Determination of UAV Flight Altitude and Time for Optimizing Variable-Rate Nitrogen Prescription Maps for Winter Wheat in the North China Plain

Abstract

An unmanned aerial vehicle (UAV) multi-spectral system provides a monitoring platform to rapidly obtain crop spectral information that can reflect crop nitrogen status for the generation of dynamic variable-rate nitrogen (VRN). To improve the accuracy of VRN prescription maps, a method of generating VRN prescription maps on the basis of the vegetation index was proposed, and the effects of UAV flight time and altitude on VRN prescription maps were analyzed. The experimental site was located in Dacaozhuang, Hebei Province, China, and the experimental crop was winter wheat (Lunxuan 145). The flight altitudes of the UAV system were set to 50, 70 and 90 m. The flight times were set to 8:00 a.m., 11:00 a.m., 2:00 p.m. and 5:00 p.m. local time. The flight area was 1.18 ha with a 60° rotation angle under a three-span center pivot irrigation system with an overhang. UAV flight missions were executed during the jointing, heading, and grain filling phases of winter wheat. There were 90 management zones with pie shapes in total, which were composed of a 10° angle in the rotation direction and 4 sprinklers along the lateral direction. The vegetation indices (VIs) which are closely related to crop nutrient status were selected and used to generate distribution maps, which were superimposed with the management zones to generate VRN prescription maps. The results demonstrated that the red-edge soil adjusted vegetation index (RESAVI) was relatively more sensitive to the nitrogen status of winter wheat than the other VIs were. The RESAVI distributions were stable during periods with a solar elevation angle greater than 50° (11:00 a.m.–2:00 p.m. local time), and the VRN prescription maps were similar, with the overlap percentage of the same fertilization grade being greater than 80% and the relative error of the fertilization amount being less than 5%. Compared with that at 2:00 p.m., the overlap percentage of the same fertilization grade was 56.6% in both seasons at 8:00 a.m., whereas flights at 5:00 p.m. exhibited overlaps of 70.9% and 44.6% in the 2023 and 2024 seasons, respectively. Conversely, the flight altitude had little influence on the fertilizer amount and VRN prescription maps. The difference in the amount of fertilizer used was less than 3% at different flight altitudes. The required time is half of that for a 50 m flight when the flight altitude is 70 m and one third of that when the flight altitude is 90 m. Our study recommended operating the UAV multi-spectral system at solar elevation angles greater than 50° when generating VRN prescription maps of winter wheat, and the flight height can be adjusted according to the field area and the endurance time of the UAV.

1. Introduction

Nitrogen (N) is a fundamental element in plants that supports crop growth. Its dynamic supply influences the accumulation of photosynthates (e.g., carbohydrates), grain development, and yield in winter wheat [1]. The traditional uniform fertilization method, in which a single fertilizer rate is applied across the entire field, is prone to wasting nitrogen resources. This phenomenon occurs in larger fields where the spatial variability in soil nutrients is more pronounced but is often neglected because of technical limitations. The traditional practice not only reduces the efficiency of nitrogen fertilizer utilization but also leads to environmental pollution, such as nitrate leaching and greenhouse gas emissions. Variable-rate nitrogen (VRN) management can spatially and temporally regulate the nitrogen supply in the soil, further increasing nitrogen use efficiency and reducing environmental risk [2].

As the core infrastructure of modern intensive agricultural production systems, center pivot sprinkler irrigation systems have evolved into critical components of agricultural water management infrastructure because of their extensive operational coverage and high-efficiency performance characteristics. Most research on precision management via center pivot sprinkler irrigation systems has focused on the development of variable-rate irrigation decision support systems to improve water productivity [3,4], whereas research on VRN technologies integrated with center pivot systems is relatively lacking. At present, VRN management is realized mostly by changing the working speed of fertilizer spreaders [5,6,7]. Owing to the limited applicability of spreaders in the middle and late growth stages of tall crops whose canopy heights exceed spreader ground clearance, VRN management is mostly confined to the early growth stages. Furthermore, implementation often relies on historical data or sensors mounted on tractors to monitor soil and crop conditions [5,8,9], which increases the complexity of field application. These limitations point to the need for a practical VRN approach suitable for integration with center pivot systems.

The development and application of unmanned aerial vehicle (UAV) multi-spectral systems provide a non-destructive method for determining crop N status and real-time guidance for in-season N management. Shi [10] reported that in the monitoring of chlorophyll content in winter wheat, the characteristic spectral bands were predominantly located within the red-edge to near-infrared (NIR) range. Liu et al. [11] reported that the vegetation index DATT, which integrates the NIR, red-edge, and red spectral bands, exhibited superior performance in estimating the nitrogen nutrition index (NNI), achieving a coefficient of determination (R²) of 0.95. Zhang et al. [12] reported that the red-edge soil adjusted vegetation index (RESAVI), which is composed of red-edge and NIR bands, achieved the highest accuracy in predicting the nitrogen content in winter wheat (R² = 0.78), and they established a variable nitrogen application model for winter wheat on the basis of RESAVI, demonstrating that the nitrogen application amount could be reduced by 15.4% without compromising yield. These findings demonstrated the feasibility of generating VRN prescription maps based on vegetation indices (VIs) obtained from an UAV multi-spectral system.

The flight time and altitude of a UAV are key technical parameters that affect the spatiotemporal resolution of multi-spectral system sampling. In VRN management, UAV multi-spectral systems are mainly used to obtain indicators such as yield, chlorophyll content, and plant nitrogen uptake [13,14,15], and the selection of flight time for UAVs generally focuses on the time period around midday. For example, when estimating winter wheat yields, Hu et al. [16] selected a flight time of 11:00 a.m. to 2:00 p.m. local time and a flight altitude of 30 m, whereas Kang et al. [17] chose a flight time between 2:00 p.m. and 3:00 p.m. local time at a flight altitude of 25 m. When estimating the chlorophyll content in winter wheat, Zhang et al. [18] selected a flight time from 11:00 a.m. to 1:00 p.m. at an altitude of 30 m, whereas Zhou et al. [19] conducted flights between 3:00 p.m. and 4:00 p.m. local time at an altitude of 60 m. Liu et al. [20] reported that the optimal prediction of the chlorophyll content was at a height of 60 m, and the most predictive result for the LAI was at 30 m. When estimating plant nitrogen uptake, Wang et al. [21] chose a flight time from 10:00 a.m. to 2:00 p.m. local time and operated at an altitude of 25 m, whereas Zhang et al. [12] conducted flights at an altitude of 80 m. The flight altitudes selected in the above studies appeared to be arbitrary. Mesas et al. [22] demonstrated that flight altitudes ranging from 60 to 100 m could distinguish vegetation from bare soil. The above conclusions were derived from the relationships between spectral indicators under different flight parameters and ground-measured data. There is a lack of guidelines for selecting these parameters specifically for generating VRN prescription maps. The influence of different flight times and altitudes on VRN prescription maps generated from VIs obtained by UAV multi-spectral systems remains to be further studied.

To realize accurate VRN management based on a UAV multi-spectral system, this paper proposes a method for generating VRN prescription maps on the basis of VIs and quantifies the effects of UAV flight time and altitude on the consistency, stability, and operational efficiency of VRN prescription maps, aiming to promote the reliable application of UAV multi-spectral systems in dynamic VRN management.

2. Materials and Methods

2.1. Experimental Site

The experiment was conducted at the Management Area of Dacaozhuang in the Hebei Province of the North China Plain (37°37′12″ N, 114°54′01″ E). This site is characterized by a temperate continental monsoon climate, with an average annual precipitation of 465 mm and a mean annual temperature of 13 °C. The experimental area was 1.18 ha, which was 1/6 of the area controlled by a three-span 150 m long center pivot machine with an overhang (Figure 1). The sprinkler system was equipped with 59 Nelson D3000 non-rotating sprinkler nozzles installed at a height of 1.5 m, with a spacing of 2.5 m between nozzles, and the water application efficiency exceeded 80%, as determined by measuring the difference in soil water content before and after an irrigation event. The soil type at the experimental site was silt loam, and the experimental crop was winter wheat. A meteorological station was installed in the central part of the experimental area to continuously monitor air temperature, relative humidity, solar radiation, wind speed, and rainfall at 15 min intervals.

2.2. Field Experiment Layout

The field experiment was conducted during the 2022–2023 and 2023–2024 growing seasons, with winter wheat (Lunxuan 145) as the experimental crop. The winter wheat was sown at 14 cm row spacing with a seeding rate of 375 kg ha⁻¹. Sowing occurred on 17 October 2022 and 12 October 2023 for the two growing seasons, with corresponding harvest dates occurring on 9 June 2023 and 7 June 2024. The duration of each phenological stage is shown in

Table 1. Duration of each phenological stage.

Growth Stage	Date
Sowing–Seedling Stage	12 October–30 November
Overwintering Period	1 December–10 February
Green-Up Stage	11 February–30 March
Jointing Stage	31 March–20 April
Heading Stage	21 April–30 April
Grain-Filling Stage	1 May–27 May
Maturity Stage	28 May–7 June

. The N application rate was calculated on the basis of the target yield and the nitrogen uptake requirements per 100 kg of grain production [23]. In both seasons, the base fertilizers applied were 70 kg ha⁻¹ N, 85 kg ha⁻¹ P₂O₅, and 26 kg ha⁻¹ K₂O. Five topdressing nitrogen rate treatments (N1–N5) were set up with equal areas. The total topdressing N rates for N1 were 160 kg ha⁻¹ and 150 kg ha⁻¹ in the 2023 and 2024 seasons, respectively, and the total topdressing nitrogen rates for N2–N5 were 75%, 50%, and 25% of those of N1, and 0 kg ha⁻¹, respectively. The plots were arranged in a complete block design with three replications, resulting in 15 plots, each covering an area of 0.08 ha. Each treatment was replicated three times, as shown in Figure 2. In the 2023 season, topdressing was applied at ratios of 20%, 55%, and 25% at the regreening stage, jointing stage, and early filling stage, respectively. In the 2024 season, topdressing was carried out at ratios of 70% and 30% at the early jointing stage and heading stages, respectively.

All the treatments in the experimental area adopted the same irrigation regime. In the 2023 season, 33 mm of emergence irrigation was applied after sowing. To meet the topdressing requirements, 20 mm, 20 mm, and 15 mm of irrigation were applied at the regreening stage, jointing stage, and early filling stage, respectively. During the filling stage, 19 mm of irrigation was applied on the basis of the soil water content. In the 2024 season, 18 mm of emergence irrigation was applied after sowing, followed by 36 mm of winter irrigation in early December. After the plants entered the regreening stage, 40 mm and 30 mm of irrigation were applied at the early jointing stage and heading stage, respectively, according to the seedling status and soil moisture conditions. During the filling stage, 25 mm of irrigation was applied on the basis of the soil water content.

2.3. Data Acquisition

2.3.1. Remote Sensing Data

VIs were collected via a DJI Phantom 4 Multi-spectral UAV system. It consists of 5 monochrome sensors of 12.08 megapixels and is configured on a 3-axis gimbal to obtain clear and stable images. The camera is equipped with a real-time kinematic (RTK) system to obtain images with centimeter-level positioning accuracy.

Table 2. DJI phantom 4 multi-spectral UAV system.

Parameters	Values
Weight	1487 g
Max Speed	48 km/h
Sensor	1/2.9-inch CMOS
Pixel resolution (px × px)	1600 × 1300
FOV (°)	62.7
Battery Type	LiPo 4S
Battery life (min)	27
Band set	Blue: 450 nm ± 16 nm Green: 560 nm ± 16 nm Red: 650 nm ± 16 nm Red Edge: 730 nm ± 16 nm NIR: 840 nm ± 26 nm

describes the technical specifications of the camera. To investigate the impact of different UAV flight times on the spatial distributions of the VIs, flight missions were executed at a constant altitude of 50 m. Flight dates were selected for sunny weather with wind speeds below 3.4 m/s, and take-off operations were carried out at 8:00 a.m., 11:00 a.m., 2:00 p.m., and 5:00 p.m. on 17 April, 25 April, and 12 May in the 2023 season and on 12 April, 20 April, and 9 May in the 2024 season. To study the effect of the flight altitude on the spatial distribution of VIs, flight altitudes of 50 m, 70 m, and 90 m were selected, considering the UAV’s battery endurance, and the flights were conducted at 11:00 a.m.–2:00 p.m. The spatial resolution of our UAV multi-spectral system is 2.6 cm at a flight altitude of 50 m, 3.7 cm at 70 m, and 4.8 cm at 90 m. The flight duration also decreases with increasing flight altitude. Specifically, the operation requires 20 min at 50 m, 10 min at 70 m, and 7 min at 90 m. The UAV flight parameters were set via DJI GS Pro software (SZ DJI Technology Co., Ltd., Shenzhen, China) and included a lateral overlap rate of 85% and a longitudinal overlap rate of 80%. Autonomous cruise flights were then conducted at a speed of 3.4 m/s. Three ground control points (GCPs) were placed at the experimental site for geometric correction. Radiometric calibration plates were photographed for radiometric correction. The radiation calibration board was placed on a flat ground surface in the study area, and it was ensured that there were no shadows covering the radiation calibration board. Images taken by the radiation calibration board were added to DJI Terra software (V4.1.12; DJI, Shenzhen, China) for radiometric calibration during the image mosaicking process, and reflectance images for different spectral bands were obtained.

The images taken by the UAV multi-spectral system were imported into DJI Terra software (V4.1.12) to generate spatial distribution maps of each spectral band across the experimental area. The images were subsequently imported into ArcGIS 10.8 software (Esri, Redlands, CA, USA) for the calculation and extraction of VIs. The selected vegetation indices included the normalized difference vegetation index (NDVI), soil adjusted vegetation index (SAVI), red-edge soil adjusted vegetation index (RESAVI), optimized soil adjusted vegetation index (OSAVI), difference vegetation index (DVI), green soil adjusted vegetation index (GSAVI) and enhanced vegetation index (EVI). These vegetation indices are all widely used to characterize crop growth [24,25,26,27,28,29,30]. The calculation formulas are shown in

Table 3. Selected optical indices used in this study.

Index Name	Formula	Reference
NDVI	(NIR − R)/(NIR + R)	[24]
SAVI	1.5 × (NIR − R)/(NIR + R + 0.5)	[25]
RESAVI	1.5 (NIR − RE)/(NIR + RE + 0.5)	[26]
OSAVI	(1 + 0.16) × (NIR − R)/(NIR + R + 0.16)	[27]
DVI	NIR − R	[28]
GSAVI	1.5 (NIR − G)/(NIR + G + 0.5)	[29]
EVI	2.5 × (NIR − R)/(NIR + 6R − 7.5B + 1)	[30]

G, R, RE, B and NIR represent the plant spectral reflectance of the green, red, red-edge, blue and near-infrared bands extracted from the UAV multi-spectral images, respectively.

.

2.3.2. Field Data Collection

To select the VIs closely related to crop nitrogen status, plant sampling was conducted on the same day or the next day after UAV flight to measure plant nitrogen uptake and establish relationships between VIs and plant nitrogen uptake. In the 2023 season, three sampling points were randomly selected for each treatment replication, and in the 2024 season, two sampling points were selected for each treatment replication. At each sampling site, representative plants from three rows by 0.15 m long were collected, and the position information of each sampling point was recorded. To prevent edge effects from adjacent plots with different treatments, the sampling points were located in the interior of each plot, maintaining a sufficient buffer distance from the boundaries. All the plant samples were subjected to enzymatic inactivation by oven–drying at 105 °C for 0.5 h, followed by further drying at 80 °C until a constant weight was reached for aboveground dry biomass determination. After the dried samples were ground, the total nitrogen content in both the plant tissues and the grains was separately quantified via a Kjeldahl nitrogen analyzer (Kjeltec 8400, Foss Inc., Hilleroed, Denmark) following the standard micro-Kjeldahl digestion procedure. Plant nitrogen uptake was calculated as the product of aboveground dry matter mass and plant nitrogen content.

2.4. Method of Generating VRN Prescription Maps on the Basis of VIs

To investigate the impact of flight times and altitudes on VRN prescription maps, this study proposed a method for generating VRN prescription maps on the basis of the VI. The specific steps are as follows:

(1)

Images taken by the UAV multi-spectral system are imported into DJI Terra software (V4.1.12), and spatial distribution maps of each spectral band in the experimental area are obtained.

(2)

The spatial distribution map of each band was imported into ArcGIS software, and the vegetation index closely related to crop nitrogen was subsequently calculated to generate the spatial distribution map.

(3)

According to the water application distribution characteristics of the center pivot irrigation system, a map of the management subzones was delineated. To ensure irrigation uniformity, management zones along the truss direction were designed as concentric rings centered on the central pivot, each comprising four adjacent sprinklers. Along the travel direction, the 60° control area was equally divided at 10° intervals, resulting in 90 management subzones [3].

(4)

The spatial distribution maps of the vegetation indices are superimposed on the base map of the management zones. The average vegetation index value for each subzone was calculated, bare soil areas near the center pivot were excluded, and the subzones were classified into three management zones via K-means clustering analysis [31]. The K-means algorithm is an unsupervised machine learning method that partitions observations into k clusters, where each observation belongs to the cluster with the nearest mean. The cluster center for each management zone was determined and denoted as Ki.

(5)

The nitrogen application rate was calculated on the basis of the target yield and the nitrogen uptake requirements per 100 kg of grain production were taken as the maximum nitrogen application rate (N_max, kg ha⁻¹). The topdressing ratio of nitrogen fertilizer was set as n, and the maximum topdressing nitrogen amount (N_max × n) was denoted as F (kg ha⁻¹). The topdressing nitrogen application rate N_i for each management zone was computed by the algorithm developed by Holland and Schepers [32]. To enhance field applicability, we assumed that the maximum and minimum VIs were observed in the field as sufficient nitrogen treatment and nitrogen-limited treatment, respectively. The calculation formula is as shown in Equation (1).

$$N_i=F\times \sqrt{{\displaystyle \frac{{VI}_{max}-K_i}{{VI}_{max}-{VI}_{min}}}}$$

(1)

where VI_max represents the maximum mean VI across all management subzones, and VI_min represents the minimum mean VI value across all managed subzones.

3. Results

3.1. Selection of Vegetation Indices for Nutrient Deficiency

To select VIs that are closely associated with nitrogen deficiency in winter wheat, the relationships between VIs and plant nitrogen uptake over two years were analyzed. Regression analysis was conducted by fitting several models, including linear, quadratic polynomial, exponential, power, and polynomial functions, to the data. The optimal model and its coefficients of determination (R²) and root mean square error (RMSE) are shown in

Table 4. The relationship between the vegetation index and plant nitrogen uptake of winter wheat.

Growth Stage	Vegetation Index	Optimal Model	Regression Equation	R2	RMSE
Jointing Stage	NDVI	Q	y = 4.31 × x2 − 7.02 × x + 2.93	0.49	0.022
	SAVI	Q	y = 6.84 × x2 − 4.26 × x + 0.76	0.51	0.017
	RESAVI	Q	y = 1.81 × x2 − 0.4982 × x + 0.10	0.56	0.016
	OSAVI	Q	y = 3.78 × x2 − 4.79 × x + 1.61	0.57	0.016
	DVI	Q	y = 4.38 × x2 − 2.17 × x + 0.36	0.46	0.018
	GSAVI	Q	y = 2.17 × x2 − 1.77 × x + 0.45	0.45	0.018
	EVI	Q	y = 1.24 × x2 − 1.10 × x + 0.33	0.49	0.022
Heading Stage	NDVI	P	y = 0.31 × x6.53	0.38	0.028
	SAVI	E	y = 0.0094 × e7.15x	0.57	0.024
	RESAVI	E	y = 0.0219 × e5.96x	0.63	0.021
	OSAVI	E	y = 0.0015 × e6.29x	0.61	0.022
	DVI	E	y = 0.0285 × e4.85x	0.52	0.024
	GSAVI	E	y = 0.0101 × e4.90x	0.59	0.023
	EVI	E	y = 0.0191 × e3.24x	0.53	0.024
Jointing to Heading Stage	NDVI	E	y = 0.0013 × e5.16x	0.3	0.017
	SAVI	Q	y = 9.14 × x2 − 5.75 × x + 1.00	0.63	0.014
	RESAVI	E	y = 0.0176 × e6.44x	0.65	0.013
	OSAVI	Q	y = 5.99 × x2 − 7.69 × x + 2.56	0.61	0.014
	DVI	Q	y = 3.58 × x2 − 1.55 × x + 0.25	0.60	0.013
	GSAVI	Q	y = 4.47 × x2 − 4.01 × x + 1.0	0.62	0.016
	EVI	Q	y = 1.74 × x2 − 1.61 × x + 0.46	0.61	0.013

Note: P denotes the power function; Q denotes the quadratic polynomial function; and E denotes the exponential function.

. Significant regression relationships were observed between VIs and plant nitrogen uptake across different growth stages at the 99% confidence level. The optimal regression models included quadratic polynomial, exponential, and power functions. The vegetation index that presented the highest correlation with plant nitrogen uptake varied across the jointing stage and heading stage. For example, at the jointing stage, the OSAVI had the highest R² and the lowest RMSE (R² = 0.57, RMSE = 0.016 t ha⁻¹), whereas the GSAVI had the lowest R² (0.45), and the NDVI and EVI had the highest RMSE (0.022 t ha⁻¹). At the heading stage, the RESAVI performed best (R² = 0.63, RMSE = 0.021), whereas the NDVI showed the weakest correlation (R² = 0.38, RMSE = 0.028 t ha⁻¹). To identify the optimal VI for generating VRN prescription maps during the critical fertilization period of winter wheat, regression relationships between VIs and plant nitrogen uptake throughout the entire jointing to heading period were fitted. The results indicated that the NDVI presented the lowest R² and highest RMSE (R² = 0.30, RMSE = 0.017 t ha⁻¹) due to its saturation effect, whereas RESAVI, which incorporates near-infrared and red-edge bands, achieved the highest R² (0.65) and lowest RMSE (0.013 t ha⁻¹). Under high canopy coverage, the absorption of the red band tends to saturate. At this stage, even with continued increases in crop nitrogen content and biomass, red light reflectance hardly decreases. This leads to the NDVI value no longer increasing with the improvement of crop growth [33,34]. The RESAVI incorporates the red-edge band, which maintains higher sensitivity to crop nitrogen content under high canopy coverage. In addition, the RESAVI includes a soil adjustment coefficient, which effectively mitigates interference from soil background effects. This finding further supports the conclusion of Zheng et al. [35] that red-edge and near-infrared bands are critical indicators for predicting leaf nitrogen content in winter wheat.

3.2. Spatial Distribution of the VI at Different Flight Times

To analyze the spatiotemporal variation in the VIs in the experimental area, the average RESAVI values and coefficients of variation (CVs) for winter wheat at different growth stages and flight times were calculated (

Table 5. The average RESAVI value and coefficient of variation in winter wheat with different flight times.

Year	Date (Day/Month)	Average Value				Coefficient of Variation
		8:00 a.m.	11:00 a.m.	2:00 p.m.	5:00 p.m.	8:00 p.m.	11:00 p.m.	2:00 p.m.	5:00 p.m.
2023	17 April	0.223	0.291	0.305	0.374	0.176	0.168	0.163	0.21
	25 April	0.341	0.315	0.305	0.310	0.142	0.139	0.136	0.144
	12 May	0.314	0.274	0.292	0.250	0.141	0.129	0.127	0.145
	Average	0.293	0.293	0.301	0.311	0.153	0.145	0.142	0.167
2024	12 April	0.286	0.284	0.279	0.156	0.190	0.182	0.183	0.185
	20 April	0.337	0.328	0.313	0.297	0.169	0.171	0.170	0.197
	9 May	0.307	0.285	0.300	0.307	0.171	0.167	0.170	0.181
	Average	0.310	0.299	0.297	0.253	0.177	0.173	0.174	0.188

). The RESAVI values initially tended to increase but then decreased across different growth stages of winter wheat, with the maximum RESAVI value observed during the heading stage. In the 2024 season, the three-day average RESAVI showed a decreasing trend with flight time. The minimum RESAVI value occurred at 5:00 p.m., which was a 14.5% reduction compared with that at 2:00 p.m., whereas the difference between the 11:00 a.m. and 2:00 p.m. measurements was minimal, with a relative error of less than 5%. This result is consistent with the findings of Zhang et al. [36], who reported similar effects of UAV flight time on the OSAVI characterizing water deficit. In the 2023 season, the decreasing trend of RESAVI values with delayed UAV flight times within a single day was not as obvious as that in 2024, but the difference in RESAVI values between 11:00 a.m. and 2:00 p.m. remained small, with a relative error still within 5%. The maximum relative error in the RESAVI values was 18.4% between 5:00 p.m. and 2:00 p.m. and 27% between 8:00 a.m. and 2:00 p.m. The significant variations in vegetation index values across different flight times are related to changes in the solar elevation angle [37,38,39]. The CVs for the RESAVI across different flight times ranged between 0.1 and 0.2, indicating moderate variability. This finding is consistent with the spatial variability of crop nitrogen uptake caused by different nitrogen application treatments in the experimental area.

To validate the correlation between the RESAVI values and N application rates, as well as their variation patterns at different flight times, the average RESAVI values for treatments N1–N5 under different flight times on different dates are shown in Figure 3. When flights were conducted at 8:00 a.m., 11:00 a.m., and 2:00 p.m., the RESAVI values tended to increase with increasing nitrogen application rates, demonstrating the feasibility of using the RESAVI for VRN management. This trend was more obvious in the 2024 season than in the 2023 season, potentially because of the higher basic soil fertility in the experimental area in the 2023 season. When the flight was conducted at 5:00 p.m., the response of the RESAVI to the nitrogen application rate was not obvious. A comparison of the differences in RESAVI values across different flight times revealed that the difference in RESAVI values between the 8:00 a.m. and 5:00 p.m. flights was the greatest, with instability during the growth period of winter wheat. Therefore, flight times should not be selected at 8:00 a.m. or 5:00 p.m. when generating VRN prescription maps.

3.3. The Effect of Flight Time on VRN Prescription Maps

The total fertilizer application rates calculated on the basis of the RESAVI distribution maps at different flight times are shown in

Table 6. Total fertilizer application rates at different flight times.

Year	Date (Day/Month)	Flight Time
		8:00 a.m.	11:00 a.m.	2:00 p.m.	5:00 p.m.
2023	17 April	0.61 F	0.66 F	0.65 F	0.59 F
	25 April	0.61 F	0.66 F	0.68 F	0.73 F
	12 May	0.59 F	0.73 F	0.70 F	0.59 F
	Average	0.60 F	0.68 F	0.68 F	0.64 F
2024	12 April	0.66 F	0.73 F	0.75 F	0.79 F
	20 April	0.74 F	0.75 F	0.79 F	0.79 F
	9 May	0.67 F	0.75 F	0.77 F	0.59 F
	Average	0.69 F	0.74 F	0.77 F	0.72 F

. The discrepancy in fertilizer application rates between flights at 11:00 a.m. and 2:00 p.m. was minimal. In the 2023 season, the values were both 0.68 F, and in the 2024 season they were 0.74 F and 0.77 F, respectively, with deviations within 5% across all dates. For flights at 8:00 a.m., the average fertilization rates were 11.8% and 10.4% lower than those at 2:00 p.m. in the 2023 and 2024 seasons, respectively. For flights at 5:00 p.m., the average fertilization rates were 5.9% and 6.5% lower than those at 2:00 p.m. in the 2023 and 2024 seasons, respectively.

The VRN prescription maps generated at different flight times at the jointing stage are illustrated in Figure 4. The VRN prescription maps for 11:00 a.m. and 2:00 p.m. flights exhibited similar zoning patterns, whereas disparities were observed in the zoning patterns of the maps for flights at 8:00 a.m. and 5:00 p.m. The number of subzones with high fertilizer application rates was lower than it was at 8:00 a.m. than at 11:00 a.m. and 2:00 p.m. The number of subzones with low fertilizer application for the 5:00 p.m. flight was lower than that for the 11:00 a.m. and 2:00 p.m. flights. To further quantify the differences in VRN prescription maps generated at various flight times, the overlap percentages of the same management classes are shown in

Table 7. Overlap percentage (%) of VRN prescription maps at different flight times.

Year	Date (Day/Month)	8:00 a.m.–11:00 a.m.	8:00 a.m.–2:00 p.m.	8:00 a.m.–5:00 p.m.	11:00 a.m.–2:00 p.m.	11:00 a.m.–5:00 p.m.	2:00 p.m.–5:00 p.m.
2023	17 April	57.5	47.9	53.1	77.7	73.1	66.7
	25 April	73.7	66.3	57.1	90.9	75.0	73.1
	12 May	57.0	55.5	66.9	94.0	75.4	72.9
	Average	62.7	56.6	59.0	87.5	74.5	70.9
2024	12 April	55.4	56.7	23.3	80.2	42.0	39.9
	20 April	78.5	64.5	66.9	75.8	76.5	63.5
	9 May	56.5	48.5	40.5	89.6	32.3	30.4
	Average	63.4	56.6	43.6	81.9	50.3	44.6

. The overlap percentage between prescription maps at 11:00 a.m. and 2:00 p.m. was 87.5% in the 2023 season and 81.9% in the 2024 season, with the overlap percentage exceeding 75% on all dates. The overlap percentage between the maps from the 8:00 a.m. and 2:00 p.m. flights was 56.6% in both seasons, whereas that between the 5:00 p.m. and 2:00 p.m. flights was 70.9% in the 2023 season and 44.6% in the 2024 season. The results indicated that the distribution of VRN prescription maps was relatively stable during flights at 11:00 a.m. and 2:00 p.m., whereas flights at 8:00 a.m. and 5:00 p.m. introduced deviations in the VRN prescription map.

To explore the causes of these variations, the diurnal variations in the solar elevation angles on the UAV flight dates were analyzed on the basis of latitude, longitude, and date (Figure 5). The solar elevation angle approached 0° at 5:00–6:00 a.m. and then exhibited an increasing trend, reaching its zenith near 12:00 before gradually declining to 0° by 6:00–7:00 p.m. local time. Among the four flight times in this study, the solar elevation angle reached its maximum at 11:00 a.m. local solar time, reaching 59.9°, 62.3°, and 66.7° on 9 April, 25 April, and 12 May in the 2023 season, respectively, and 58.5°, 61.0°, and 66.2° on 12 April, 20 April, and 9 May in the 2024 season, respectively. Higher solar elevation angles reduce shadow interference and maximize the illumination intensity, thereby optimizing spectral reflectance capture. At 2:00 p.m., the solar elevation angle ranged from 50.7° to 57.5°, which was slightly lower than that at 11:00 a.m. but with a difference of less than 10°, which explained the minimal difference in the VRN prescription maps between these two flight times. The 5:00 p.m. flight had the lowest solar elevation angles (17.3–22.6°), whereas the 8:00 a.m. flight presented solar elevation angles of 29.1–34.5°. Low solar elevation angles elongate interplant shadows, obscure canopy microstructures and potentially mask nitrogen deficiency signatures in leaves, leading to diagnostic inaccuracies. Furthermore, beyond this predictable systemic effect, momentary fluctuations in illumination also contributed to deviations. The temporal variation pattern of global solar radiation is largely consistent with that of the solar elevation angle, as shown in Figure 6. The relatively weak illumination intensity and directional nature of sunlight at 8:00 a.m. and 5:00 p.m. local time may lead to nonuniform light exposure across canopy segments, reducing the consistency of the spectral reflectance measurements. Fluctuations in global solar radiation were observed between 3:00 p.m. and 5:00 p.m. on 25 April, 12 May in 2023 and 12 April, 9 May in 2024, and between 6:30 a.m. and 8:00 a.m. on 20 April in 2024. These fluctuations increase data instability, which is caused primarily by intermittent cloud cover. Zhu et al. [40] demonstrated that solar irradiance fluctuates more rapidly near sunrise and sunset, significantly impacting reflectance estimation. In summary, a low solar angle introduces a consistent deviation, while passing clouds superimpose additional variability and uncertainty. While this study focused on winter wheat in the North China Plain, future work in other regions could optimize flight times on the basis of solar elevation. When UAV multi-spectral systems are used to generate VRN prescription maps, flights should be avoided when the solar elevation angle is less than 30°, preferably above 50°.

3.4. The Effect of Flight Altitude on VRN Prescription Maps

The spatial resolution of a multi-spectral camera mounted on a UAV decreases as the flight altitude increases. The spatial resolution of our UAV multi-spectral system is 2.6 cm at a flight altitude of 50 m, 3.7 cm at 70 m, and 4.8 cm at 90 m. The flight duration also decreases with increasing flight altitude. Specifically, the operation requires 20 min at 50 m, 10 min at 70 m, and 7 min at 90 m. The total fertilizer application rates calculated from VRN prescription maps generated at different flight altitudes are presented in

Table 8. Total fertilizer application rates at different flight altitudes.

Year	Date (Day/Month)	Flight Altitude (m)
		50	70	90
2023	17 April	0.65 F	0.69 F	0.69 F
	25 April	0.68 F	0.69 F	0.70 F
	12 May	0.70 F	0.71 F	0.70 F
	Average	0.68 F	0.70 F	0.70 F
2024	12 April	0.73 F	0.73 F	0.74 F
	20 April	0.74 F	0.77 F	0.77 F
	9 May	0.75 F	0.72 F	0.74 F
	Average	0.74 F	0.74 F	0.75 F

. The total fertilizer application rate did not show a discernible variation pattern as the flight altitude increased, and the differences in total fertilizer amounts across different altitudes were minimal. The mean total fertilizer application rates at flight altitudes of 50 m, 70 m, and 90 m were 0.68 F, 0.70 F, and 0.70 F, respectively, in 2023 and 0.74 F, 0.74 F, and 0.75 F, respectively, in 2024. These results demonstrate that the flight altitude had a minimal influence on the total fertilizer requirements.

Taking the jointing stage as an example, VRN prescription maps were generated at different flight altitudes (Figure 7), and the overlap percentages of the same management zones across these altitudes were analyzed (

Table 9. Overlap percentage (%) of VRN prescription maps at different flight altitudes.

Year	Date (Day/Month)	50–70 m	50–90 m	70–90 m
2023	17 April	82.6	80.3	90.3
	25 April	95.6	96.4	96.0
	12 May	89.2	84.9	93.1
	Average	89.1	87.2	93.1
2024	12 April	78.0	92.7	77.9
	20 April	76.9	84.4	92.3
	9 May	73.5	87.6	78.0
	Average	76.2	88.2	82.7

). The prescription maps showed similarities at different flight altitudes, with the average overlap percentage varying from 76.2% to 93.1%. Consistent with our findings, Mesas et al. [22] demonstrated that the NDVI values derived from flight altitudes of 60 m, 80 m, and 100 m effectively distinguished vegetation from bare soil. In summary, the flight altitude has a minor influence on VRN prescription maps. The operational efficiency increases with increasing flight altitude. Taking the control area of this center pivot system as an example, when the flight altitude is 50 m, it takes the UAV 120 min to monitor the entire area. When the flight altitude is increased to 70 m, the flight time is reduced to 60 min, a 50% reduction compared with that at 50 m. When the altitude is further increased to 90 m, the flight time is only 42 min, which is 35% of the flight time at 50 m. An increase in the flight altitude can expand the flight area. Zhu et al. [40] demonstrated that changes in solar radiation during prolonged UAV flight duration can impact reflectance estimation. Higher flight altitudes improve the operational efficiency and reduce interference from solar radiation fluctuations. Therefore, the flight altitude can be appropriately increased according to the field area and the endurance time of the UAV.

4. Discussion

This study proposed a method for generating VRN prescription maps based on vegetation index and evaluated the impact of UAV flight parameters on these prescription maps, providing technical support for precision fertilization. For flight time, the study recommends conducting UAV flights between 11:00 a.m. and 2:00 p.m. local time. During this period, the solar elevation angle is greater than 50°, ensuring a stable spatial distribution of VIs. When the solar elevation angle is less than 30°, deviations occur in the prescription maps compared to those generated at midday. The experiment was conducted in the North China Plain, and when this method is applied in other regions, the solar elevation angle can be calculated based on the local latitude, longitude, and flight date to determine the optimal flight time. For the flight altitude, the study covered a range of 50 m to 90 m, corresponding to a spatial resolution of 2.6 cm to 4.8 cm. The average overlap percentage at different flight altitudes is 86.1%. To provide a more rigorous assessment, we calculated the Kappa coefficient for the scenario with the lowest overlap percentage across different flight altitudes. The Kappa coefficient is a statistical measure of classification agreement that has been effectively applied in agricultural zoning management [41]. For the case with an overlap percentage of 76.2%, the Kappa coefficient was 0.66. According to Landis and Koch [42], this value indicates a high degree of consistency, confirming that flight operations between 50 and 90 m are feasible for practical applications. If different multi-spectral cameras are used, the flight altitude can be adjusted according to their specific parameters and target resolution, along with the field area and the UAV endurance time, optimizing monitoring efficiency while ensuring monitoring accuracy. In addition to flight time and altitude, meteorological conditions affect the quality of UAV observations. All flights in this study were conducted under clear weather with wind speeds below 3.4 m/s to minimize environmental interference. In practical applications, complex weather conditions such as cloudy weather and gusty winds are often encountered, and their impacts on VRN prescription maps require further investigation.

The basic nitrogen application rate in this method was calculated on the basis of the target yield and the nitrogen requirement per 100 kg of grain production, which met the nutrient balance requirements [23]. The spatially optimized allocation of fertilizer can reduce greenhouse gas emissions and excessive soil nitrogen residue. It is also applicable to regenerative practices. This method has been successfully applied in the VRN management of winter wheat and summer maize in the North China Plain (relevant data are not presented in this study), which verifies its application potential. Nevertheless, its applicability to other crops and under different climatic conditions still needs further validation. In addition, to enhance the adoption of this technology, especially among smallholder farmers, training can be provided to farmers, or professional services can be offered by local cooperatives. Future studies could also investigate the use of satellite data to reduce operational costs for farmers. Furthermore, with the development of emerging technologies such as artificial intelligence and the agricultural Internet of Things, machine learning algorithms and multi-source sensor data can be integrated to realize VRN management based on yield prediction, thereby further improving the accuracy of fertilization decision-making and maximizing ecological benefits.

In addition to monitoring crop nitrogen status, UAV multi-spectral systems can also detect changes in leaf structure, leaf water content, and leaf color caused by pests and diseases. These physiological alterations manifest as variations in spectral reflectance. The methodology can be further developed to generate prescription maps specifically for pests and diseases, enabling precision agronomic management practices such as precision pesticide application.

5. Conclusions

As a rapid monitoring tool for characterizing the spatial distribution of nutrient deficiencies in field crops, the UAV multi-spectral system serves as a critical technology for generating dynamic VRN prescription maps in sprinkler irrigation systems. To advance the development of VRN methodologies based on UAV multi-spectral systems, this study proposed a method for generating VRN prescription maps via a vegetation index. This study further investigated the optimal values for key operational parameters of the UAV multi-spectral system, specifically the flight time and flight altitude. The primary conclusions are as follows:

(1)

The flight time significantly influences the VRN prescription maps. The spatial distributions of the vegetation index and VRN prescription patterns showed minimal discrepancies during periods with solar elevation angles greater than 50°, which corresponded to 11:00 a.m.–2:00 p.m. local time. Under these conditions, the overlap percentage between prescription maps exceeded 80%. Our study recommended the use of a UAV multi-spectral system (DJI Phantom 4 Multi-spectral) at solar elevation angles greater than 50° to generate VRN prescription maps of winter wheat.

(2)

The flight altitude (50–90 m) of the UAV multi-spectral system had a slight effect on the total fertilizer application rates, with differences of less than 3%. The overlap percentage ranged from 76.2% to 93.1% at different flight altitudes, and the flight duration at 90 m was 35% of that at 50 m. The flight altitude can be adjusted according to the field area and the endurance time of the UAV.