WUE in various ways, from leaf-level instantaneous WUE to whole plant scales. The light wavelength of photosynthetically active radiation ranges from 400 to 700 nm in plants^[85,86]. Some indices can indirectly estimate canopy photosynthesis, such as DVI, NDVI, Red Difference Vegetation Index (RDVI), SR, SAVI, and Transformed Soil Adjusted Vegetation Index (TSAVI)^[87,88] (Fig. 1; Table 1). PRI estimates the radiation-use efficiency by plants, related to photosynthetic rate and chlorophyll fluorescence^[89]. Some vegetative indices, such as red edge NDVI (NDRE), PRI, and EVI, showed significant correlation (R², 0.88−0.49) to chlorophyll fluorescence (estimated as F_v/F_m) in aspen (Populus tremuloides) and cherry (Prunus avium) tree leaves under drought^[90]. PRI also showed a significant correlation of r = 0.77 with epoxidation state and zeaxanthin in a cereal canopy^[91]. NDVI, SR, and GNDVI were significantly and positively correlated (r ranges from 0.58−0.68) with photosynthetic rate under both well-watered and water-deficient conditions in different tomato (Solanum lycopersicum) genotypes^[92]. In a winter wheat study, 12 out of 330 vegetation indices showed the highest correlation with instantaneous WUE (A/E) under drought, with R² values between 0.53 and 0.57^[39]. These indices were ranked as (normalized difference of the double-peak areas based on REPig [red edge positions derived by maximum inverted Gaussian fitting] division) NDDAig, the minimum spectral reflectance corresponding to the chlorophyll absorption well (Lo), EVI-1, RDVI, PRI, Readone, Green Vegetation Index (GVI), Plant Senescence Reflectance Index (PSRI), Structure Insensitive Pigment Index (SIPI), TSAVI, and DVI, furthermore, the water efficiency index (WEI = NDDAig/FWBI) were constructed for better sensitivity of WUE changes^[39,42]. Those VIs of estimating canopy photosynthesis, chlorophyll fluorescence, and radiation-use efficiency, along with specialized metrics like WEI, offer valuable tools for monitoring photosynthetic performance and WUE across different crops and water conditions, particularly under drought stress.

Vegetation index related to stomatal conductance

Stomatal conductance regulates the flux of water vapor and CO₂ in plants, which are highly associated with WUE. Vegetation indices, OSAVI and Normalized Green Red Difference Index (NGRDI), were selected as the best indicators for stomatal conductance under mild to moderate drought in sugarcane^[13]. NDVI, SR, and GNDVI were significantly and positively correlated with net photosynthetic rate and stomatal conductance under both well-watered and water-deficient conditions in tomato, with r values ranging between 0.58−0.73^[92]. NDWI and simple ratio water index (SRWI) had the highest prediction (with high regression values of R² = 0.65, 0.66, respectively) for intrinsic WUE tested with 40 cotton cultivars in the field condition, while NDWI 2130 and NDWI 1640 showed regression values of R² = 0.36 for both) didn't seem to be effective in this study^[50]. Overall, VIs, such as OSAVI, NGRDI, NDVI, SR, GNDVI, NDWI, and SRWI, have demonstrated strong potential for monitoring stomatal behavior and predicting WUE across different crops and water conditions, though their effectiveness may vary depending on the specific index and crops studied.

Vegetation index related to leaf and canopy water content

Canopy water status-related vegetation indices include SRWI, NDWI, SWWI, normalized water index (NWI), NDMI, WBI, WEI (Fig. 1; Table 1). Some vegetation indices using near-IR wavelengths, like NDWI, could penetrate deeper to see through multiple leaf layers, which may benefit the estimation of canopy water status for dense and overlapping turfgrass leaves^[93,94]. NDWI 1,630 nm and NDWI 2,130 nm had shown promising potential to estimate vegetation water content^[19,52,95]. Leaf water potential, RWC, and canopy temperature were significantly and strongly correlated with NWI-3 (r = −0.49, −0.24, 0.49) in wheat under drought^[96]. In maize, canopy water content was found strongly correlated (R² > 0.72) with Clred edge, NDVI, WBI, and NDWI under drought^[95]. WBI was used to predict soil water content in crops and turfgrass, but it was found to be significantly and highly correlated (r = 0.52 − 0.85) to total chlorophyll and tissue water content in creeping bentgrass^[97]. A strong negative correlation (r = −0.52) was found between WBI and T under drought in some plant species, such as corn^[14]. WBI varies among plant species and is significantly correlated with water and CO₂ flux under drought, suggesting the potential for estimating WUE^[98]. The SRWI uses reflectance bands (typically R₈₅₈/R₁₂₄₀) where water absorption features in the SWIR region are prominent. While some water indices designed for the leaf scale may not perform well at the canopy scale, the SRWI has demonstrated good performance at the canopy level, likely due to its strong response to the significant changes in canopy water content that occur in a canopy over time or under stress conditions, suggesting the reliable measurement of canopy water content from SRWI^[99]. The VIs, particularly those utilizing near-infrared and shortwave infrared wavelengths, such as WBI, SWWI, NDWI, and SRWI, are valuable for assessing water content in dense canopies, penetrating multiple leaf layers, and capturing temporal changes under stress. Their correlations with leaf water potential, relative water content, canopy temperature, transpiration, and stomatal conductance highlight the utility for estimating WUE and managing crop water status across diverse plant species and environmental conditions.

Vegetation index related to carbon isotope discrimination (Δ¹³C)

Carbon isotope discrimination was used to estimate WUE among six creeping bentgrass cultivars to reveal their drought-tolerant mechanisms^[100]. However, the spectral indices associated with Δ¹³C are not well studied in grass species under drought. A UAV remote sensing study demonstrated that vegetation indices such as NDVI, SR, and GNDVI were significantly correlated with A (r = 0.58−0.68), g_s (r = 0.58−0.73), WUE_intrisic (r = −0.66−0.51), and Δ¹³C (r = −0.52 to −0.17) under both well-watered and water deficient conditions in tomato^[92]. Under drought, CVI has shown strong negative correlation (r = −0.75) with δ¹³C in wheat^[27]. The correlation between NDVI and δ¹³C was significant under drought but not promising in well-watered conditions in durum wheat^[101]. Besides NDVI, RGB-based vegetation index green area (GA) and greener area (GGA)^[102,103] showed significant correlation (r = −0.49 to −0.44) with δ¹³C under rainfed and irrigated conditions in bread wheat^[104]. In a panel of 368 wheat genotypes, several water-related indices [NDMI, NWI, SWWI, WBI, NDSI] and vegetative indices (MSAVI, Weighted Difference Vegetation Index [WDVI], SAVI, MTVI, and OSAVI) exhibited strong correlations with Δ¹³C (r up to 0.76), under mild drought stress or irrigated conditions, outperforming NDVI (r = 0.51)^[23]. However, Lobos et al.^[23], also reported that these indices were not effective predictors under severe drought. Visible-near infrared spectral data showed accurate and precise estimation of E, A, g_s, Δ¹³C, WUE_intrinsic, WUE_instantenous, with the R² values ranging from 0.85 to 0.91 across different drought stress conditions, based on partial least squares regression (PLSR) model using spectroradiometer in foxtail millet (Setaria italica)^[105].

While Δ¹³C has been effectively used to characterize physiological responses to water stress, the integration of spectral indices as indirect, high-throughput proxies remains insufficiently explored in grass species. Evidence from other crops demonstrates that several vegetation indices—such as NDVI, SR, GNDVI, and RGB-based indices (GA and GGA)—can show significant relationships with Δ¹³C and WUE-related traits, particularly under drought or mild stress conditions. However, the strength and reliability of these relationships often vary with stress severity, as some indices lose predictive power under severe drought.

In summary, as depicted in Fig. 1, LAI was closely associated with MSAVI, MTVI, SAVI, SR, DVI, EVI, RVI, TVI, PRI, NDVI, TSAVI, GNDVI, and OSAVI. Chlorophyll content correlated with NDRE, SIPI, CVI, GNDVI, Lo, MCARI/OSAVI, MCARI2, NDDAig, NDVI, PSSRa, and Readone. Photosynthetic activity was linked to ARI, NDSI, PRI, PSRI, R701, TSAVI, WBI, NDRE, and SIPI, while stomatal conductance was associated with OSAVI, NGRDI, and R702. Canopy and leaf water status showed strong relationships with WBI, NDSI, WEI, SRWI, NWI3, NDWI, and Clred edge. Δ¹³C related VIs include NMI, NWI, WDVI, NDSI, WBI, CVI, NDVI, GNDVI, MSAVI, MTVI, NDMI, OSAVI, SAI, SR, and SWWI. The indices NDRE, NDSI, PRI, R701, SIPI, TSAVI, WBI, OSAVI, MTVI, MSAVI, CVI, SAVI, SR, NDMI, SWWI, GNDVI, and NDVI are clearly linked to two or more traits associated with WUE (Table 2; Fig. 1). Indices like NDSI, OSAVI, NDVI, and GNDVI are the most versatile, covering three traits. SWWI, NDMI, NDWI, and WBI are particularly valuable for water-related monitoring. OSAVI links canopy leaf areas and stomatal behaviors, making it useful under drought conditions. These indices offer a reliable framework for monitoring physiological and water-related traits, ranging from LAI and chlorophyll content to A, g_s, water status, and Δ¹³C, enabling accurate assessment of WUE and drought tolerance across diverse species.

Although numerous VIs have been reported to correlate with WUE-related traits, their performance is not ubiquitous in all circumstances and is influenced by both physiological and methodological factors. The effectiveness and accuracy of VIs to estimate WUE depend on plant species and genotype, as well as environmental conditions, including atmospheric and soil factors. Remote sensing-based VIs have been successfully used to detect genotypic variations in durum wheat, bread wheat, and bermudagrass in varying climatic conditions, suggesting a promising potential in accelerating breeding programs^[106−108]. The sensitivity of VIs varies with drought progression. In early/mild drought, physiological changes such as reductions in photosynthetic efficiency and initial declines in leaf water status occur before visible structural damage; therefore, pigment- and water-sensitive indices, such as the PRI, Clred-edge, SIPI, GNDVI^[109−112], and water-related indices (such as NDWI)^[113,114], are most responsive. As drought progresses, photosynthetic rates decline further, accompanied by chlorophyll degradation, change of leaf angle (leaf rolling or wilting), and leaf abscission (Table 3). Under prolonged or severe drought, more pronounced structural changes occur, including leaf senescence, canopy loss, and reduced leaf area, which become dominant drivers of spectral signals. Consequently, structural indices, including NDVI, EVI, and SAVI^{[110,115−117]}, along with water-related indices (such as WBI, NDWI)^[113,118], generally provide more reliable estimates of WUE-related traits at this stage (Table 3).

Improving plant drought tolerance has become an urgent priority under increasingly adverse climatic and environmental conditions, with WUE serving as a central trait for adaptation. WUE can be assessed through instantaneous gas exchange measurements or through Δ¹³C, which reflects integrated, whole-plant performance over time. However, conventional approaches for measuring WUE are labor-intensive and time-consuming, limiting their application in large-scale screening and breeding programs. This review highlights the potential of remote sensing–derived VIs to estimate key physiological components associated with WUE, including LAI, A, g_s, water status, chlorophyll content, and Δ¹³C, across grass species ranging from annual cereal crops to perennial turf and forage grasses. However, the limited research in grass species, especially under varying drought intensities, underscores the need for further investigation to validate and optimize spectral indices for reliable drought tolerance assessment in these systems. Integrating UAV-based multispectral or hyperspectral imagery with advanced machine learning models could improve prediction accuracy and scalability. Additionally, exploring the physiological basis linking spectral reflectance to Δ¹³C will enhance mechanistic understanding and model transferability across species. For instance, stomatal regulation affects intercellular CO₂ concentration and transpiration, thereby influencing both Δ¹³C and spectral indices (such as OSAVI) associated with leaf water content^[53]. Long-term and multi-environment validation studies are also essential to determine the reliability of spectral proxies for breeding and management applications. Ultimately, refining non-destructive, high-throughput phenotyping tools for Δ¹³C and WUE estimation will support precision irrigation strategies and accelerate the selection of drought-tolerant turfgrass cultivars.