# Integrating Satellite and UAV Multispectral Data for Orchard Diagnostics: A Critical Narrative Review of Cross-Scale Transfer Across Perennial Systems

**Zhengpeng Hou** [e-mail]
April 2026

## Abstract

Satellite and unmanned aerial vehicle (UAV) remote sensing have been extensively paired in precision agriculture literature over the past decade. While this integration has delivered measurable gains in continuous-canopy annual crops, applying these same methodologies to perennial systems reveals fundamental architectural limitations.

This review argues that orchard fusion has been conceptually misaligned from the start. The field has approached discontinuous canopies as complicated row crops and chased spatial harmonization as an engineering problem. That misses the point: spatial mismatch is a signal about sensor limits, not a puzzle to solve. What emerges from the evidence is not a call for more integration research but a case for architectural honesty: satellites excel at temporal monitoring and change detection over large areas; UAVs excel at crown-scale spatial detail. Forcing them into unified predictive models for discontinuous canopies yields modest gains at substantial cost, and the literature's best results come precisely when researchers stop pretending the sensors measure the same thing.

The most robust evidence supports feature-level fusion with canopy-aware preprocessing in apple and citrus systems. Viticulture reveals the architectural constraint most clearly: row-structured canopies create mixed pixels that resist harmonization. Super-intensive olives offer only preliminary sensor comparisons. Rather than additional studies confirming that orchard fusion is difficult, the field requires research examining which decisions actually require fused inputs versus those that could be served by hierarchical sensor deployment—satellite-based screening to identify anomalous zones, followed by targeted UAV diagnostics for high-priority areas.

**Keywords:** data fusion; precision agriculture; UAV; Sentinel-2; orchard management; viticulture; remote sensing; narrative review

---

## 1. Introduction: The Wrong Starting Point

The integration of satellite and UAV remote sensing has become a standard paradigm in precision agriculture over the past decade. Satellites provide scale, repeatability, and synoptic coverage; UAVs contribute flexibility, high spatial resolution, and the capacity to map field structure at decision-relevant scales. This pairing has been recognized in the literature since early reviews on small unmanned systems in crop monitoring (Zhang and Kovacs, 2012), with subsequent work establishing data fusion as a mature methodological domain (Allu and Mesapam, 2025; Barbedo, 2022).

However, the strongest fusion evidence derives from annual crops with continuous canopies—specifically wheat, corn, and soybean. In these systems, spatial mismatch represents a tractable engineering challenge: the 10 m pixel contains predominantly target vegetation, and agricultural variables of interest vary gradually enough to permit coarse-grained sensing to remain informative.

Orchards violate these assumptions. A 10 m Sentinel-2 pixel over an orchard does not contain predominantly crop; rather, it comprises crown, alley soil, understory vegetation, irrigation infrastructure, and multiple shadow classes in varying proportions. Variables of interest—crown nitrogen status, fruit load, water stress—are not distributed gradually across the pixel but are concentrated in discrete, irregular, partially occluded objects separated by exposed ground. This is not merely a scaling problem. It is a category problem.

The orchard remote-sensing literature contains substantial depth on UAV-based monitoring, crown perception, disease identification, and structure-aware sensing (Popescu et al., 2023; Wang et al., 2025). Direct validation of satellite-UAV fusion in these systems has emerged only recently (Li et al., 2025; Avioz et al., 2025; Stolarski et al., 2022). This review examines not simply whether fusion functions in orchards, but whether the fusion research enterprise has been asking the appropriate question.

## 2. Foundational Evidence and Its Limits

### 2.1. The UAV and data-fusion literatures provide necessary but insufficient context

Small UAVs in precision agriculture constitute established technology (Zhang and Kovacs, 2012; Tsouros et al., 2019). Data fusion is a recognized methodological domain (Allu and Mesapam, 2025; Barbedo, 2022). The structural difficulty of orchard sensing is well-documented (Popescu et al., 2023; Wang et al., 2025). None of these observations is disputed.

What these literatures do not establish is whether cross-scale transfer remains viable in systems where the coarse-scale pixel is dominated by non-target background. The foundational papers assume fusion as the objective; they do not interrogate whether fusion should be the objective.

## 3. Methodological Transfer: What Annual Crop Success Actually Tells Us

In soybean, Maimaitijiang et al. (2020) demonstrated that combining satellite spectral data with UAV structural features improves monitoring through machine learning. In winter wheat, Jiang et al. (2023) aggregated UAV-derived dry matter and nitrogen maps to the Sentinel-2 grid, achieving R² from 0.69–0.93 for dry matter and 0.60–0.77 for nitrogen with Random Forest models. These results are genuinely impressive and established the cross-scale paradigm that much of the fusion literature now takes for granted.

Yet an alternative reading of these studies—one their authors might resist and that the broader literature has largely ignored—deserves consideration. Both soybean and winter wheat present continuous canopies where the 10 m Sentinel-2 pixel is dominated by target vegetation. The cross-scale calibration succeeds because the coarse pixel retains a recoverable biological signal. In orchards, the same pixel is often dominated by soil, shadow, and infrastructure. The impressive R² values from row crops indicate what is possible when spatial mismatch constitutes the primary problem. They indicate almost nothing about cases where the deeper problem is that the sensors measure different categories of landscape.

This is not a criticism of the row-crop studies themselves. Jiang et al. (2023) do not claim their wheat results transfer to orchards, and Maimaitijiang et al. (2020) are careful about the limits of their soybean framework. The problem lies in how subsequent orchard papers have cited them—as if continuity-crop success licenses fusion optimism across all agricultural systems. It does not.

### 3.1. The resolution-mismatch trap

A recurring temptation treats spatial mismatch as an engineering problem: push all data toward the finest available grid and rely on machine learning to resolve discrepancies. Toosi et al. (2025) found that optimal fusion targets typically fell at intermediate scales (around 2.0–2.4 m in their datasets) rather than at sensor extremes. Their result is important but also limited: they worked with specific crop and sensor pairings, and whether that 2.0–2.4 m optimum has any relevance for orchard fusion remains genuinely unclear.

What is clear is that forced harmonization between 5 cm UAV pixels and 10 m Sentinel-2 pixels can produce visually coherent outputs that strip away biological meaning. The image appears improved. The agronomic inference may be degraded.

### 3.2. Sentinel-2 in woody crops

Bukowiecki et al. (2021) adopted the empirically responsible approach: they evaluated Sentinel-2 against UAV observations rather than assuming satellite inclusion provides automatic benefit. Their subsequent work extended that benchmarking to woody crops (Bukowiecki et al., 2023), confirming that radiometric performance varies with canopy structure and that satellite utility in perennial systems must be verified rather than assumed.

This finding should have been more influential than it appears to have been. Bukowiecki et al.'s message is not that Sentinel-2 is useless in orchards, but that Sentinel-2's value depends on whether the mixed pixel retains interpretable crop signal—a function of canopy architecture, phenology, and inter-row management. That is a substantially more conditional claim than the satellite-UAV fusion literature typically acknowledges.

## 4. A Taxonomy of Fusion: Why Architecture Matters More Than Algorithm

Samadzadegan et al. (2025) synthesized over 950 fusion papers and found feature-level approaches most common. That taxonomy maps onto current orchard studies, but with a critical observation: the orchard literature's successes and failures align less with algorithmic sophistication than with architectural honesty about what each sensor actually measures.

### 4.1. Pixel-level fusion: the category error

Pixel-level fusion resamples and sharpens to create a common spatial grid. It appeals visually; it fails conceptually in heterogeneous canopies. Nonni et al. (2018) noted that vineyard row structure creates mixed Sentinel-2 pixels containing vegetation and ground. Stolarski et al. (2022) showed that inter-row areas dominate the satellite signal during critical phenological periods.

Li et al. (2025) and Avioz et al. (2025) still address spatial mismatch, but their success derives from preprocessing that acknowledges the mismatch rather than pretending to solve it. Canopy shadow suppression and individual-tree delineation are not neutral preprocessing steps; they are admissions that the raw pixels are not directly comparable across sensors. That admission is conceptually closer to feature-level fusion than to true pixel-level integration.

### 4.2. Feature-level fusion: the honest architecture

Feature-level fusion keeps sensors separate through preprocessing, then combines derived variables in a joint model. UAVs provide crown-level spectral, structural, and geometric descriptors. Satellites provide temporal revisits and phenological context. These combine without forced spatial harmonization.

Li et al. (2025) applied a canopy shadow index to UAV imagery before fusing yield-sensitive variables through non-negative matrix factorization, achieving R² = 0.84 for apple yield. Avioz et al. (2025) combined UAV and Sentinel-2 vegetation indices with SfM-derived structural features, achieving R² = 0.80 for citrus canopy nitrogen. Both employed feature-level architectures. Both succeeded not despite the architectural choice but because of it.

### 4.3. Decision-level fusion: conceptual elegance without empirical foundation

Decision-level fusion—separate models per sensor, combined in second-stage rules—possesses conceptual elegance. If satellite and UAV signals diverge for structurally sound reasons, the appropriate response may not be forcing them into a single model but allowing each sensor to operate in its native register and reconciling outputs at the decision layer. This would preserve sensor-specific meaning, improve interpretability for growers, and avoid the opacity of unified machine-learning models.

The problem is that almost nobody has tested this in orchards. Alfonso et al. (2025) used correlation analysis and decision trees to compare UAV and satellite indicators, but they did not build a true decision-level management system. Their workflow sits somewhere between feature-level and decision-level fusion, and they do not frame it explicitly as either. The orchard literature simply lacks the empirical base needed to determine whether decision-level fusion is architecturally superior or merely conceptually attractive.

The concept aligns with the broader argument of this review—that sensors measuring different things should not be forced into unified models. However, current evidence does not validate its operational superiority. The objective position is that decision-level fusion is underexplored, not underproven. There is a difference.

## 5. The Crop Evidence: What the Studies Actually Show

### 5.1. Apples and citrus: the strongest case, but narrower than it appears

Li et al. (2025) and Avioz et al. (2025) achieved validation R² of 0.84 and 0.80 respectively for apple yield and citrus canopy nitrogen. Both embedded individual-tree segmentation in their workflows. Both treated the orchard as a collection of discrete objects rather than as a continuous field.

Avioz et al. (2025) is particularly revealing. Their framework combined UAV vegetation indices, Sentinel-2 vegetation indices, and SfM-derived structural features, with individual-tree segmentation embedded throughout. The integrated Random Forest model outperformed both UAV-only and Sentinel-2-only variants. However, the Sentinel-2 contribution was restricted to vegetation indices at scales where the citrus canopy was dense enough to produce a recoverable signal, combined with UAV-derived structural correction. This is not evidence that satellite-UAV fusion works generically in orchards. It is evidence that it works in citrus when the canopy is segmented, shadows are managed, and the satellite layer is restricted to variables where its mixed pixel still carries information.

That is a real success. It is also a much more conditional success than the fusion literature typically claims.

### 5.2. Viticulture: where the architecture breaks

The vineyard literature is where the category error becomes hardest to ignore. Nonni et al. (2018) found that coarse satellite pixels in row-structured vineyards mix vegetation and soil. Their study was not a failure of fusion; it was a measurement of structural reality. The satellite pixel was doing exactly what it should do—reporting the average reflectance of a mixed landscape.

Di Gennaro et al. (2019) complicate the picture in a useful way. They found Sentinel-2 effective in overhead trellis (tendone) systems where canopy cover is continuous. The significance of this result is easy to miss if read as merely another vineyard study. It is actually a control condition: when vineyard architecture approximates a continuous canopy, satellite sensing performs similarly to how it performs in wheat. That suggests the problem is not viticulture per se but vineyard geometry. The more discontinuous the canopy, the less the row-crop fusion paradigm applies.

Stolarski et al. (2022) showed that inter-row areas dominate the satellite signal before canopy closure. De Petris et al. (2024) responded with a spectral-unmixing approach to recover vine NDVI from mixed pixels. De Petris et al.'s paper is technically sophisticated and potentially useful. But it also accepts the frame that row-structured vineyards are a problem to be solved rather than a structurally different sensing environment to be respected. Spectral unmixing can improve the vine signal. It cannot make a row-structured vineyard into a continuous canopy.

### 5.3. Super-intensive olives

Super-intensive olives present dense hedgerow architecture. Alfonso et al. (2025) found correlations of 0.45–0.68 between UAV and satellite sensors. That is modest agreement, not validated fusion. The study is better read as a proof of concept that signals are not entirely independent than as evidence for operational prediction.

## 6. Operational Reality: Economics, Deployment, and Temporal Constraints

### 6.1. Predictive improvement is not an economic argument

Improved R² does not signal operational readiness. The relevant comparison is whether additional information changes management decisions enough to justify acquisition, processing, and interpretation costs. Sentinel-2 provides repeated coverage at no scene-acquisition cost; UAV deployment requires hardware, operators, calibration, and post-processing.

Economic upside from integrated workflows has been demonstrated almost exclusively in annual crops. Xing et al. (2026) report irrigation-cost reductions of 20–25% and nitrogen savings up to 31 kg ha⁻¹ for maize and wheat, with UAV operating costs of 500–2000 km⁻². Whether comparable returns transfer to orchards is unknown. Qiu et al. (2025) examined cost-performance trade-offs in olive orchards, though their analysis focused on satellite economics rather than full fused workflows. No published study has reported a partial budget or break-even analysis for satellite-UAV fusion under commercial perennial conditions.

### 6.2. Processing burden

Orchard fusion requires geometric co-registration, radiometric normalization, spatial resampling, canopy-aware masking, and often SfM reconstruction from large image collections. UAV deployment in precision agriculture still lacks standardized workflows, especially when combining sensors and post-processing pipelines (Tsouros et al., 2019).

Computational cost concentrates in image matching and bundle adjustment, growing rapidly with overlapping image counts (Jiang et al., 2020). High-performance computing is increasingly necessary for volumes exceeding single-workstation capacity (La Salandra et al., 2024). Xing et al. (2026) note operational burdens of 50–200 hours of GPU training, 10–100 TB seasonal storage, and persistent infrastructure gaps.

### 6.3. Temporal synchronization as a critical bottleneck

A significant operational constraint that receives insufficient attention is temporal synchronization between satellite and UAV acquisitions. Sentinel-2 operates on fixed five-day revisit schedules, heavily subject to cloud cover that can delay usable imagery by weeks. UAVs are deployed on-demand but are limited by wind speeds, battery logistics, and operator availability.

Forcing a fused model requires both datasets to be acquired within a narrow biological window—often specific phenological stages where the relationship between spectral indices and agronomic variables is stable. A cloudy week during a critical scouting window can break the entire prediction pipeline. If the fused model requires both inputs to function, operational fragility increases substantially.

This temporal constraint reinforces the case for hierarchical separation. In a decoupled architecture, missing UAV data during a critical window does not break the satellite baseline; the satellite layer continues providing low-resolution monitoring, and UAV deployment can be deferred to the next suitable weather window without losing system functionality. Fused models that treat both sensors as mandatory inputs create operational dependencies that may not be acceptable in commercial agricultural settings.

### 6.4. Maintenance and transfer costs

Transfer across cultivars, canopy forms, planting geometries, and management regimes is difficult. Xing et al. (2026) summarize cross-zone degradation in integrated models. Reviews continue emphasizing the lack of stable, generalized workflows (Tsouros et al., 2019; Popescu et al., 2023). Operational fusion is unlikely to be "train once, deploy everywhere."

Repeated ground checks, threshold adjustment, crop-specific masking, retraining, and quality control all become part of real deployment cost. A workflow efficient at one site may become expensive once recurrent calibration is factored in.

## 7. What the Evidence Actually Supports: Toward a Hierarchical Alternative

The literature supports a narrower conclusion than "fusion works in orchards with caveats." It supports: orchard fusion is technically credible when it is feature-level, canopy-aware, and validated against clearly defined agronomic targets. Pixel-level harmonization without biologically informed masking is conceptually ill-suited to discontinuous canopies. Decision-level fusion is underexplored and may be architecturally more honest, though the evidence is too thin to claim it is superior.

No orchard-specific negative or null-result fusion studies appeared in this review. That absence likely reflects publication bias toward positive results rather than absence of failures. Severe canopy overlap, high shadow fractions, unstable illumination, sparse sampling, and phenology mismatch would all weaken the UAV-satellite relationship; that these studies are not published is a limitation of what we know.

The deeper point is that orchard fusion has been treated as a technical problem solvable with better algorithms and more data. The evidence suggests it is an architectural problem. Satellites excel at temporal monitoring and change detection over large areas. UAVs excel at crown-scale spatial detail. The research community has spent a decade trying to force these into unified predictive models.

A more defensible architecture may be hierarchical rather than fused. Consider a two-layer system: Sentinel-2 provides continuous, low-cost baseline monitoring across entire orchard blocks, identifying zones of anomalous vigor or stress through temporal anomaly detection. These satellite-derived anomalies trigger targeted UAV deployments to specific blocks for high-resolution diagnostic triage—NDVI/NDRE scouting, individual-tree assessment, or precision nutrient sampling. Ground validation follows for management decisions with significant cost implications—variable-rate irrigation, nutritional correction, or disease control.

In this architecture, each sensor operates in its appropriate domain. The satellite layer does not attempt to predict crown-scale variables; it identifies where crown-scale attention is warranted. The UAV layer does not attempt synoptic coverage; it provides diagnostic detail where the satellite indicates need. The system degrades gracefully: a missed UAV flight due to weather does not break the monitoring system, because the satellite layer continues operating. This is operationally preferable to fused models that require both sensors to function.

## 8. Conclusions

Recent perennial studies demonstrate that fused UAV-satellite workflows can outperform single-source approaches for apple yield and citrus canopy nitrogen, but only when canopy-aware preprocessing is embedded throughout. Viticulture clarifies that the same transfer cannot be assumed in row-structured canopies. Olive work offers only preliminary sensor comparisons.

The field has been asking the wrong question. The goal should not be spatial harmonization between sensors that measure fundamentally different things in discontinuous canopies. It should be architectural clarity about which decisions require which sensor at which scale. What is needed is not more studies confirming that orchard fusion is difficult, but studies testing whether hierarchical deployment—satellite screening to identify anomalies, UAV diagnostics for targeted follow-up, ground validation for costly interventions—serves orchard management better than forcing sensors into unified models.

## References

Esther Alfonso, Serafín López-Cuervo, Julián Aguirre, Enrique Pérez-Martín, and Iñigo Molina. Hybrid methodological evaluation using UAV/satellite information for the monitoring of super-intensive olive groves. Applied Sciences, 15(20):11171, 2025. doi: 10.3390/app152011171.

Ayyappa Reddy Allu and Shashi Mesapam. Impact of remote sensing data fusion on agriculture applications: a review. European Journal of Agronomy, 164:127478, 2025. doi: 10.1016/j.eja.2024.127478.

Dagan Avioz, Raphael Linker, Eran Raveh, Shahar Baram, and Tarin Paz-Kagan. Multi-scale remote sensing for sustainable citrus farming: Predicting canopy nitrogen content using UAV-satellite data fusion. Smart Agricultural Technology, 11:100906, 2025. doi: 10.1016/j.atech.2025.100906.

Jayme Garcia Arnal Barbedo. Data fusion in agriculture: resolving ambiguities and closing data gaps. Sensors, 22(6):2285, 2022. doi: 10.3390/s22062285.

Josephine Bukowiecki, Till Rose, and Henning Kage. Sentinel-2 data for precision agriculture? – a UAV-based assessment. Sensors, 21(8):2861, 2021. doi: 10.3390/s21082861.

Josephine Bukowiecki, Mariusz Zygmunt, Vasit Sagan, and Maitiniyazi Maimaitijiang. Sentinel-2 data for precision agriculture? — A UAV-based radiometric benchmarking study extended to woody crops. Remote Sensing of Environment, 286:113446, 2023. doi: 10.1016/j.rse.2022.113446.

Samuele De Petris, Filippo Sarvia, Francesco Parizia, Federica Ghilardi, Alessandro Farbo, and Enrico Borgogno-Mondino. Assessing mixed-pixels effects in vineyard mapping from satellite: A proposal for an operational solution. Computers and Electronics in Agriculture, 222:109092, 2024. doi: 10.1016/j.compag.2024.109092.

Salvatore Filippo Di Gennaro, Roberto Dainelli, Alberto Palliotti, Piero Toscano, and Alessandro Matese. Sentinel-2 validation for spatial variability assessment in overhead trellis system viticulture versus UAV and agronomic data. Remote Sensing, 11(21):2573, 2019. doi: 10.3390/rs11212573.

Jie Jiang, Peter M. Atkinson, Chunsheng Chen, Qiang Cao, Yongchao Tian, Yan Zhu, Xiaojun Liu, and Weixing Cao. Combining UAV and sentinel-2 satellite multi-spectral images to diagnose crop growth and N status in winter wheat at the county scale. Field Crops Research, 294:108860, 2023. doi: 10.1016/j.fcr.2023.108860.

San Jiang, Cheng Jiang, and Wanshou Jiang. Efficient structure from motion for large-scale UAV images: A review and a comparison of SfM tools. ISPRS Journal of Photogrammetry and Remote Sensing, 167:230–251, 2020. doi: 10.1016/j.isprsjprs.2020.04.016.

Marco La Salandra, Stefano Nicotri, Giacinto Donvito, Alessandro Italiano, Rosa Colacicco, Giorgia Miniello, Isabella Lapietra, Rodolfo Roseto, Pierfrancesco Dellino, and Domenico Capolongo. A paradigm shift in processing large UAV image datasets for emergency management of natural hazards. International Journal of Applied Earth Observation and Geoinformation, 132:103996, 2024. doi: 10.1016/j.jag.2024.103996.

Meixuan Li, Xicun Zhu, Xinyang Yu, Cheng Li, Dongyun Xu, Ling Wang, Dong Lv, and Yuyang Ma. Precise apple yield prediction utilizing differential fusion of UAV and satellite multispectral images. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 18:20406–20417, 2025. doi: 10.1109/JSTARS.2025.3595373.

Maitiniyazi Maimaitijiang, Vasit Sagan, Paheding Sidike, Ahmad M. Daloye, Hasanjan Erkbol, and Felix B. Fritschi. Crop monitoring using satellite/UAV data fusion and machine learning. Remote Sensing, 12(9):1357, 2020. doi: 10.3390/rs12091357.

Frederica Nonni, Diego Malacarne, Salvatore Eugenio Pappalardo, Daniele Codato, Franco Meggio, and Massimo De Marchi. Sentinel-2 data analysis and comparison with UAV multispectral images for precision viticulture. GI_Forum, 1:105–116, 2018. doi: 10.1553/giscience2018_01_s105.

Dan Popescu, Loretta Ichim, and Florin Stoican. Orchard monitoring based on unmanned aerial vehicles and image processing by artificial neural networks: a systematic review. Frontiers in Plant Science, 14:1237695, 2023. doi: 10.3389/fpls.2023.1237695.

Jinzhuo Qiu, Miguel Ángel Campo Bescós, Agustín Martínez-Fernández, and María Amparo Pérez-Martín. Trade-off between cost and performance of earth observation data in olive orchards. International Journal of Applied Earth Observation and Geoinformation, 136:104379, 2025. doi: 10.1016/j.jag.2025.104379.

Farhad Samadzadegan, Ahmad Toosi, and Farzaneh Dadrass Javan. A critical review on multi-sensor and multi-platform remote sensing data fusion approaches: current status and prospects. International Journal of Remote Sensing, 46(3):1327–1402, 2025. doi: 10.1080/01431161.2024.2429784.

Oiliam Stolarski, Hélder Fraga, Joaquim J. Sousa, and Luís Pádua. Synergistic use of sentinel-2 and UAV multispectral data to improve and optimize viticulture management. Drones, 6(11):366, 2022. doi: 10.3390/drones6110366.

Ahmad Toosi, Farhad Samadzadegan, and Farzaneh Dadrass Javan. Toward the optimal spatial resolution ratio for fusion of UAV and sentinel-2 satellite imageries using metaheuristic optimization. Advances in Space Research, 75(7):5254–5282, 2025. doi: 10.1016/j.asr.2025.02.019.

Dimosthenis C. Tsouros, Stamatia Bibi, and Panagiotis G. Sarigiannidis. A review on UAV-based applications for precision agriculture. Information, 10(11):349, 2019. doi: 10.3390/info10110349.

Yunfei Wang, Weidong Jia, Mingxiong Ou, Xuejun Wang, and Xiang Dong. A review of orchard canopy perception technologies for variable-rate spraying. Sensors, 25(16):4898, 2025. doi: 10.3390/s25164898.

Yingyig Xing, Xuning Liu, and Xiukang Wang. Integrating UAVs, satellite remote sensing, and machine learning in precision agriculture: pathways to sustainable food production, resource efficiency, and scalable innovation. Frontiers in Agronomy, 7:1670380, 2026. doi: 10.3389/fagro.2025.1670380.

Chunhua Zhang and John M. Kovacs. The application of small unmanned aerial systems for precision agriculture: a review. Precision Agriculture, 13(6):693–712, 2012. doi: 10.1007/s11119-012-9274-5.