Impact of Biostimulation on Floricane Raspberries Assessed Using Drone-Based Remote Sensing

Abstract

The effects of biostimulants on raspberry remain insufficiently understood, particularly in relation to cultivar-specific responses and the potential of remote sensing for treatment evaluation. The aim of this study was to assess the impact of foliar-applied biostimulants on yield and physiological responses of two floricane raspberry cultivars, Glen Ample and Przehyba, using UAV-based multispectral imaging under field conditions. Five treatment variants were tested, including a control and four biostimulant formulations based on animal-derived amino acids, plant-derived amino acids, seaweed extract, and seaweed extract combined with animal-derived amino acids. Biostimulant application significantly affected yield and fruit number per plant, whereas fruit weight remained unchanged. The highest yield values per plant regardless of cultivar were associated with treatments based on plant-derived amino acids (2578.78 g) and seaweed-containing (2483.36–2544.49 g) formulations, while the lowest values were recorded in the control (2335.66 g) and after the application of animal-derived amino acids (2351.80 g). Multispectral analysis revealed treatment-dependent temporal changes in vegetation indices, with clearer trends emerging after aggregation of relative percentage changes across measurement intervals. These results indicate that biostimulant effectiveness in floricane raspberry is strongly dependent on cultivar, formulation type, and temporal context. UAV-based multispectral imaging proved to be a promising non-destructive tool for tracking physiological responses to biostimulation under field conditions.

1. Introduction

Agricultural production constitutes the basis of global food security; however, its long-term sustainability has become a major concern across scientific, socioeconomic, and policy-related domains. The transition toward sustainable agriculture extends beyond sectoral reform and represents a critical component in achieving the United Nations Sustainable Development Goals, as well as in fostering resilient and inclusive societies. In the context of rapid population growth and intensifying climate change, the need for sustainable agricultural development has become increasingly urgent [1]. Although policy frameworks clearly emphasize sustainability, the effective implementation of integrated, science-based solutions in agriculture still requires substantial progress [2]. The shift toward sustainable agricultural systems is constrained by multiple challenges, including rising food demand, food insecurity, climate variability, biodiversity loss, and food waste [3]. The strong reliance of agriculture on synthetic fertilizers and pesticides has intensified concerns regarding its long-term sustainability. Biostimulants offer a more environmentally compatible strategy for improving crop yields [4]. Biostimulants have only recently gained increased attention in agricultural research, and their potential contribution to enhancing the sustainability of production systems remains insufficiently understood. Moreover, from an applied research perspective, there is a clear need to generate robust evidence regarding their effects across various aspects of agricultural systems [5]. Once regarded chiefly as inputs for organic agriculture, biostimulants have since gained broad application in conventional and integrated production, both in field-grown crops and under protected cultivation [6]. Biostimulants derived from natural sources are typically formulated using materials such as seaweed extracts, microorganisms, humic compounds, and protein hydrolysates [7,8]. Protein hydrolysate-based biostimulants are produced through enzymatic or chemical degradation of protein-rich plant or animal materials, yielding products composed primarily of peptides and free amino acids [9]. Seaweed positioned as the basis for ecological and sustainable food production [10], contain not only a multitude of substances, mostly organic, but also trace amounts of inorganic nutrients [11]. Of the seaweed species examined, Ascophyllum nodosum showed the most stable biostimulant performance, while effects linked to Laminaria, Gracilaria, and Kappaphycus were less consistent [12]. Growing interest in algal- and protein-derived formulations reflects their reported capacity to support crop performance across a wide range of environmental conditions, including those associated with abiotic stress [13,14]. As the range of biostimulant formulations continues to expand, reliable evaluation methods are becoming increasingly important. Their effects cannot be assumed to be uniform across plant material, since responsiveness may differ between species and cultivars. Consequently, rapid analytical approaches capable of capturing these genotype-dependent responses are needed to support meaningful assessment of product performance.

The integration of automation with advanced technologies, including artificial intelligence, drones, robotics, and data-driven approaches, supports the development of modern agriculture, enabling site-specific management and contributing to the sustainability of agricultural systems [15]. Digital agriculture depends on the effective acquisition, processing, storage, and analysis of data, as well as on the practical implementation of derived insights to support optimal decision-making in farming systems [16]. The application of remote sensing technologies has reshaped contemporary agriculture by enabling the acquisition of timely and accurate data at scale, thereby improving decision-making processes in crop production systems [17]. Despite these advantages, the implementation of remote sensing technologies in agriculture is constrained by several limitations, including restricted spatial and temporal coverage, frequent cloud interference, and variability in data quality [18]. Technologies based on spectral data currently show considerable potential for assessing the chemical composition as well as the physicochemical and antioxidant properties of seaweeds, plants, and food products, and may be applied at every stage of the production chain [19,20,21,22,23,24].

Spectral imaging is a technique that enables the acquisition and analysis of light spectra reflected, transmitted, or emitted by an object. It integrates imaging and spectroscopic methods within a single system, providing both spatial and spectral information about the investigated material. The distinction between multispectral and hyperspectral imaging is primarily determined by the number of spectral bands captured [25]. Multispectral imagery combines a limited number of spectrally distinct, spatially aligned bands, allowing the same target area to be characterized at different wavelength ranges [26]. Multispectral imaging typically involves the acquisition of images across 3 to 8 distinct spectral bands [27]. In this context, the development of efficient and scalable methods for evaluating plant responses to biostimulants becomes increasingly important. The integration of remote sensing technologies with advanced data analysis enables high-throughput, non-destructive monitoring of plant performance, offering a promising approach for assessing treatment effects under field conditions. Such approaches may help overcome the limitations of conventional experimental methods by providing detailed spatial and temporal information on crop responses.

Raspberry and blackberry cultivation is increasingly affected by climatic variability, stricter fruit-quality standards, and a narrowing range of permitted chemical inputs. Under these conditions, biostimulants and plant growth regulators may provide additional management options to support crop performance [28].

The application of UAV-based remote sensing in raspberry production remains at a relatively early stage of implementation. At the same time, biostimulant use is gaining importance in modern raspberry cultivation, with the growing availability of commercial products requiring continuous optimization of their application strategies.

Previous research by the authors evaluated foliar biostimulation and UAV-based multispectral monitoring in primocane raspberry cultivars [29]; however, whether comparable treatment-related responses can be identified in floricane cultivars remains insufficiently explored. This distinction is particularly relevant because floricane and primocane production systems differ in their fruiting cycle, which may influence both plant responses to biostimulation and their spectral expression.

In this context, the objective of this study was to determine the effects of foliar biostimulation on yield magnitude and fruit quality in floricane cultivars of Rubus idaeus L., while also evaluating the usefulness of drone-acquired multispectral data for monitoring plant responses.

2. Materials and Methods

2.1. Overview of the Research Area

The study was carried out on a raspberry plantation located in eastern Poland within the Lublin Upland macroregion (51°11′ N, 21°49′ E), established in May 2023 using plug plants. The experimental site was characterized by soil classified as class IIIa according to the Polish soil valuation system. Two floricane raspberry cultivars were evaluated (

Table 1. Raspberry cultivars.

Cultivar	Country of Origin	Breeding Program
Przehyba	Poland	Sadowniczy Zakład Doświadczalny Brzezna
Glen Ample	Scotland	James Hutton Institute

). The selection was based on their importance in commercial production systems in Poland and their suitability for both fresh consumption and processing.

Plants were cultivated in rows aligned along the east–west direction, with a spacing of 3.0 m between rows (Figure 1). Experimental treatments were assigned within rows to the respective cultivars. Cane number was standardized at four per shrub across all combinations.

Each treatment combination included 18 plants organized into three replicates of six plants each. Plants were spaced at 0.40 m within the row, corresponding to a density of 8000 plants ha⁻¹. Yield measurements were conducted at the replicate level, which served as the experimental unit. In commercial raspberry production, plants are grown in hedgerow systems that form a continuous canopy across adjacent individuals. This structural arrangement limits the possibility of isolating yield components at the level of single plants or canes without interfering with canopy integrity. Therefore, yield assessment was performed on groups of plants rather than individual units, allowing the preservation of natural canopy structure and providing results representative of field conditions. This approach ensured that treatment effects were evaluated without introducing artefacts associated with canopy manipulation. Irrigation was provided using a drip system consisting of two lines equipped with emitters delivering 1 L h⁻¹ at 20 cm spacing. This system ensured stable soil moisture conditions, particularly during periods of reduced rainfall.

Thermal conditions at the experimental site were monitored with a sensor (AGRONETPRO Sp. z o.o., Dęblin, Poland) positioned 1 m above the soil surface. During the 2025 growing season, thermal conditions showed marked variability, particularly in the early developmental stages. The winter period was characterized by alternating warm and cold phases, with frequent episodes of sub-zero temperatures, indicating a potential risk of cold stress. This period was followed by a gradual increase in temperature and moderate daily fluctuations. In contrast, the summer months were relatively stable, with only occasional exceedance of 30 °C. Overall, the season can be described as having a variable early phase followed by a more stable thermal regime during the main period of plant development, which may have influenced plant growth and phenological responses. A detailed overview of daily minimum, maximum, and mean air temperatures is presented in Figure 2, where shaded areas indicate temperature extremes and the solid line represents mean values. Threshold levels of 0 °C and 30 °C are marked with dashed lines, and monthly averages are included for reference.

Early-season frost protection consisted of covering entire rows with two layers of white non-woven fabric weighing 23 g m⁻². With further crop development, this method was replaced by sprinkler-based frost protection applied whenever sub-zero temperatures occurred.

2.2. Biostimulation

Five treatment variants were applied for each cultivar, including a control without biostimulation and four foliar biostimulant applications. The tested products represented different categories of biostimulants, namely animal-derived amino acids (NaturalCrop SL; NaturalCrop Poland Sp. z o.o., Warsaw, Poland), plant-derived amino acids (Kaishi; Sumi Agro Poland Sp. z o.o., Warsaw, Poland), seaweed extract (Valkiria Power Alg; CALFERT Sp. z o.o., Warsaw, Poland), and a combined formulation of seaweed extract and animal-derived amino acids (Phylgreen Kuma; TRADE CORPORATION INTERNATIONAL, S.A.U., Madrid, Spain). Control plants were treated with water only. The applied products differed in origin and production process. NaturalCrop SL consists of amino acids obtained from collagen through enzymatic hydrolysis, whereas Kaishi is based on plant protein hydrolysates. Valkiria Power Alg is derived from seaweed extracts, while Phylgreen Kuma combines extracts of A. nodosum L. produced using a low-temperature extraction process with animal-derived amino acids obtained via acid hydrolysis.

The treatments were applied at the maximum routine rates specified by the manufacturers, excluding doses intended for intervention use. Spray solutions were prepared on the basis of a total application volume of 500 L ha⁻¹ (

Table 2. Biostimulants.

Biostimulant Type	Trade Name	Application Dose (L/ha)
Animal-derived amino acids	NaturalCrop SL	1.5
Plant-derived amino acids	Kaishi	2.0
Seaweed extract	Valkiria Power Alg	2.0
Seaweed extract + Animal-derived amino acids	Phylgreen Kuma	3.0

).

The products selected for the experiment represented major horticultural biostimulant categories, including plant- and animal-origin protein hydrolysates and algal formulations. Products supplying considerable quantities of mineral nutrients were omitted in order to limit direct fertilization effects, particularly those containing elevated levels of K, P, Mg, Ca, S, or micronutrients commonly delivered through foliar feeding. This approach minimized the contribution of direct nutrient supply and enabled a clearer evaluation of biostimulant activity. Nevertheless, it should be acknowledged that biostimulants derived from biological raw materials may inherently contain trace amounts of mineral components and relatively high levels of organic nitrogen.

Applications were restricted to evening periods with no precipitation and low-wind conditions to reduce potential weather-related effects on treatment performance. Rainwater was used for the preparation of spray solutions. Treatments were applied using a motorized backpack mist blower (Hortmash 3WF-600, HORTMASZ Sp. z o.o., Skierniewice, Poland) to approximate application conditions typical of tractor-mounted air-blast sprayers used in fruit production. Each cultivar received four applications during the growing season. The first treatment was performed at the leaf development stage. Subsequent applications were carried out at the onset of inflorescence development, during flowering, and at the fruit development stage (Figure 3). Phenological stages were determined according to the BBCH scale for fruit crops [30]. This application schedule ensured coverage of key developmental phases of raspberry plants. During spraying, adjacent treatment units were physically separated with portable screens measuring 2.0 × 1.0 m to minimize cross-contamination caused by spray drift.

2.3. Measurements

2.3.1. Yield Parameters

The response of yield to biostimulant application was evaluated during the 2025 growing season. Harvesting was conducted at intervals of one to three days, depending on fruit ripening and prevailing weather conditions. Fruit mass was determined with an accuracy of 0.1 g using an electronic scale (Steinberg Systems SBS-LW-600, Expondo Polska sp. z o.o. sp.k, Zielona Góra, Poland). In addition, fruit number was recorded to calculate mean fruit weight.

2.3.2. Multispectral Imaging

Multispectral data acquisition was carried out during the 2025 growing season using a DJI Mavic 3 Multispectral unmanned aerial vehicle (SZ DJI Technology Co., Ltd., Shenzhen, China) (Figure 4). The platform integrates an RGB imaging system (4/3-inch CMOS, 20 MP; 5280 × 3956 pixels) with a multispectral sensor (1/2.8-inch CMOS, 5 MP; 2592 × 1944 pixels). Reflectance was recorded in four spectral regions, including near-infrared (860 ± 26 nm), red edge (730 ± 16 nm), red (650 ± 16 nm), and green (560 ± 16 nm). Accurate georeferencing was achieved using an onboard real-time kinematic positioning system, which ensured precise spatial alignment of the acquired imagery [31]. Incoming solar radiation was continuously monitored by an onboard light sensor, and the recorded values were stored as metadata. These data were subsequently used for radiometric correction during image processing, contributing to improved consistency of the multispectral dataset.

Multispectral imagery was collected under two flight scenarios designed to provide different spatial resolutions. The flights were performed at altitudes of approximately 30.4 m and 49.9 m above the take-off point, resulting in orthomosaic ground sampling distances of 1.40 cm and 2.30 cm per pixel, respectively. Identical flight conditions were maintained for both scenarios, including consistent image overlap, a fixed flight path, and speed adapted to altitude. Precise georeferencing of the imagery was ensured by real-time kinematic positioning. The imagery was georeferenced using the UAV’s onboard RTK positioning system. Automatic white balance was applied during image capture, while dewarping was disabled to preserve the original sensor geometry. Acquisition settings for both flight scenarios are summarized in

Table 3. Multispectral imaging mission parameters.

Ortho GSD (cm/pixel)	1.40	2.30
Route altitude (m)	30.4	49.9
Course angle (°)	111	111
Speed (m/s)	1.3	2
Frontal overlap ratio (%)	90	90
Side overlap ratio (%)	70	70
RTK	On	On
White balance	Auto	Auto
Dewarping	Off	Off

.

For each application cycle, multispectral imagery was collected over a four-day period beginning on the treatment date. Flights took place between 11:00 and 13:00 under relatively stable lighting conditions to limit the influence of canopy shadows. Data were processed at the level of entire experimental plots without further subdivision. Plot boundaries were identified using concrete support posts, which served as reference points during spatial data verification. The image set included RGB data together with four spectral channels representing the green, red, red-edge, and near-infrared regions for every treatment combination. Examples of the resulting imagery are shown in Figure 5.

Before every imaging mission, reflectance calibration was conducted with a MicaSense reference panel (MicaSense, Inc., Seattle, USA). The panel was photographed at two distances from the take-off level, 0.6 and 1.0 m, using the RGB camera and all four multispectral channels (G, R, RE, and NIR). Both the calibration target and the UAV irradiance sensor remained fully exposed to incident light during image capture. These measurements were subsequently incorporated into the radiometric correction workflow to improve comparability among datasets.

2.4. Analysis of Results

2.4.1. Statistical Analysis

Statistical analyses were conducted in RStudio (Posit PBC, Boston, MA, USA). Treatment effects on yield variables were examined using one-way analysis of variance (ANOVA). In addition, a two-factor analysis of variance was conducted to evaluate the main effects of cultivar and treatment, including their interaction. Prior to analysis, model assumptions were evaluated. Residual normality was tested using the Shapiro–Wilk test, while variance homogeneity was assessed with Levene’s test. The assumptions were met for all analyzed variables. Mean separation was performed using Tukey’s honestly significant difference test. Results are expressed as homogeneous groups indicated by letters, where identical letters denote the absence of significant differences. Lowercase letters correspond to comparisons within columns. F-test p-values are reported for each analysis, and statistical significance was defined at p ≤ 0.05. Yield results were presented using line and bar plots generated in PyCharm (JetBrains s.r.o., Amsterdam, The Netherlands). Details of the software environment, programming languages, and associated libraries used for data processing and visualization are summarized in

Table 4. Software.

Programming Language R (4.4.0)
Integrated development environment	Rstudio (2025.9.2.418)
Packages	Readxl (1.4.5), Dplyr (1.1.4), Emmeans (2.0.1), Multcomp (1.4.29), MultcompView (0.1.10), ComplexHeatmap (2.22.0), Circlize (0.4.17)
Programming language Python (3.12.0)
Integrated development environment	PyCharm (2025.3.2.1)
Libraries	Pandas (2.2.3), Seaborn (0.13.2), Matplotlib (3.9.2), NumPy (2.2.0)

.

2.4.2. Multispectral Data Analysis

Multispectral datasets were processed using Pix4Dfields software (Pix4D S.A., Lausanne, Switzerland; version 2.12.0) with the Accurate processing workflow. The workflow produced detailed orthomosaics and digital surface models while reducing spatial misalignment and terrain-related geometric error [32]. Radiometric calibration was based on measurements obtained from a calibrated reflectance panel and incorporated sensor calibration parameters together with illumination data, including incident solar radiation and sun position recorded by the onboard DLS sensor. GPU acceleration was applied to increase computational performance. Detailed processing settings are provided in

Table 5. Data processing settings.

Parameter	Setting
Processing pipeline	Accurate
reflectanceTargetUsed	True
enableRadiometry	True
enableGPU	True
orhtoMinGsd	0
orthoMaxSizeMPixels	0
enablePanSharpening	False
enableOrthoMinGsd	False
enableOrthoMaxSizeMPixels	False

.

A key limitation in multispectral studies of biostimulation is the variability of atmospheric conditions during data acquisition. While treatment applications are performed under controlled and optimal conditions aligned with plant developmental stages, measurement campaigns are often subject to less stable environmental conditions. For this reason, radiometric calibration is critical to ensure data comparability. Processing settings were selected according to the prevailing sky conditions at the time of image capture, distinguishing between clear and overcast acquisitions. Configuration details for each measurement campaign are provided in

Table 6. Weather conditions during measurements.

Treatment	Measurement	Processing Settings
1	1	clear sky
	2	overcast
	3	overcast
	4	overcast
2	1	clear sky
	2	clear sky
	3	clear sky
	4	overcast
3	1	clear sky
	2	overcast
	3	clear sky
	4	clear sky
4	1	clear sky
	2	overcast
	3	clear sky
	4	clear sky

.

Pix4Dfields provides multiple annotation options for multispectral image analysis. The annotation approaches were chosen to ensure efficient labeling while maintaining reliable data extraction from hedgerow raspberry canopies. Two annotation strategies were considered. In the first strategy, data extraction was limited to the inner part of the hedgerow to reduce contamination from the soil background. This restriction, however, resulted in the omission of some protruding cane tips. The second approach covered the full hedgerow width, ensuring complete inclusion of plant material at the expense of increased background signal. The initial imagery served as the basis for mask definition. These masks were then transferred to datasets of differing ground sampling distance for each acquisition date, requiring only positional correction. Due to slight misalignment of imported masks in subsequent datasets, manual repositioning was performed for each measurement. Annotations were further updated over time to reflect changes in canopy structure resulting from plant growth, cane displacement, and fruit load. Artificial adjustment of hedgerow shape was avoided because it would have altered the natural canopy structure encountered in commercial plantations, where fruit development depends on maintaining a favorable internal canopy microclimate. The applied annotation approaches are illustrated in Figure 6.

Multispectral data were interpreted using vegetation indices derived from spectral reflectance values (

Table 7. Formulas of vegetation indices.

Vegetation Index	Formula
Leaf Chlorophyll Index (LCI)	${\displaystyle \frac{\mathrm{N}\mathrm{I}\mathrm{R}-\mathrm{R}\mathrm{E}}{\mathrm{N}\mathrm{I}\mathrm{R}+\mathrm{R}}}$
Normalized Difference Red Edge (NDRE)	${\displaystyle \frac{\mathrm{N}\mathrm{I}\mathrm{R}-\mathrm{R}\mathrm{E}}{\mathrm{N}\mathrm{I}\mathrm{R}+\mathrm{R}\mathrm{E}}}$
Normalized Difference Vegetation Index (NDVI)	${\displaystyle \frac{\mathrm{N}\mathrm{I}\mathrm{R}-\mathrm{R}}{\mathrm{N}\mathrm{I}\mathrm{R}+\mathrm{R}}}$
Green Normalized Difference Vegetation Index (GNDVI)	${\displaystyle \frac{\mathrm{N}\mathrm{I}\mathrm{R}-\mathrm{G}}{\mathrm{N}\mathrm{I}\mathrm{R}+\mathrm{G}}}$
Modified Chlorophyll Absorption in Reflective Index (MCARI)	$1.2[2.5\left(\mathrm{N}\mathrm{I}\mathrm{R}-\mathrm{R}\right)-1.3\left(\mathrm{N}\mathrm{I}\mathrm{R}-\mathrm{G}\right)]$
Modified Chlorophyll Absorption in Reflective Index 2 (MCARI2)	${\displaystyle \frac{1.5[2.5\left(\mathrm{N}\mathrm{I}\mathrm{R}-\mathrm{R}\right)-1.3\left(\mathrm{N}\mathrm{I}\mathrm{R}-\mathrm{G}\right)]}{\sqrt[]{{(2\mathrm{N}\mathrm{I}\mathrm{R}+1)}^2-\left(6\mathrm{N}\mathrm{I}\mathrm{R}-5\sqrt[]{\mathrm{R}}\right)-0.5}}}$
Optimized Soil-Adjusted Vegetation Index (OSAVI)	${\displaystyle \frac{\mathrm{N}\mathrm{I}\mathrm{R}-\mathrm{R}}{\mathrm{N}\mathrm{I}\mathrm{R}+\mathrm{R}+0.16}}$
Structure Intensive Pigment Index 2 (SIPI2)	${\displaystyle \frac{\mathrm{N}\mathrm{I}\mathrm{R}-\mathrm{G}}{\mathrm{N}\mathrm{I}\mathrm{R}-\mathrm{R}}}$

). The selected indices are commonly applied in remote sensing studies and were calculated according to their standard mathematical expressions. In these formulations, NIR, R, G, and RE correspond to reflectance in the near-infrared, red, green, and red-edge spectral bands, respectively. The layers generated during multispectral data processing in Pix4Dfields are shown in Figure 7.

Each acquisition generated multiple dataset variants defined by cultivar, image resolution, and annotation approach. Their consistency was assessed from the dispersion of vegetation index values using SD and CV, pooling observations across measurement dates and treatment combinations. It should be noted that the conducted analysis did not allow for a clear determination of which ground sampling distance (GSD) and annotation method provided the most stable results in terms of SD and CV. Therefore, the final configuration was selected using secondary methodological criteria. The 1.40 cm pixel⁻¹ GSD was chosen because it provided the highest available spatial resolution, while the bounded annotation method was selected because it minimized the contribution of soil and other background elements to the extracted vegetation index values. This configuration was subsequently used for further analysis.

For the multispectral measurement series, relative percentage changes in vegetation index values between consecutive measurement dates were calculated to the following Equations (1)–(3), where m₁, m₂, m₃ and m₄ represent the vegetation index values obtained during four consecutive measurements for each treatment, respectively.

$$\mathrm{D}12\left(\mathrm{\%}\right)={\displaystyle \frac{m_2-m_1}{m_1}}\times 100$$

(1)

$$\mathrm{D}23\left(\mathrm{\%}\right)={\displaystyle \frac{m_3-m_2}{m_2}}\times 100$$

(2)

$$\mathrm{D}34\left(\mathrm{\%}\right)={\displaystyle \frac{m_4-m_3}{m_3}}\times 100$$

(3)

The parameters D12, D23, and D34 represent the relative percentage change between the first and second, second and third, and third and fourth measurements, respectively, for each biostimulant treatment. Positive and negative changes were analyzed separately and denoted as D12+, D12−, D23+, D23−, D34+, and D34−. Positive values indicate an increase in vegetation index values, whereas negative values reflect a decrease between consecutive measurement dates.

Relative percentage changes in vegetation index values between consecutive measurement dates were used to assess temporal plant responses. These changes were subsequently analyzed and visualized in two complementary ways: as detailed responses for each individual measurement interval and treatment, and as aggregated values obtained by summing changes within each measurement interval across all treatments. This approach enabled the identification of both fine-scale dynamics and general patterns of variation across treatments, cultivars, and time. In the case of the aggregation approach heatmap values represent the cumulative sum of increases or decreases across all application dates for each cultivar and treatment combination. Within each column, rankings were assigned to facilitate comparison between combinations and to highlight dominant response patterns. This approach enabled simultaneous evaluation of both the magnitude and direction of temporal changes following biostimulant application. Using sequential relative changes reduced the contribution of initial differences in vegetation index magnitude among treatments. Observed variability reflects both structural heterogeneity of the canopy, including differences in cane orientation and inclination, and physiological responses of plants to biostimulant treatments. The separation and aggregation of positive and negative changes allowed for a more integrated interpretation of temporal dynamics rather than reliance on individual measurement values.

The methodology used for the analysis of results obtained in Pix4Dfields was based on an approach described in previous studies [29,33]. Due to the large number of data combinations, defined as cultivar × treatment combination × vegetation index × GSD × annotation method, the complete results was not included in the main manuscript. Several of the analyzed vegetation indices showed similar response patterns across treatments and measurement intervals; therefore, to avoid redundant presentation of comparable results, five representative indices were selected for detailed presentation in the main text. This selection was intended to capture the principal response patterns observed across the complete dataset rather than to indicate superior performance of the selected indices. The remaining indices were retained in the complete analysis, and the full results for all analyzed configurations are available in a publicly accessible repository [34].

3. Results

3.1. Yield Parameters

Yield-related traits for Glen Ample and Przehyba are summarized in

Table 8. Yield.

Cultivar	Combination	Yield per Plant (g)	Number of Fruits per Plant (pcs.)	Fruit Weight (g)
Glen Ample	CT	2795.42 b	604.54 b	4.62 a
	AAA	2846.70 b	627.53 b	4.54 a
	PAA	3101.24 a	696.82 a	4.45 a
	SW	3022.43 ab	675.64 a	4.47 a
	SW + AAA	3123.25 a	698.00 a	4.48 a
	p-value	0.004	<0.001	0.353
Przehyba	CT	1875.89 b	336.22 b	5.58 a
	AAA	1856.89 b	335.05 b	5.54 a
	PAA	2056.32 a	374.79 a	5.51 a
	SW	1944.28 ab	340.18 b	5.72 a
	SW + AAA	1965.73 ab	358.10 ab	5.49 a
	p-value	0.018	0.002	0.289

Control (CT); Animal-derived amino acids (AAA); Plant-derived amino acids (PAA); Seaweed extract (SW); Seaweed extract + Animal-derived amino acids (SW + AAA).

. In Glen Ample, biostimulant application significantly influenced average plant yield and fruit number. Plant-derived amino acid treatment produced the greatest yield response, while the control and animal-derived amino acid treatment were associated with the lowest yield levels. The number of fruits per plant was lowest in the control and in the combination based on animal-derived amino acids, whereas the remaining combinations were characterized by significantly higher values of this parameter and did not differ significantly from one another. Fruit mass remained statistically comparable across treatments.

In Przehyba, biostimulant application significantly affected mean yield per plant and the number of fruits per plant. The highest values of both parameters were obtained in the combination based on plant-derived amino acids, whereas the lowest yield and the smallest number of fruits were recorded in the control and after the application of animal-derived amino acids. No statistically significant variation in average fruit mass was detected across treatments.

Figure 8 shows the yield per 1 ha. In Przehyba, the highest yield per hectare was obtained following the application of a biostimulant based on plant-derived amino acids, whereas the lowest yield was recorded for the combination with animal-derived amino acids, with a difference of 1595.44 kg·ha⁻¹. For Glen Ample, the highest yield was achieved in the combination with seaweed extract combined with animal-derived amino acids, while the lowest value was observed in the control, with a difference of 2622.64 kg·ha⁻¹.

Figure 9 presents yield parameters depending on the harvest date. In both cultivars, despite observable periodic differences among the combinations for all analyzed parameters, the course of changes was primarily driven by the harvest date. The temporal pattern of yield, fruit number, and mean fruit weight remained consistent across treatments, with differences between combinations mainly reflected in the magnitude rather than the overall trend of the curves.

Table 9. Main factorial effects on yield parameters.

		Yield per Plant (g)	Number of Fruits per Plant (pcs.)	Fruit Weight (g)
Cultivar (A)	Przehyba	1939.82 b	348.87 b	5.56 a
	Glen Ample	2977.81 a	660.51 a	4.51 b
	p-value	<0.001	<0.001	<0.001
Combination (B)	CT	2335.66 c	470.38 c	5.10 a
	AAA	2351.80 bc	481.29 c	5.04 a
	PAA	2578.78 a	535.80 a	4.97 a
	SW	2483.36 ab	507.91 b	5.09 a
	SW + AAA	2544.49 a	528.05 ab	4.98 a
	p-value	<0.001	<0.001	0.265
A × B	p-value	0.162	<0.001	0.362

Control (CT); Animal-derived amino acids (AAA); Plant-derived amino acids (PAA); Seaweed extract (SW); Seaweed extract + Animal-derived amino acids (SW + AAA).

shows the main factorial effects on yield parameters. Glen Ample was characterized by a significantly higher yield per plant and number of fruits per plant, whereas a significantly greater mean fruit weight was observed in Przehyba.

Regarding the effect of combinations, the highest yield was recorded for treatments based on plant-derived amino acids and seaweed extract combined with animal-derived amino acids, while the lowest yield was observed in the control. In terms of fruit number, the highest value was obtained following the application of plant-derived amino acids, whereas the lowest values were recorded in the control and in the treatment based on animal-derived amino acids. Average fruit mass remained statistically comparable across all treatments. The cultivar × treatment interaction was statistically relevant only for fruit number per plant.

3.2. Vegetation Indices

Figure 10 presents relative percentage changes in the LCI vegetation index. In Glen Ample, the highest increases in D12+ were observed in combinations based on animal-derived and plant-derived amino acids, intermediate values were recorded in the control, whereas the lowest increases occurred in combinations involving seaweed extract. In D23+, the highest increases were observed in the combination based on seaweed extract, while lower values were recorded in the remaining treatments. For D34+, clearly higher increases were observed in all biostimulant treatments compared with the control.

For negative changes, the largest decreases in D12− were recorded in combinations based on seaweed extract and plant-derived amino acids, whereas the smallest decrease was observed in the control. In D23−, no decreases were recorded. In D34−, the smallest decreases were observed in the control and in combinations based on animal-derived amino acids.

In Przehyba, similar trends were observed for all positive changes as well as for negative changes in D12− and D23−. In D34−, the smallest decreases were recorded in the control and in combinations based on animal-derived amino acids.

Figure 11 presents relative percentage changes in the NDRE vegetation index. In Glen Ample, the highest increases in D12+ were observed in combinations based on animal-derived and plant-derived amino acids, intermediate values were recorded in the control, whereas the lowest increases occurred in combinations involving seaweed extract. In D23+, the highest increases were observed in the combination based on seaweed extract, while lower values were recorded in the remaining treatments. For D34+, clearly higher increases were observed in all biostimulant treatments compared with the control. For negative changes, the largest decreases in D12− were recorded in combinations based on seaweed extract and plant-derived amino acids, whereas the smallest decrease was observed in the control, with intermediate values for animal-derived amino acids. In D23−, no decreases were recorded. In D34−, the smallest decreases were observed in the control and in combinations based on animal-derived amino acids, while clearly greater decreases were recorded for plant-derived amino acids and seaweed extract.

In Przehyba, similar trends were observed for all positive changes as well as for negative changes in D12− and D23−. In D34−, the smallest decreases were recorded in combinations based on plant-derived amino acids and seaweed extract combined with animal-derived amino acids.

Figure 12 presents relative percentage changes in the NDVI vegetation index. For Glen Ample in D12+, the highest increases were associated with combinations based on seaweed extract, intermediate values were observed for treatments based on plant-derived and animal-derived amino acids, whereas the lowest increases occurred in the control. In D23+, the highest increases were recorded for seaweed extract and plant-derived amino acids, intermediate values were observed for the combination of seaweed extract with animal-derived amino acids, while the lowest values were associated with the control and animal-derived amino acids. In D34+, the pattern shifted, with the lowest increases observed in the combination based on seaweed extract combined with animal-derived amino acids, whereas the highest values were recorded in the control.

For negative changes, in D12− the largest decrease was observed in the control, while the smallest decrease was associated with animal-derived amino acids, with intermediate values for the remaining treatments. In D23−, the most pronounced decreases were observed in combinations based on plant-derived amino acids and seaweed extract, whereas the smallest decreases were recorded for animal-derived amino acids and the control. In D34−, clearly greater decreases were observed in combinations based on plant-derived amino acids, while the remaining treatments showed comparable values.

In Przehyba, in D12+ the lowest increases were recorded in the control, with relatively similar values observed in the remaining treatments. In D23+ and D34+, the observed patterns were consistent with those recorded in Glen Ample.

For negative changes, in D12− the largest decreases were observed in the control and in the combination based on seaweed extract combined with animal-derived amino acids, whereas the smallest decreases were recorded in treatments based on plant-derived and animal-derived amino acids. In D23−, the largest decrease was associated with plant-derived amino acids, while the smallest was observed for animal-derived amino acids. In D34−, a pattern similar to that observed in Glen Ample was maintained.

Figure 13 presents relative percentage changes in the MCARI2 vegetation index. In Glen Ample, the highest increases in D12+ were observed in combinations based on seaweed extract, intermediate values were recorded for plant-derived amino acids, whereas the lowest increases were observed for animal-derived amino acids and the control. In D23+, the same pattern was maintained. In D34+, this trend was reversed, with the highest increases recorded in the control and in combinations based on animal-derived and plant-derived amino acids, while the lowest values were observed in treatments involving seaweed extract. For negative changes, no decreases were recorded in D12−. In D23−, clearly greater decreases were observed in combinations based on plant-derived amino acids and seaweed extract. In D34−, the largest decreases were recorded in the control and in combinations based on animal-derived amino acids.

In Przehyba, the highest increases in D12+ were observed in combinations based on seaweed extract and plant-derived amino acids, whereas the lowest values were recorded in the control and in combinations based on animal-derived amino acids. In D23+, clearly higher increases were observed in the control compared with biostimulant treatments. In D34+, the lowest increases were observed in combinations involving seaweed extract, while the remaining treatments showed higher values. For negative changes, the observed patterns were consistent with those recorded in Glen Ample.

Figure 14 presents relative percentage changes in the SIPI2 vegetation index. In Glen Ample, higher increases in D12+ were observed in all biostimulant treatments compared with the control. In D23+, the highest increases were recorded in the control and in the combination based on seaweed extract combined with animal-derived amino acids, intermediate values were observed for seaweed extract, whereas the lowest values were recorded for combinations based on animal-derived and plant-derived amino acids. The same pattern was observed in D34+. For negative changes, the smallest decreases in D12− were recorded in the control and in combinations based on animal-derived amino acids, whereas the largest decreases were observed in treatments involving seaweed extract. In D23−, greater decreases were observed in biostimulant treatments compared with the control. In D34−, the smallest decreases were recorded in combinations based on plant-derived and animal-derived amino acids, whereas the largest decreases were observed in treatments involving seaweed extract.

In Przehyba, the highest increases in D12+ were observed in all biostimulant treatments compared with the control. In D23+, the lowest increase was recorded for plant-derived amino acids, whereas a clearly higher increase was observed for the combination based on seaweed extract combined with animal-derived amino acids. In D34+, the lowest increases were observed in treatments involving seaweed extract, while the highest values were recorded in the control. For negative changes, greater decreases in D12− were observed in biostimulant treatments compared with the control. In D23−, the largest decrease was recorded in the combination based on seaweed extract, whereas the remaining treatments showed smaller and similar values. In D34−, the smallest decreases were observed in combinations based on plant-derived and animal-derived amino acids, while greater decreases were recorded in the remaining treatments.

4. Discussion

The present study provides evidence that the response of floricane raspberry to foliar biostimulant application is shaped by a complex interplay among formulation type, cultivar-specific traits, and temporal dynamics during the production season. Although biostimulants have been increasingly investigated in horticultural systems, including several fruit crops, their effects in raspberry remain insufficiently resolved, particularly under field conditions and at the level of cultivar-dependent responses. At the same time, the integration of unmanned aerial vehicle-based multispectral imaging into biostimulant research is still at an early stage, despite its clear potential to capture treatment-related physiological variation in a rapid and non-destructive manner. In this context, the present work contributes not only to the understanding of biostimulant performance in floricane raspberry, but also to the development of remote sensing-based approaches for tracking crop responses under practical production conditions.

A significant effect of biostimulant application on yield and fruit number per plant was observed in both Glen Ample and Przehyba. In both cultivars, the highest values of these parameters were consistently associated with biostimulant treatments based on plant-derived amino acids and containing seaweed extract, whereas the lowest values were recorded in the control and in combinations based on animal-derived amino acids. In contrast, fruit weight remained unaffected by biostimulant application across all combinations. Across both cultivars, the yield response was expressed primarily through fruit number rather than fruit weight, suggesting that biostimulant action was more closely associated with the regulation of reproductive intensity than with fruit enlargement itself. At the same time, the dominant role of harvest timing and cultivar confirms that biostimulants do not override the intrinsic developmental program of the crop, but rather modulate its expression within biologically defined limits.

The analysis of main effects showed that cultivar was the primary determinant of yield components, with Glen Ample producing higher yield and fruit number, while Przehyba exhibited greater fruit weight. Regardless of the cultivar, the highest yield was achieved for plant-derived amino acids and seaweed extract combined with animal-derived amino acids. A significant interaction between cultivar and combination was only demonstrated for fruit quantity.

Available evidence suggests that raspberry responses to biostimulant application are often favorable, although the magnitude and nature of the effect depend on cultivar, formulation, and production context. In the primocane cultivar Vajolet, foliar treatment with a biostimulant derived from the brown alga A. nodosum significantly improved yield and fruit size-related traits, while also enhancing fruit nutritional quality through increased magnesium content [35]. Positive responses have also been reported in floricane raspberry, including the cultivar Meeker and the breeding selections WSU 2188 and ORUS 4715-2, where a preparation containing A. nodosum extract and a protein hydrolysate, applied both to foliage and soil, improved physiological performance, increased total biomass, and promoted higher anthocyanin accumulation in fruit [36]. At the same time, not all reported responses have translated directly into improved productivity. In the cultivar XZ28, seaweed-based foliar biostimulants modified fruit biochemical composition, including flavonoids, anthocyanins, glutamic acid, palmitic acid, and soluble solids content, but did not significantly affect yield or growth [37]. Under stress conditions, however, a seaweed enzyme-based biostimulant applied to the cultivar XE25 increased photosynthetic rate, indicating that at least part of the biostimulant effect may be expressed through changes in physiological activity rather than through immediate agronomic responses alone [38]. Our findings were also partially corroborated by previous research on primocane raspberry cultivars, in which the effects of foliar biostimulation were shown to depend strongly on cultivar, environmental conditions, and formulation type. In that study, the most favorable responses were generally associated with treatments based on plant-derived amino acids and seaweed-containing products [29]. Other studies have also shown that the effectiveness of biostimulant application in raspberry cultivation depends on complex physiological and genotype-related mechanisms, as well as on the interaction between the plant and external environmental conditions. In addition, the importance of adaptive strategies based on modern agronomic approaches tailored to climatic conditions in raspberry production has been emphasized [39].

A positive effect of using biostimulants containing amino acids and seaweed extract on the yield was also noted in the case of fruit species such as apples [40,41,42], strawberries [43,44,45], blueberries [46,47,48], vines [49,50,51], bananas [52,53], oranges [54,55] and cherries [56,57,58].

Taken together, the yield data suggest that biostimulant action in raspberry is expressed less as a universal productivity effect than as a context-dependent reorganization of plant performance, shaped by genotype and developmental timing. From this perspective, harvest outcomes may capture only the terminal manifestation of a broader cascade of treatment-induced physiological responses. It is therefore reasonable to expect that part of this response may become detectable earlier and more sensitively at the spectral level, which provides the rationale for examining vegetation index dynamics in greater detail.

Consistent with this interpretation, the effects of foliar biostimulant application were also expressed in the relative percentage changes observed between consecutive multispectral measurements. A detailed analysis of relative percentage changes revealed considerable variability in responses, influenced not only by cultivar, vegetation index and combination, but also by specific time intervals and individual measurement points. However, an analytical approach based on the aggregation of relative percentage changes separately for positive and negative values across all treatments within each measurement interval revealed more distinct trends. The obtained results indicate patterns that were particularly evident in positive changes. However, these patterns were strongly dependent on the vegetation index and, despite partially similar trends, also on the cultivar. In most cases, the combinations were grouped into two or three main recurring response patterns. The first group generally included the control and animal-derived amino acids, while the second comprised plant-derived amino acids and biostimulants based on seaweed extract, including those combined with animal-derived amino acids.

When one group exhibited higher values in the initial measurement intervals, this trend was typically reversed in the final interval, where the same group showed the lowest values. In contrast, for negative changes, the patterns were less pronounced and considerably less structured. This may be explained by the greater sensitivity of decreases in index values to short-term environmental fluctuations, whereas increases are more closely linked to coordinated physiological responses induced by biostimulant application [29]. The recurrence of similar treatment-related response patterns across indices suggests that biostimulant responses may converge at the level of coordinated physiological regulation, even when individual measurement points remain highly variable. In this sense, the observed spectral patterns can be viewed as the canopy-scale expression of treatment-induced changes in plant physiological status.

These patterns can be interpreted as the optical manifestation of underlying physiological changes induced by biostimulant application. Biostimulants are known to affect key processes such as chlorophyll synthesis, photosynthetic activity and pigment balance. As plant responses alter the multidirectional interaction of tissues with electromagnetic radiation, including its absorption, reflection, and scattering, corresponding shifts may be expressed in vegetation index values. Because vegetation indices are calculated from combinations of these spectral bands (green, red, red-edge, and near-infrared), even subtle physiological and structural changes at the leaf level could potentially contribute to detectable differences at the canopy scale. Consequently, the observed variability and emerging patterns in vegetation indices likely reflect dynamic, treatment-dependent shifts in photosynthetic efficiency and canopy structure, which are indirectly captured through changes in spectral reflectance. However, without direct physiological measurements, these spectral responses cannot be unequivocally attributed to specific processes.

Research on the use of multispectral imaging as a tool to support the analysis of the effects of biostimulation on plants is still very limited. However, the results obtained indicate the enormous potential of this technology in this regard. This is especially important considering that spectral imaging is not limited to UAV platforms, but can also be conducted using ground-based and satellite systems. Furthermore, hyperspectral imaging may provide additional advantages due to the greater number of recorded bands. This may be particularly important in the context of the rapidly expanding biostimulant market, which demands more efficient and faster approaches for evaluating their effects on plant systems.

Previous findings further indicate that multispectral responses in raspberry reflect not only physiological activity at the level of individual canes, but also broader changes in canopy structure and seasonal growth dynamics. Changes in cane number were associated with systematic shifts in index values, spatial heterogeneity, and temporal variability, suggesting that remote sensing can capture integrated structural and functional responses at the plantation scale [59].

The potential of UAV-based multispectral imaging for assessing biostimulant effects has also been demonstrated in other crops, including soybean, where significant treatment-related differences were detected in indices such as NDRE, LCI, and OSAVI [60]. However, its sensitivity may be context-dependent, as studies in grapevine showed that although NDVI and NDRE reflected general treatment patterns, they were less effective in distinguishing between individual biostimulation application rates [61]. Particularly noteworthy is the potential of spectral data-based technologies to become an important tool in biostimulant research, as evidenced by their already widespread application in agrotechnical treatments. Numerous studies indicate that spectral remote sensing technologies can support the spatial identification and characterization of plants within agricultural systems, with applications in plant-specific nutrient management as well as pest and disease monitoring. In this context, high-resolution remote sensing systems mounted on UAVs are increasingly being used for large-scale crop mapping and the characterization of within-field variability [62,63,64,65,66,67]. One of the most developed areas of spectral imaging application is precision nutrient management, where numerous studies have demonstrated their considerable potential for optimizing nitrogen fertilization in fruit crops [68,69,70,71,72]. The successful application of spectral imaging in nitrogen monitoring therefore points to considerable potential for its use in evaluating organic nitrogen-containing biostimulants, especially amino acid- and protein hydrolysate-based formulations. This approach may help determine whether observed plant responses are associated with changes in nitrogen status, broader biostimulatory activity, or an interaction between these mechanisms.

An important aspect is also the further development of remote sensing data analysis, in which vegetation indices, despite their concise mathematical form and clear interpretation, represent a powerful tool for vegetation monitoring [73].

Moreover, realizing the full potential of biostimulants requires coordinated efforts across the sector, including robust and transparent scientific research, targeted education for growers and agronomists, and the development of regulatory frameworks that classify products according to their biological function [74]. Although biostimulants have considerable potential to support the development of sustainable agriculture by promoting plant growth, improving nutrient use efficiency, and enhancing tolerance to environmental stresses. However, their broader adoption remains limited due to challenges such as variability in product formulations, inconsistencies in regulatory frameworks, and an incomplete understanding of their mechanisms of action [75].

The present study has several limitations that should be considered when interpreting the results. The experiment included only two floricane raspberry cultivars, which restricts the broader applicability of the findings to other genotypes and production contexts. Moreover, the multispectral indices used in this work provided only indirect information on plant physiological status and were not supported by direct physiological measurements such as gas exchange, chlorophyll fluorescence, nutrient analysis, or direct chlorophyll determination. The analytical framework was further limited to UAV-based multispectral sensing and to biostimulants based on amino acids and seaweed extracts, without comparison to hyperspectral, ground-based, or satellite approaches, or to other categories of biostimulant products. Finally, the study did not address fruit quality, health-related composition, sensory traits, or economic performance, which remain important areas for future investigation.

The future success of biostimulant research, particularly for commercial products, will depend fundamentally on greater transparency in formulation composition. Such transparency would enable a more mechanistic understanding of product action and support the rational matching of formulations to specific agronomic objectives, climatic contexts, crop species, and even cultivars. In the longer term, this could open the way to crop-specific and potentially cultivar-tailored biostimulant programs. Equally important will be the continued development not only of remote sensing technologies themselves, but also of the analytical approaches required to interpret their outputs, so that their full scientific and agronomic potential can be realized. This challenge is especially significant because spectral signals measured in crop canopies ultimately arise from one of the most complex processes in biology, namely photosynthesis, together with its associated pigment dynamics, structural adjustments, and water relations, all of which shape the interaction of plant tissues with electromagnetic radiation. In this sense, progress in spectral phenotyping will depend not only on improvements in sensor performance, but also on deeper physical understanding of light–matter interactions and on more robust methods for translating these signals into biologically meaningful information. In the longer perspective, advances in the physical understanding of electromagnetic signal formation, including increasingly refined descriptions of light–matter interactions across scales, may substantially improve the interpretation of plant spectral responses. Such progress will likely require not only the refinement of existing analytical workflows, but also the development of new vegetation indices and data integration strategies tailored to specific agronomic interventions and plant responses. In turn, this shift could improve the precision of product assessment and accelerate the development of crop-specific, context-adapted biostimulant strategies for sustainable agriculture.

5. Conclusions

The present study showed that biostimulant effects in floricane raspberry were dependent on cultivar and treatment formulation, while UAV-based multispectral imaging captured temporal variation in canopy spectral responses following biostimulant application. Although the multispectral component remains exploratory, the observed temporal patterns demonstrate the value of extending biostimulant assessment beyond endpoint classical physiological and agronomic measurements.

Beyond the immediate scope of raspberry cultivation, these findings point to a broader conceptual shift in how biostimulants may be studied and ultimately deployed. Rather than being evaluated solely through endpoint agronomic traits, their action may increasingly be understood as a temporally structured physiological process that can be tracked through non-destructive optical signals. This perspective is particularly relevant in the context of the rapidly expanding biostimulant market, where faster, more mechanistically informed, and more scalable methods of assessment are urgently needed. The results therefore suggest that multispectral sensing has the potential to become more than a supplementary analytical tool. When combined with robust physiological interpretation and advanced data integration, it may provide the basis for a new generation of biostimulant research, linking crop performance with real-time functional plant status. Under such a framework, biostimulants should no longer be viewed simply as auxiliary inputs, but as responsive components of data-driven crop management systems designed to improve resilience, efficiency, and sustainability under increasingly complex production conditions.