Ⅰ. INTRODUCTION

Italian ryegrass (IRG) has emerged as a crucial component in the domestic forage market, with its cultivation area steadily increasing. For instance, the area under IRG cultivation expanded from 61,000 hectares in 2019 to 73,000 hectares in 2021. This expansion represents a key strategy aimed at enhancing forage self-sufficiency, with IRG playing a pivotal role in forage production. However, environmental factors such as climate change, drought, and abnormal low temperatures may disrupt the supply of IRG, which could severely impact the domestic forage market. Consequently, the importance of efficient production and management of IRG has been underscored, necessitating the development of precise yield prediction technologies. Vegetation indices serve as important indicators for assessing plant distribution, vitality, chlorophyll content, and photosynthetic activity, with over 100 indices developed to date. Rouse et al. (1974) introduced the normalized difference vegetation index (NDVI), which generalized vegetation reflectance characteristics by comparing red and near-infrared wavelengths. Traditionally, data over large areas have been collected through satellite and aerial imaging; however, obtaining data at the desired time points remains a challenge, and the high cost of these technologies limits their accessibility to general users (Gitelson et al., 1996;Ahmed et al., 2019). Recently, the integration of unmanned aerial vehicle (UAV) remote sensing technology with geographic information systems (GIS) has enabled more precise, location-based analysis (NIPA, 2017). In South Korea, spectral data collection and dry matter yield (DMY) estimation for food and forage crops have been conducted based on growth stages. For example, in a study aimed at estimating cereal crop conditions using UAVs, a linear relationship between vegetation indices and crop growth factors was reported (Na et al., 2016). In addition, research on winter forage crops using automatic ground-based spectral measurement systems revealed that NDVI reached its peak during the early growth stages of IRG, rye, and barley, rendering DMY predictions inaccurate at later stages (Shin et al., 2021). International studies have also demonstrated the utility of UAVs in estimating crop growth characteristics and yields (Xiang and Tian, 2011;Bendig et al., 2014). In the Netherlands, attempts have been made to integrate crop and soil estimation models into decision support systems for crop management, while studies on the application of vegetation indices using optical sensing technology have also been conducted (Gimplinger and Kaul, 2009). While prediction technologies for food crops such as cereals and cabbage have been developed to address supply instability, the development of similar technologies for forage crops like IRG remains limited. Thus, the accurate prediction of IRG yield is essential for stabilizing the supply of domestic forage.

Therefore, this study aims to accurately assess the growth characteristics of IRG using UAV-based vegetation index measurements, and to predict DMY based on these indices.

Ⅱ. MATERIALS AND METHODS

1. Study site and plant material

The experiment was conducted at the experimental fields of the National Institute of Animal Science, from September 19, 2022, to May 19, 2023. The IRG variety ‘Kowinnerly’ was used as the experimental crop. Seeds were sown using row planting methods at a rate of 30 kg/ha.

2. Soil environmental data collection

Soil temperature and moisture were monitored continuously using soil sensors (DT-SMS01B, Damoa Tech, Korea) installed at two depths: 15 cm and 30 cm. These sensors continuously transmitted data throughout the cultivation period.

3. Soil composition and chemical analysis

Soil samples were collected and analyzed for physical and chemical properties. Soil composition was analyzed for various parameters, including moisture content, pH, ammonium nitrogen (NH₄-N), phosphorus pentoxide (P₂O₅), exchangeable potassium (Ex-K), total nitrogen (T-N), organic matter (OM), cation exchange capacity (CEC), and electrical conductivity (EC).

4. UAV data acquisition and hyperspectral post-processing

A total of six field investigations were carried out to collect UAV observation and growth data, with four datasets utilized for the analysis. For each investigation, data were collected from six 1 m² sample points per observation (Fig. 1). The primary UAV used for data collection was equipped with a hyperspectral camera (miscoHSI 410 SHARK, Corning, USA), mounted on a DJI Matrice 300 RTK UAV (DJI, China). The UAV was flown at an altitude of 40 meters on clear days around noon, with a longitudinal and lateral overlap of more than 70%, ensuring optimal data acquisition conditions. The hyperspectral data were collected across a wavelength range of 400–1000 nm, divided into 150 spectral bands. A tarp with uniform reflectance values of 11%, 30%, and 56% was installed within the shooting range for atmospheric correction, and a GPS surveying device (Trimble R4s, Texas, USA) was used for GPS correction. Raw hyperspectral data were subsequently processed using specialized software (ENVI 5.7, NV5 Geospatial, USA) to conduct radiometric calibration, geometric correction, image mosaicking, atmospheric correction, and vegetation index calculation. The vegetation indices used in this study are detailed in Table 1. NDVI and NDRE, two of the most widely used vegetation indices for biomass estimation and time-series analysis, were utilized in this study. The GLI, developed for the assessment of wheat cover, was also employed; negative values indicate soil and non-living features, while positive values correspond to green leaves and stems. In addition, optimized soil-adjusted vegetation index (OSAVI), optimized for relatively sparse vegetation where soil is visible through the canopy, was applied.

5. Vegetation cover analysis

To assess the vegetation changes before and after overwintering, the GLI was utilized to calculate the vegetation cover fraction. The GLI index, commonly used to distinguish between vegetative areas and non-vegetative areas, was applied to segment the crop areas (GLI > 0) from the other regions (GLI ≤ 0). The time-series vegetation cover fraction (VF) was then calculated by dividing the vegetative area by the total field area using the following formula:

6. Plant growth and nutritional analysis

Plant growth data, including measurements of plant height and DMY, were collected periodically during April and May, when crop growth was most vigorous after overwintering, and were matched with the UAV observational data. Growth stages were identified based on visual assessments, and DMY was calculated by harvesting plants at specific stages and drying them in an oven at 70°C for 72 h.

In addition, chemical composition analyses were conducted for crude protein (CP), neutral detergent fiber (NDF), and acid detergent fiber (ADF) contents. CP was determined using the Kjeldahl method (AOAC, 1990), and NDF and ADF were measured using an Semi-Automated Crude & Detergent Fiber (Ankom 200 fiber analyzer, Ankom Technology, USA). The results of the chemical composition analysis for different growth stages are summarized in Table 3.

7. Statistical analysis

Data such as chemical compositions were tested using one-way analysis of variance (ANOVA), and significant differences among the means were identified using Duncan’s multiple range test. Correlation between the vegetation indices and growth parameters, such as plant height and DMY, was also determined using SAS Enterprise Guide software (ver. 9.1, SAS Institute, USA). The strength of the relationships between each vegetation index and the growth parameters was evaluated using the coefficient of determination (R²), with the statistical significance of these relationships assessed through p-values (p<0.05).

Ⅲ. RESULTS AND DISCUSSION

1. Soil environmental data and composition analysis

Over the course of the study, soil temperature gradually increased, reaching a maximum of 16°C, while soil moisture content fluctuated based on depth (Fig. 2). Soil moisture ranged between 29% and 39% at a depth of 15 cm, whereas at 30 cm, it remained between 35% and 38%. The 15 cm depth was more susceptible to external rainfall, as indicated by the recorded data. As shown in Table 2, the soil used in the experiment was classified as sandy loam, with a pH of 7.86, a moisture content of 4.21%, and an average P₂O₅ concentration of 249.98 mg/kg. Other relevant characteristics, including NH₄-N (11.70 mg/kg) and Ex-K (0.15 cmolc/kg), were measured to understand the nutrient availability during the cultivation period.

2. Chemical composition analysis

The chemical composition of IRG was analyzed at four key time points during the growing season (April 10, April 21, April 28, and May 4, 2023). As shown in Table 3, moisture content peaked on April 21 (82.01%) and then declined toward the end of the study period. The crude protein content reached its maximum one week before heading (April 28) at 15.98%, followed by a rapid decline to 7.32% by May 4. This trend indicates a correlation between plant maturity and nutritional composition, where fiber components, including NDF and ADF, increased as the growth cycle progressed. NDF and ADF values rose steadily after April 21, showing peaks of 51.43% and 25.18%, respectively, before further increases were observed in May.

3. Correlation analysis between growth parameter and observation data

An analysis of the relation between plant height and DMY revealed a linear increase in both parameters with growth stages. Both plant height and DMY showed a strong correlation (R² = 0.975), as illustrated in Fig. 3. This indicates a high degree of reliability in predicting DMY based on plant height measurements. In Fig. 4, a correlation analysis between five vegetation indices and plant height showed that both NDVI and GLI had relatively high coefficients of determination. However, NDVI exhibited saturation at later growth stages, while GLI maintained the strongest correlation throughout the study period (R² = 0.948). GLI demonstrated a consistent upward trend as plant height increased, indicating its suitability for accurately reflecting plant growth in IRG. These findings align with those reported by Na et al. (2016), who identified NDVI and GLI as effective indicators for predicting the growth stages of barley and wheat using remote sensing technology.

This study further applied UAV-based remote sensing to analyze the relationship between IRG growth characteristics and DMY using various vegetation indices, including NDVI and GLI. Among these indices, GLI exhibited the highest correlation with DMY (Fig. 5). The GLI-based DMY estimation model derived in this study demonstrated a high degree of accuracy, with a coefficient of determination (R² = 0.971), which represents a significant finding.

GLI, which reflects the distribution and vitality of green vegetation, proved to be particularly effective in assessing the growth status of IRG, a critical forage crop. Throughout the study period, GLI consistently increased in tandem with IRG growth stages, showing a strong correlation between plant height and DMY. In contrast, NDVI experienced saturation at certain growth stages, resulting in reduced accuracy in yield prediction. However, GLI consistently maintained a high level of accuracy, effectively overcoming the limitations posed by NDVI. The results of Na et al. (2016) further support the conclusions of this study, confirming the utility of NDVI and GLI for predicting the growth of barley and wheat. Similarly, Bendig et al. (2014) demonstrated the value of UAV-based RGB imaging in estimating barley biomass and highlighted the potential of UAV data collection for crop growth analysis. While Bendig et al. (2014) utilized RGB imaging, this research employed hyperspectral cameras, which provided greater precision in data collection and allowed for more accurate predictions of IRG DMY.

The GLI-based DMY estimation model derived from this research offers a practical tool for predicting forage yields, especially under fluctuating environmental conditions such as those caused by climate change. Compared to other vegetation indices (e.g., NDVI, OSAVI), GLI demonstrated the highest coefficient of determination (R² = 0.971), confirming its reliability as an indicator for forage crop management. This result is consistent with the findings of Rondeaux et al. (1996), who introduced the OSAVI (Optimized Soil Adjusted Vegetation Index), suggesting that vegetation indices can serve as useful tools for predicting crop yields at various growth stages. Therefore, this suggests that UAV-based vegetation index analysis can serve as an essential tool for managing forage production effectively.

4. Development of DMY estimation models

Using the vegetation indices employed in this study, DMY estimation models were derived as follows:

To quantitatively assess plant growth before and after the winter season, vegetation coverage was analyzed. UAV data collected on November 24, 2022 (before winter) and March 27, 2023 (after winter) were converted into GLI vegetation index maps. Pixels with GLI values exceeding 0.1 were counted to estimate vegetation coverage. While a GLI value greater than 0 is typically used to determine plant presence, in this study, a threshold of 0.1 was set to clearly differentiate vegetation from non-vegetative areas such as soil. Compared to the coverage on November 24, the post-winter coverage on March 27 was approximately 54%, corresponding to an area of 1.08 hectares.

The GLI-based DMY estimation model, which had the highest coefficient of determination (R²) among the relationships between vegetation indices and DMY, was selected, and the DMY estimation process is presented in Fig. 6. By applying the GLI-based DMY estimation model to the experimental plots, the estimated yields were 1,860.3 kg/ha on April 28 and 2,681.9 kg/ha on May 4. These results suggest that UAV-based remote sensing technology could serve as a valuable tool for predicting the yield of forage crops like IRG. Ahmed et al. (2019) demonstrated the feasibility of using remote sensing for crop yield estimation by monitoring corn yield variability with Sentinel-2 and machine learning techniques. Although their study used satellite data, both their research and ours successfully predicted crop yields based on high-precision remote sensing data. The primary distinction lies in the platform and crop type: Ahmed et al.'s study focused on corn, a major staple crop, whereas this study concentrated on forage crops like IRG. Moreover, Bendig et al. (2014) explored UAV-based RGB imaging for barley biomass estimation, highlighting the potential of UAVs in crop growth monitoring. While their study employed relatively cost-effective RGB imaging, the hyperspectral cameras used in this study provided higher spectral resolution, enabling more accurate analysis of crop growth. When compared to previous studies, this research demonstrates a unique contribution to forage crop analysis, specifically by leveraging UAV-based hyperspectral data for more precise prediction of plant growth and DMY. This represents a significant advancement in the field. The technical information on GLI derived from hyperspectral data further demonstrates the advantages of this approach. The results of this study confirm that GLI-based DMY estimation can be highly accurate across different growth stages of IRG. This approach has the potential to significantly improve future crop management and supply forecasting.

Ⅳ. CONCLUSIONS

This study successfully demonstrated the effectiveness of utilizing the Green Leaf Index (GLI) for predicting the growth and DMY of IRG through unmanned aerial vehicle (UAV)-based remote sensing technology. The GLI-based DMY estimation model exhibited a high degree of accuracy, with a coefficient of determination (R² = 0.971), making it a practical and reliable tool for predicting forage crop yields, especially in response to environmental variability. Among the vegetation indices tested, GLI consistently outperformed others such as NDVI and OSAVI in both growth stage monitoring and yield estimation. These findings suggest that GLI can be applied as an essential tool for improving crop yield predictions and enhancing forage production management, particularly under the influence of climate change.