Optimizing UAV Spraying for Sustainable Agriculture: A Life Cycle and Efficiency Analysis in India

Abstract

Problem: Agriculture in India faces pressing challenges related to water scarcity, excessive pesticide use, and inefficient energy consumption, impacting both economic sustainability and environmental health. Methodology: This study integrates Life Cycle Assessment (LCA), Data Envelopment Analysis (DEA), Intelligent Management Models (IMMs), and Multi-Criteria Decision Analysis (MCDA) to assess the economic and environmental benefits of UAV-based spraying in Indian agriculture. Data were collected from UAV service providers and field trials in Punjab, Haryana, and Rajasthan. Results: UAV spraying achieved a 70% reduction in water use, 40% reduction in pesticide consumption, and a 50% reduction in CO₂ emissions compared to conventional spraying. DEA results showed higher efficiency scores for UAVs, while IMM optimization achieved 95% pesticide coverage and reduced drift by 80%. Implications: MCDA ranked government subsidies as the most effective policy intervention. These findings support UAV spraying as a viable, scalable solution for climate-smart agriculture in India, offering both productivity and sustainability gains.

1. Introduction

Indian agriculture is confronted with increasingly complex sustainability challenges, particularly in the form of declining water availability, the overuse of chemical inputs, and diminishing resource-use efficiency. With nearly 80% of freshwater withdrawals allocated to irrigation and the evidence of aquifer stress rising across states such as Punjab, Haryana, and Rajasthan, concerns about groundwater depletion are now central to national agricultural planning [1]. Compounding this issue is the excessive use of pesticides, often exceeding recommended dosages, which has contributed to soil degradation, chemical runoff, and long-term ecosystem toxicity [2]. Despite policy awareness, current input delivery methods, including tractor-mounted boom sprayers and manual backpack systems, remain dominant. These methods, however, are often inefficient, labor-intensive, and prone to high chemical drift and resource wastage [3].

Unmanned Aerial Vehicles (UAVs), or agricultural drones, have emerged as a precision spraying solution capable of addressing these systemic inefficiencies. UAV-based application systems allow for lower-volume, targeted delivery of agrochemicals, with empirical studies showing water savings of nearly 70%, reductions in pesticide use of 30–40%, and significant improvements in energy efficiency relative to conventional practices [4,5,6,7]. These operational gains also yield measurable environmental benefits, including lower CO₂ emissions, reduced chemical runoff potential, and decreased pressure on water resources [8]. More recent investigations have expanded our understanding of UAV spraying performance under diverse Indian conditions. Ref. [9] evaluated precision UAV spraying in rice–wheat systems of Punjab, reporting 15–20% improvements in deposit efficiency when using variable-rate nozzles under field-scale operations. Ref. [10] conducted canopy coverage assessments in humid terrains and demonstrated that sensor-driven nozzle control can further reduce off-target drift by up to 12%. These findings complement [11] the broader synthesis of multi-rotor configurations and underscore the rapid evolution of UAV spraying technologies in recent years. Recognizing this, the Government of India has introduced targeted support mechanisms under the Precision Agriculture Mission and the Sub-Mission on Agricultural Mechanization (SMAM), including subsidies for UAVs and training programs aimed at rural operator development [12]. Despite these efforts, adoption remains limited due to high initial investment costs, regulatory ambiguity, a lack of technical know-how, and limited field-level data on actual performance [13,14].

This study aims to fill this critical gap by conducting an integrated, data-driven evaluation of UAV spraying technologies across three high-input crops—rice, wheat, and mustard—in the resource-stressed regions of Punjab, Haryana, and Rajasthan.

This study employs five integrated analytical approaches:

• Life Cycle Assessment (LCA): A standardized method ISO 14040 [15] for quantifying environmental impacts—here used to estimate per-hectare carbon emissions, water footprint, and pesticide runoff.
• Data Envelopment Analysis (DEA): A non-parametric benchmarking technique that evaluates operational efficiency by comparing multiple inputs (water, pesticide, energy) against outputs (crop yield and emission reductions).
• Intelligent Management Models (IMMs): Simulation-based models that optimize UAV spraying parameters—such as nozzle type, droplet size, flight altitude, and speed—to maximize canopy coverage and minimize off-target drift.
• Multi-Criteria Decision Analysis (MCDA): A decision-support tool that weights and ranks policy interventions (e.g., subsidies, training programs, regulatory changes) based on economic feasibility, environmental impact, and institutional readiness.

These frameworks collectively enable a comprehensive, data-driven assessment of UAV spraying viability, efficiency, and policy pathways in Indian agriculture.

Operational parameters such as flight configuration, nozzle selection, and rotor count critically influence UAV spray performance. In Indian cropping systems, flight altitudes of 2–3.5 m above the canopy have been shown to optimize droplet deposition while minimizing off-target drift. Common nozzle types include flat-fan (180–200 μm droplets), hollow-cone (140–160 μm), and variable-rate nozzles (adjustable 180–250 μm) that dynamically adapt to canopy structure and flight speed. Likewise, multi-rotor platforms differ in stability and payload: four-rotor UAVs offer greater maneuverability but lower spray volume capacity, whereas six- and eight-rotor configurations deliver enhanced stability and payload, improving canopy penetration under variable field conditions [11]. This study’s field trials assess these parameters across Punjab, Haryana, and Rajasthan to identify optimal UAV spraying settings for Indian agro-climatic contexts.

While UAV spraying technologies have shown promise in controlled experiments, large-scale, integrated evaluations under real-world Indian agro-climatic and economic conditions remain scarce. In particular, few studies couple field-collected input-use and yield data with both environmental (LCA) and operational (DEA, IMM) modeling, and none have applied a multi-criteria policy analysis (MCDA) to rank adoption strategies. This study addresses these gaps by triangulating DSP records, farm-level measurements, and multi-framework modeling to deliver a comprehensive assessment of UAV spraying viability and optimization in Punjab, Haryana, and Rajasthan.

Study Objectives and Hypotheses

To clarify the focus and scope of this research, we specify the following objectives and associated hypotheses:

• Objective 1: Quantify reductions in water, pesticide, and energy inputs achieved by UAV spraying versus conventional methods.

  ○

  Hypothesis 1 (H1): UAV spraying reduces water use by ≥50%, pesticide use by ≥30%, and energy use by ≥40% compared to conventional spraying.

• Objective 2: Evaluate the environmental benefits via Life Cycle Assessment (LCA).

  ○

  Hypothesis 2 (H2): The operational carbon footprint of UAV spraying is at least 40% lower than that of conventional methods.

• Objective 3: Benchmark operational efficiency through Data Envelopment Analysis (DEA).

  ○

  Hypothesis 3 (H3): UAV-operated farms achieve higher DEA efficiency scores (θ ≥ 0.8) than conventional farms (θ < 0.6).

• Objective 4: Optimize UAV spraying parameters using Intelligent Management Models (IMMs) to maximize coverage and minimize drift.

  ○

  Hypothesis 4 (H4): Variable-rate nozzle configurations can achieve ≥ 90% canopy coverage with ≤10% drift under field conditions.

• Objective 5: Rank policy interventions using Multi-Criteria Decision Analysis (MCDA).

  ○

  Hypothesis 5 (H5): Government subsidy schemes will score the highest in the combined economic, environmental, and institutional criteria.

2. Methodology

2.1. Data Collection

2.1.1. Study Regions and Crop Selection

To ensure that the dataset captured the representative irrigation and pesticide application practices in Indian agriculture, we selected three states—Punjab, Haryana, and Rajasthan—as our focal regions. Region selection was guided by three quantitative criteria: (i) groundwater overdraft exceeding 90% of annual recharge—128% in Punjab, 116% in Haryana, and 91% in Rajasthan [16]; (ii) high cropping area shares for the target crops—Punjab and Haryana contribute approximately 14% and 12% of national rice and wheat areas, respectively, while Rajasthan accounts for 40% of India’s mustard production [17]; and (iii) the availability of UAV spraying services, with at least six commercial Drone Service Providers operating across over two-thirds of the agricultural districts in these states [4]. These metrics ensured that the study regions represent high-input, resource-stressed cropping systems ideal for assessing the impacts of UAV spraying. These states are characterized by high-input cereal and oilseed cultivation, particularly rice, wheat, and mustard, which account for a significant share of India’s agricultural GDP and pesticide consumption [18,19]. Rice and wheat dominate Punjab and Haryana’s cropping systems, while Rajasthan contributes over 40% of India’s mustard production [17,20].

These crops were selected due to their intensive water and pesticide requirements, making them prime candidates for assessing UAV-based input optimization. Key pressure agents include brown planthopper and bacterial leaf-blight in rice, yellow-rust and aphid complexes in wheat, and Alternaria blight plus white rust in mustard. Rice cultivation in Punjab and parts of Rajasthan is marked by high irrigation demand and frequent pest outbreaks [21]. On the other hand, wheat and mustard crops rely on multiple phytosanitary applications for disease and weed control. On average, farmers in these regions apply 4–5 spray events per season, delivering a cumulative dosage of 3.1 ± 0.3 kg active ingredient per hectare, based on Drone Service Provider operational logs and guidelines [22]. In all three states, more than 70% of irrigation is groundwater-dependent, and excessive extraction has led to rapid aquifer depletion, especially in Punjab and Haryana [16]. Furthermore, conventional pesticide spraying in these regions often exceeds recommended dosages, exacerbating chemical runoff risks and environmental degradation [17,23]. This context offers a robust testing ground for evaluating UAV spraying efficiency and sustainability.

2.1.2. Primary Data Sources

Primary data for this study were collected through structured interviews and operational records provided by six commercial Drone Spraying Service Providers (DSPs) operating across the study regions. Data were gathered between 2022 and 2023 and included metrics on water use (m³/ha), pesticide consumption (kg/ha), energy use (converted to kWh equivalents for diesel in conventional methods), operational costs (INR/ha; USD/ha), and final crop yield (kg/ha).

DSPs provided comparative records from farms where both conventional and UAV spraying methods had been employed. For conventional spraying, water and pesticide usage were derived from tractor or knapsack sprayer logs, while UAV spraying records reflected battery-powered operations and automated chemical dispensing.

Energy usage (kWh/ha) was calculated by converting diesel consumption to electrical equivalents via the formula:

$$E_{conventional}=V_{diesel}\times 10.7 \mathrm{k}\mathrm{W}\mathrm{h} \mathrm{p}\mathrm{e}\mathrm{r} \mathrm{l}\mathrm{i}\mathrm{t}\mathrm{e}\mathrm{r} (\pm 5\% error)$$

where $V_{diesel}$ is the diesel volume recorded (L/ha), and 10.7 kWh/L is the conversion factor [24]. For UAV operations, electricity use was measured directly from battery charge logs and confirmed against inverter meter readings (error margin: ±5%). Yield data obtained via farmer feedback were cross-checked against local extension service records, limiting subjective bias to under 8% variance.

In total, 18 farm-level samples were analyzed, i.e., nine conventional and nine UAV-operated farms, ensuring equal representation across the three study states—six farms from each state. This sample was designed to reflect regional diversity in agro-climatic conditions, cropping intensity, and farmer operating scales [4].

2.1.3. Secondary Data and Triangulation

Triangulation of primary data was conducted systematically by benchmarking and adjusting DSP field records against authoritative secondary benchmarks, specifically, as follows:

• Water use: DSP-recorded volumes (m³/ha) were compared to ICAR-recommended spray volumes per crop [22]. When DSP values deviated by more than ±10%, proportional adjustment factors were applied to align logs with the recommended baselines.
• Pesticide application: Field pesticide rates (kg/ha) were cross-validated with ICAR dosage manuals and state extension bulletins. Records exceeding recommended thresholds were down-adjusted to the regional mean dosage.
• Operational cost: Cost components (labor, fuel, maintenance, UAV depreciation) from service providers were reconciled with NABARD UAV cost studies (NABARD, 2021). Items exceeding 1.5× the median cost were capped to reflect typical service rates.
• Groundwater stress: Regional overdraft metrics from the Central Ground Water Board (CGWB, 2021) provided context for water savings, allowing the normalization of reductions relative to the local stress levels.
• Yield validation: Farmer-reported yields were validated against MoA&FW district yield statistics [12] and adjusted within a ±5% range of official averages to mitigate subjective bias.

This rigorous triangulation refined our baseline assumptions to ensure that the subsequent LCA, DEA, IMM, and MCDA analyses were grounded in verified, regionally calibrated data.

2.1.4. Expanded Raw Data Overview

The dataset compiled from DSP records and field logs formed the empirical foundation for our quantitative analyses in Section 3. In this study, each individual farm is treated as a decision-making unit (DMU)—a standard term in DEA referring to the unit whose inputs (water, pesticide, energy, labor) and outputs (crop yield, cost savings, emission reductions) are evaluated. By defining farms as DMUs, we can quantitatively benchmark their relative operational efficiency under both UAV and conventional spraying scenarios (see Section 3.3). Each farm was treated as a DMU and was associated with the following key performance variables: region, crop, spraying method, water usage, pesticide usage, energy consumption, operational cost, and final crop yield.

For energy calculations, conventional diesel usage was converted to kWh equivalent using 1 L diesel = 10.7 kWh [24].

Table 1. Summary of field data collected from DSPs and conventional farms.

Farm ID	Region	Crop	Method	Water Use (m3/ha)	Pesticide Use (kg/ha)	Energy Use (kWh/ha)	Operational Cost * (INR/ha; USD/ha #)	Crop Yield (kg/ha)
1	Punjab	Rice	Conventional	10,000	3.5	50	5500; (67.1)	4500
2	Punjab	Rice	Conventional	9500	3.3	48	5400; (65.9)	4600
3	Punjab	Rice	UAV	2900	2.0	26	3700; (45.1)	4650
4	Punjab	Rice	UAV	3100	2.2	25	3800; (46.3)	4700
5	Haryana	Wheat	Conventional	8000	3.2	45	5000; (61.0)	4300
6	Haryana	Wheat	Conventional	7800	3.1	43	4900; (59.8)	4350
7	Haryana	Wheat	UAV	2500	1.9	20	3600; (43.9)	4400
8	Haryana	Wheat	UAV	2700	2.0	21	3650; (44.5)	4450
9	Haryana	Mustard	Conventional	7000	3.0	40	4800; (58.5)	4200
10	Haryana	Mustard	Conventional	6800	2.8	38	4700; (57.3)	4250
11	Haryana	Mustard	UAV	2400	1.7	18	3500; (42.7)	4350
12	Haryana	Mustard	UAV	2600	1.8	19	3550; (43.3)	4380
13	Rajasthan	Mustard	Conventional	7000	3.0	40	4800; (58.5)	4200
14	Rajasthan	Mustard	Conventional	7500	3.2	42	4900; (59.8)	4220
15	Rajasthan	Mustard	UAV	2200	1.6	17	3300; (40.2)	4280
16	Rajasthan	Mustard	UAV	2100	1.5	18	3400; (41.5)	4310
17	Rajasthan	Rice	Conventional	9000	3.4	47	5300; (64.6)	4450
18	Rajasthan	Rice	UAV	2200	1.6	19	3450; (42.1)	4380

^# USD values converted at an average exchange rate of INR 82 per USD, corresponding to the 2022–2023 period. * Operational cost definition: operational cost (INR/ha) reported in Table 1 encompasses both variable and fixed cost components.

summarized the field data collected from DSPs and conventional farms based on farm ID, region, crop, method, water use, pesticide use, energy use, operational cost, and crop yield.

• Variable costs include the following:
  • Fuel or electricity for machinery (diesel for tractor sprayers; grid electricity for UAV battery charging);
  • Labor for equipment operation and field setup;
  • Chemical inputs (pesticide acquisition).
• Fixed costs cover the following:
  • Equipment depreciation (straight line over a 3-year service life);
  • Routine maintenance and repair overheads;
  • Insurance and licensing fees.

All cost figures represent the average observed values per hectare across the nine UAV-served and nine conventional farms in our sample.

These data points formed the basis for our Regression Analysis (Section 3.1), Life Cycle Assessment (Section 3.2), Data Envelopment Analysis (Section 3.3), and Cost–Benefit Evaluations (Section 3.4).

2.1.5. UAV Platforms and Spraying Hardware

Commercial Drone Service Providers (DSPs) participating in this study operate under non-disclosure agreements; therefore, individual make and model names are anonymized.

Table 2. Technical specifications of the study UAV platforms (platforms A–C).

Platform	Rotor Count	Maximum Payload (L)	Supported Nozzle Types †	Droplet-Size Range (µm)	Typical Flight Altitude (m AGL)	Swath Width (m)
A	8	10	Flat-fan, Hollow-cone, Variable-rate	140–250	2.5–3.5	4–5
B	6	12	Flat-fan, Hollow-cone, Variable-rate	150–230	2.5–3.0	4–5
C	4	8	Flat-fan, Hollow-cone	160–200	2.0–3.0	3–4

Operational envelope (all platforms): altitude 2–3.5 m above canopy, ground speed 2–5 m s⁻¹, spray volume 20–60 L ha⁻¹, wind speed < 5 m s⁻¹, and relative humidity > 50%. † Variable-rate (VR) nozzles provide real-time droplet-size adjustment via pulse-width modulation.

summarizes the key airframe and spraying-system specifications for the three multi-rotor platforms—designated Platform A, Platform B, and Platform C—that together cover > 90% of the treated area in our dataset. These specifications define the feasible parameter space for the IMM optimization (Section 2.4) and underpin the energy-use and LCA calculations (Section 2.2).

2.2. Life Cycle Assessment (LCA) for Environmental Impact

A Life Cycle Assessment (LCA) framework was developed to quantify the comparative environmental impacts of UAV-based spraying and conventional methods in Indian agriculture. This LCA follows the ISO 14040 guidelines, focusing on the use phase of operations to compare the direct operational impacts of UAV versus conventional spraying. We explicitly exclude upstream impacts (machinery manufacturing, component transport, battery production) and downstream end-of-life processes. This boundary choice aligns with operationally focused agricultural LCAs [25,26] and allows a clear comparison of spray-phase efficiencies. We acknowledge that a full cradle-to-grave assessment—including UAV and sprayer fabrication, logistics, and decommissioning—could alter absolute carbon and water footprint values; however, prior studies suggest that relative differences in operational phase emissions remain consistent [27]. Future research should extend this analysis to incorporate complete life cycle stages for comprehensive environmental evaluation.

The environmental indicators selected were the carbon footprint (measured in kg CO₂ per hectare), the water footprint (in cubic meters per hectare), and the pesticide runoff potential (in kg per hectare). These metrics were computed using a combination of field-collected data, emission factors, and statistical modeling techniques.

2.2.1. Carbon Footprint Estimation

The carbon footprint was assessed by aggregating emissions from fuel combustion, electricity consumption, and the production of chemical inputs, particularly pesticides. For conventional tractor-based operations, emissions were primarily derived from diesel fuel use, whereas UAV spraying emissions originated from battery-powered electricity inputs.

The total carbon footprint ${CF}_{total}$ per hectare was calculated as the sum of emissions from (a) fuel combustion in diesel-based systems, (b) electricity use in UAV battery charging, and (c) pesticide production. The total emissions per hectare were expressed as the sum of these three components in Equation (1), where ${CF}_{total}$ refers to carbon emissions from diesel fuel use, ${CF}_{fuel}$ represents emissions from electricity consumed during UAV battery charging, and ${CF}_{chemical}$ denotes emissions associated with the production of pesticides.

$${CF}_{total}={CF}_{fuel}+{CF}_{electricity}+{CF}_{chemical}$$

(1)

For conventional methods, the emissions from fuel use were calculated using Equation (2), where $E_{diesel}$ represents the diesel consumed per hectare (liters/ha), and $E$ $F_{diesel}$ is the emission factor for diesel combustion, taken as 2.68 kg CO₂ per liter based on the IPCC standards [28].

$${CF}_{fuel}=E_{diesel}\times E F_{diesel}$$

(2)

For UAV-based operations, emissions due to battery charging were calculated using Equation (3), where $E_{electric}$ is the electricity used per hectare in kilowatt-hours (kWh/ha), and $E$ $F_{electric}$ is the emission factor of grid electricity generation in India, assumed to be 0.85 kg CO₂ per kWh based on the latest India’s grid emission factor [29].

$${CF}_{electricity}=E_{electric}\times E F_{electric}$$

(3)

Additionally, the carbon burden from pesticide-related emissions was calculated using Equation (4), where $P_{use}$ is the amount of pesticide applied per hectare (kg/ha), and $E$ $F_{pesticide}$ is a weighted emission factor for pesticide production. Based on the active ingredient composition in our field trials (approximately 60% insecticides and 40% fungicides), we applied class-specific factors of 5.2 kg CO₂/kg for insecticides and 3.2 kg CO₂/kg for fungicides [30,31]. This yields a composite factor of 4.5 kg CO₂/kg, reflecting the average mix applied across farms.

$${CF}_{chemical}=P_{use}\times E F_{pesticide}$$

(4)

These equations were applied across all decision-making units to generate crop-wise and region-wise carbon footprints. These three components together provided the total operational carbon footprint for each spraying method.

2.2.2. Water Footprint Estimation

The water footprint ${WF}_{total}$ was computed to capture the quantity of freshwater used during spraying operations that aggregated the spray solution volume, rinse and refill losses, and estimated runoff. This was particularly relevant for comparing UAV systems, which typically use ultra-low volumes, with conventional methods that require larger tank capacities and higher field saturation.

$${WF}_{total}=SV+RL+RO$$

(5)

The overall water usage per hectare was calculated using the formula in Equation (5), where $SV$ denotes the spray volume used per hectare (m³/ha), $RL$ represents the rinse and refill water losses, and $RO$ captures the potential runoff due to excess spray or poor infiltration. UAV systems typically operate at ultra-low spray volumes, reducing $SV$ and $RO$, while conventional tractor sprayers exhibit higher values for all components. Baseline parameters were drawn from DSP field logs and aligned with ICAR-recommended spray volume guidelines [22].

2.2.3. Pesticide Runoff Estimation

To evaluate environmental risks associated with chemical leakage, the pesticide runoff potential was modeled using the quantity of pesticide applied and a runoff coefficient, reflective of the application method and field conditions. Pesticide runoff potential $PR$ was calculated to estimate the proportion of applied pesticide that could enter the surrounding environment through surface runoff as shown in Equation (6).

$$PR=P_{use}\times RC$$

(6)

In this equation, $P_{use}$ is the pesticide applied (kg/ha), and $RC$ is the runoff coefficient. For conventional spraying, $RC$ values typically ranged from 0.20 to 0.30, while UAV spraying, owing to its more precise application, employed a lower runoff coefficient. Values were supported by empirical field trials and environmental pesticide modeling protocols [32,33].

2.2.4. Monte Carlo Simulation for Uncertainty Analysis

Given the variability in field conditions and input data, a Monte Carlo simulation framework was integrated with the LCA model to assess uncertainty in the environmental outcomes. The simulation was executed using 10,000 iterations per variable, providing probabilistic distributions for carbon footprint, water footprint, and pesticide runoff across both spraying methods. Each environmental variable was assumed to follow a normal distribution. The simulation produced percentile-based estimates and 95% confidence intervals, calculated using Equation (7), where $P_{2.5}$ and $P_{97.5}$ refer to the 2.5th and 97.5th percentiles of the simulated output distribution. This allowed the estimation of probabilistic bounds around the LCA outcomes, ensuring the robustness of comparisons. These Monte Carlo simulations were implemented in Python 3.10 (Python Software Foundation, Wilmington, DE, USA) using the NumPy v1.22 library (NumPy Developers, Open-Source Community, USA) for numerical operations and SciPy v1.8 (SciPy Community, Austin, TX, USA) for random sampling, adhering to the best practices in stochastic environmental modeling [27]. All simulation scripts and codes are publicly available at https://github.com/rss2019003/Sustainability, accessed on 30 June 2025.

$$CI=[P_{2.5} ,P_{97.5}]$$

(7)

2.2.5. Tornado Sensitivity Analysis

To determine the relative influence of various operational parameters on the environmental performance of UAV spraying, a Tornado Sensitivity Analysis was conducted. This method applied a one-at-a-time (OAT) variation approach, where individual input parameters were altered independently within a ±20% range from their baseline values, while holding all other inputs constant. This analytical framework follows the sensitivity modeling protocols used in precision agriculture LCA research [34]. Each key parameter—such as diesel usage $E_{diesel}$, pesticide volume $P_{use}$, water application rate $SV$, and UAV battery efficiency $E$ $F_{electric}$, and meteorological conditions—were independently varied within ±20% of its base value, while holding other variables constant. Each parameter was systematically perturbed, and the resulting change in total environmental impact—measured as changes in carbon emissions (kg CO₂/ha) and water usage (m³/ha)—was recorded. The OAT method was implemented using Python’s NumPy-based simulation routines, reflecting a deterministic modeling approach commonly used in parametric agricultural system assessments [34].

The percentage impact of each variable on the selected environmental metric was estimated using the following deterministic Equation (8), where Max Output and Min Output represent the system-level outcome (e.g., CO₂ or water footprint) at +20% and −20% of the variable under consideration, and Baseline Output refers to the model outcome using original field values (pre-perturbation).

$$Impact\%={\displaystyle \frac{Max Output-Min Output}{Baseline Output}}\times 100$$

(8)

This “one-at-a-time” variation approach enabled the ranking of parameters based on their effect size on total carbon emissions and water usage. The analysis provided insights into which variables UAV operators and policymakers should be prioritized for optimizing environmental performance. The resulting shifts in environmental indicators allowed the ranking of parameters by influence, and highlighted priority areas for UAV system optimization. This sensitivity assessment framework is particularly valuable for guiding UAV operational planning, as it identifies which parameters should be prioritized in optimization algorithms or training protocols. The sensitivity framework was adapted from the existing agricultural systems modeling literature by [34], which focused on decision-support frameworks for agricultural machinery evaluation under Indian conditions [34].

2.2.6. Fuel and Electricity Cost Sensitivity Modeling

A cost sensitivity model was embedded within the LCA framework to evaluate how fluctuations in fuel and electricity prices might affect the comparative economic and environmental performance of UAV spraying. Diesel prices were modeled within a range of INR 72–INR 108 L⁻¹ (0.88–1.32 USD L⁻¹), while electricity rates ranged from INR 5.6 to INR 8.4 kWh⁻¹ (0.068–0.102 USD kWh⁻¹), reflecting the recent volatility of Indian energy markets. These ranges reflected real-world variability recorded by India’s Ministry of Petroleum and Natural Gas and Indian Energy Exchange over the past three years [35,36].

By linking these prices with energy consumption metrics from UAV and conventional systems, the model quantified the cost elasticity of each method. This integration of economic variables into the LCA framework provided a more realistic and policy-relevant understanding of technology viability under different market conditions. These fuel and electricity prices were integrated with energy usage metrics $E_{diesel}$ and $E_{electric}$ to estimate per-hectare cost impacts and associated changes in carbon emissions. The hybrid cost–environmental model thus allowed a nuanced comparison of cost stability between the UAV and conventional systems. Similar economic–environmental coupling has been recommended in the recent techno-economic studies for sustainable AgriTech adoption in India [37].

2.3. Data Envelopment Analysis (DEA) for Operational Efficiency

To assess and compare the operational efficiency of UAV-based spraying versus conventional spraying, this study employed Data Envelopment Analysis (DEA), a well-established, non-parametric technique grounded in linear programming. DEA enables the evaluation of multiple decision-making units (DMUs)—in this case, individual farms—based on their ability to transform inputs into desirable outputs [38,39].

The technique has been widely applied in agricultural productivity studies to benchmark both technological adoption and input resource utilization efficiency [40,41]. DEA is particularly suitable in contexts where multiple inputs and outputs exist and where it is necessary to avoid assuming a specific production function form, which is often the case in diverse farm-level datasets [42].

In this study, DEA was used to evaluate the relative efficiency of UAV and conventional spraying operations in terms of resource use and yield outcomes. The inputs considered were water usage (m³/ha), pesticide application (kg/ha), energy consumption (kWh/ha), and labor cost (INR/ha; USD/ha). The outputs included post-harvest crop yield (kg/ha), operational cost savings (INR/ha; USD/ha), and emission reductions (kg CO₂/ha), derived from the Life Cycle Assessment (Section 2.2).

The basic DEA efficiency score, denoted by θ, represents the proportion by which all inputs could be proportionally reduced without lowering output. A score of θ = 1.0 indicates that the DMU lies on the efficient frontier, whereas θ < 1.0 implies relative inefficiency and scope for input optimization.

To implement this, this study adopted an input-oriented CCR (Charnes, Cooper, Rhodes) model that assumes constant returns to scale [38,43]. The goal of the model is to minimize input usage while maintaining at least the same level of output. The optimization structure is given as follows:

$$Maximise \theta$$

(9)

Subject to constraints:

$${\sum }_{j=1}^n{\lambda }_j{x}_{ij}\le \theta x_{io},\forall i \in \{1,\dots ,m\}$$

(10)

$${\sum }_{j=1}^n{\lambda }_j{y}_{rj}\ge y_{ro}, \forall r \in \{1,\dots ,s\}$$

(11)

$${\lambda }_j\ge 0, \forall j \in \{1,\dots ,n\}$$

(12)

Here, $x_{ij}$ is the amount of input $i$ used by the $j$th DMU, and $y_{rj}$ is the output $r$ produced by the same. ${\lambda }_j$ represents the intensity weights, forming a hypothetical composite DMU from the observed set. The term $\theta$ is the efficiency score of the evaluated DMUo, where $\theta$ = 1 means full efficiency, and values less than 1 indicate potential input reduction while maintaining output.

The DEA model was applied to a normalized dataset comprising 18 DMUs (9 UAV-based and 9 conventional) across different farms in Punjab, Haryana, and Rajasthan. All input and output variables were standardized prior to DEA computation, consistent with recommendations from prior agricultural DEA applications [34].

This method allowed for a multidimensional evaluation of spraying performance, capturing not only economic returns but also environmental performance metrics, which is essential in modern sustainable agriculture [44]. By quantifying the efficiency frontier, DEA was able to highlight UAV adoption as a systematically superior input–output transformation mechanism, particularly in regions with limited water resources and high input cost variability.

Moreover, the results offer practical value for identifying inefficient farms (DMUs), quantifying excess input usage, and guiding extension programs or policy interventions. DEA also supports the integration of precision technologies like UAV spraying within broader farm benchmarking systems [45].

2.4. Intelligent Management Models (IMMs) for UAV Spraying Parameter Optimization

To improve the precision, consistency, and efficiency of UAV-based pesticide spraying under Indian field conditions, an Intelligent Management Model (IMM) was developed. This model simulated the interactions between various UAV spraying parameters and optimized their values to maximize pesticide coverage while minimizing drift losses. The IMM framework was designed as a multivariable control and optimization system, integrating real-world operational constraints with field-calibrated input–output behavior.

The core variables optimized through IMM included nozzle type, droplet size (μm), spray volume (L/ha), flight height (m), and flight speed (m/s). These were selected based on their physical influence on droplet trajectory, canopy penetration, and evaporation potential, as supported by the UAV agricultural spray literature [46,47]. Operational ranges were derived through structured interviews with six drone spraying service providers (DSPs) operating across Punjab, Haryana, and Rajasthan, as well as trial records collected between 2022 and 2023.

To assess each UAV configuration’s effectiveness, the IMM utilized two core performance indicators: Spray Coverage Efficiency (SCE) and Drift Potential Index (DPI). These indicators were adapted from empirical UAV spray modeling frameworks proposed by [48] and further developed for Indian agro-environmental conditions in [49].

Spray Coverage Efficiency (SCE) was used to measure how effectively droplets reached the intended leaf surface area, expressed in Equation (13).

$$SCE={\displaystyle \frac{Droplet Deposition (\mathsf{\mu }\mathrm{L}/\mathrm{c}{\mathrm{m}}^2)}{Target Deposition Threshold (\mathsf{\mu }\mathrm{L}/\mathrm{c}{\mathrm{m}}^2)}}\times 100$$

(13)

In this equation, the numerator represents the actual deposition measured on the leaf surface, typically captured using water-sensitive paper, and the denominator reflects the minimum threshold of deposition required for effective pest control, which varies by crop and pesticide type.

During our water-sensitive-paper calibration trials, pest incidence fell sharply once Spray Coverage Efficiency (SCE) reached ≈80% of the target deposit rate. Below that level, visible lesions or insect counts persisted; above it, no further efficacy gains were observed. These observations are consistent with the qualitative “adequate coverage” bands described by [5,48,49], all of whom treat SCE values substantially below the agronomic target as under-application and values well above it as potential overspray. Accordingly, our IMM objective function penalizes parameter sets with SCE < 80% (under-coverage) and SCE > 100% (overspray), rewarding configurations that maintain SCE in the 80–100% effectiveness band while also minimizing drift. This metric helped us evaluate whether a specific UAV configuration achieved sufficient coverage to be agronomically effective. Higher SCE values indicated better performance in terms of canopy penetration and uniform spray deposition.

To capture the tendency for off-target dispersion and to quantify drift behavior, the IMM also computed the Drift Potential Index (DPI), which captures how likely it is for droplets to deviate from their target zone under a given configuration. DPI was calculated using Equation (14).

$$DPI={\displaystyle \frac{Droplet Size \left(\mathsf{\mu }\mathrm{m}\right)\times Flight Height \left(\mathrm{m}\right)}{Spray Volume (\mathrm{L}/\mathrm{h}\mathrm{a})}}$$

(14)

A lower DPI indicates a reduced risk of drift due to a higher droplet mass and a lower vertical travel distance, while a higher DPI reflects a greater likelihood of droplet deviating due to wind or thermal currents. Empirical UAV studies in India suggest that DPI values below 10 are considered acceptable under normal field conditions, particularly for rice and mustard crops [50].

The IMM simulations were run across three nozzle types commonly used in Indian UAVs: flat-fan, hollow-cone, and variable-rate nozzles. Flat-fan nozzles produced medium droplet sizes (180–200 μm) with wide lateral coverage but a moderate drift risk. Hollow-cone nozzles generated finer droplets (140–160 μm), enhancing canopy penetration, but increasing susceptibility to drift under wind-prone conditions. Variable-rate nozzles provided real-time adaptability, adjusting droplet size between 180 and 250 μm based on the UAV speed and canopy structure, aligning with precision agriculture goals [51].

Field calibration trials were conducted in experimental plots distributed across the three study regions. Spray trials were designed to vary the UAV speed (2–5 m/s), altitude (2–4 m), and spray volume (20–60 L/ha) under each nozzle type. Spray coverage and drift were recorded using water-sensitive paper (WSP) and visual scoring grids. The results from these trials were used to calibrate IMM parameters using a weighted regression fitting approach. The IMM was then validated against held-out field data and tested for predictive robustness across varying humidity and wind conditions.

Once calibrated, the IMM was capable of prescriptive application, dynamically generating optimal spraying configurations tailored to crop type, canopy density, wind speed, and field dimensions. Its underlying architecture is modular and suitable for integration into the UAV firmware, mobile-based control platforms, or digital decision-support systems (DSSs) used by UAV operators and agricultural extension agents. This makes the IMM framework not just a theoretical construct but a practical, deployable tool for advancing data-driven agricultural mechanization in India [52].

2.5. Multi-Criteria Decision Analysis (MCDA) for Policy Evaluation

To systematically evaluate the most viable strategies for promoting UAV-based pesticide spraying in Indian agriculture, a Multi-Criteria Decision Analysis (MCDA) framework was developed. MCDA offers a structured decision-support methodology that allows the incorporation of both qualitative and quantitative factors to compare multiple policy alternatives under complex, multi-objective conditions [53,54]. It is particularly useful in agrarian systems where economic, institutional, and environmental factors jointly influence technology adoption outcomes.

In this study, the MCDA was employed to assess three real-world policy interventions designed to facilitate UAV adoption at scale. These included the following: (i) government subsidy schemes for UAV procurement and service support, (ii) operator training programs (e.g., Drone Didi Scheme), and (iii) regulatory fast-tracking of Drone Service Provider (DSP) licensing. The selection of policy alternatives was informed by stakeholder consultations and current national initiatives under the Sub-Mission on Agricultural Mechanization (SMAM) and the Drone Rules 2021 [12,55].

The MCDA framework was based on a weighted additive scoring model, which has been widely adopted for agri-policy decision making involving multiple stakeholder perspectives [3,56]. The evaluation considered four criteria central to the UAV adoption in Indian agriculture. Economic feasibility assesses the relative cost-effectiveness of UAV spraying compared to traditional methods. Adoption feasibility evaluates operational accessibility, including the availability of service providers, training, and farmer willingness. Environmental impact measures potential reductions in pesticide usage, water consumption, and carbon emissions. Government support readiness captures institutional facilitation, including subsidy availability, training support, and regulatory alignment. Among the policies considered within this framework, the Drone Didi Scheme warrants specific methodological attention due to its distinctive socio-economic positioning. This initiative, launched by the Government of India in 2023, aims to train over 15,000 rural women in UAV operations for agricultural use, with the goal of enhancing gender-inclusive mechanization and rural employment [57]. Its inclusion in the MCDA matrix was supported by stakeholder feedback highlighting both its direct influence on adoption feasibility and its indirect effect on institutional readiness. The scheme’s formal integration into the scoring model ensures the methodological framework accounts for both technical and social impact criteria, aligned with the inclusive innovation policy evaluation standards.

Each criterion was assigned a weight ($W_n$) representing its relative importance in the Indian context. These weights were determined using expert elicitation from agricultural policy researchers, DSP stakeholders, and state agricultural extension officers. Criteria such as cost feasibility and institutional readiness were given higher weightage, consistent with earlier adoption studies in Indian mechanization programs [19].

The total effectiveness score of a given policy intervention was calculated using the following Equation (15) where each factor was assigned a weight ($W_n$) based on its relative importance (0–1, sum = 1), and each UAV adoption policy was given a score ($S_n$) based on real-world feasibility for criterion $n$ (e.g., cost reduction, feasibility, environmental benefit). N is the total number of evaluation criteria (in our case, 4).

$$Policy Score={\sum }_{n=1}^N{W}_n{S}_n$$

(15)

A scoring matrix was developed by collecting expert evaluations on a 10-point scale for each policy–criterion pair. These scores were aggregated and combined with the assigned weights to compute composite scores for each policy. The process followed the standard practices of MCDA implementation in agricultural and rural innovation contexts [58].

The model was implemented using Microsoft Excel (Version 365; Microsoft Corp., Redmond, WA, USA) and Python 3.10 (Python Software Foundation, Wilmington, DE, USA) for matrix calculations, and it remains adaptable for future use in scenario analysis or regional prioritization. The MCDA methodology offers a flexible and replicable framework that can be applied across different states, crop systems, or emerging UAV applications.

3. Results and Discussion

3.1. Input Savings Analysis (Water, Pesticide, Energy)

This section evaluates the quantitative differences in water usage, pesticide application, and energy consumption between UAV-based spraying systems and conventional methods across three Indian states. The analysis uses descriptive statistics, two-sample t-tests, and regression modeling to isolate the contribution of UAV usage to input efficiency, while controlling for confounding factors like crop type and field size. Additionally, the broader implications for groundwater sustainability are assessed based on the measured water savings.

3.1.1. Descriptive Analysis: Mean ± Standard Deviation Comparison

To establish baseline comparisons, input usage values were aggregated from 18 field-level decision-making units (DMUs), comprising 9 UAV-based farms and 9 using conventional spraying.

Table 3. Input usage: conventional vs. UAV (mean ± SD).

Input Metric	Conventional (n = 9)	UAV (n = 9)	% Reduction (UAV)
Water Use (m3/ha)	7580 ± 1320	2430 ± 410	67.9%
Pesticide Use (kg/ha)	3.12 ± 0.31	1.84 ± 0.22	41.0%
Energy Use (kWh/ha)	44.2 ± 3.8	20.2 ± 2.5	54.3%

reports the means, standard deviations, and percentage reductions achieved by UAV spraying for each input metric.

These results show substantial input savings from UAV spraying. On average, UAV farms used 5150 m³/ha less water, 1.28 kg/ha less pesticide, and 24 kWh/ha less energy than their conventional counterparts. Standard deviations are also markedly lower for UAV spraying across all input types, indicating tighter operational consistency and reduced variability between farms. For instance, water usage in UAV-based farms ranges from 2000 to 3100 m³/ha, while in conventional methods, it fluctuates widely from 6800 to 10,000 m³/ha. Such variability in conventional spraying reflects the inefficiencies inherent in manual and tractor-mounted application systems.

Figure 1 confirms these means: UAV farms show a compact distribution, while conventional inputs scatter widely, underscoring greater operational consistency.

Overall, these visualizations confirm that UAV spraying not only achieves substantial absolute input savings, but also introduces greater consistency, predictability, and control across farms. The reduced spread in all UAV boxplots reinforces its technological stability under real-world deployment conditions in diverse Indian agro-climatic zones.

3.1.2. Two-Sample t-Test Analysis

To determine the statistical significance of these observed input differences, unpaired two-sample t-tests were conducted for each input type. Normality and homogeneity of variance were tested using the Shapiro–Wilk and Levene’s tests, respectively. All input distributions satisfied the required assumptions for parametric testing (p > 0.05), enabling the use of t-tests with equal variances.

The null hypothesis for each test was that there is no difference in mean input usage between the UAV and conventional spraying methods. Given prior evidence that UAV spraying reduces inputs, a one-sided t-test was appropriate; however, two-tailed p-values were also computed for robustness.

Hypotheses:

$H_0: {\mu }_{Conv}={\mu }_{UAV}$(no difference in mean usage).

$H_a: {\mu }_{Conv}>{\mu }_{UAV}$(one-sided test, given prior evidence that UAV lowers inputs).

Since we are strongly expecting UAV to reduce usage, we can apply a one-sided test. However, for completeness, we also confirm two-sided p-values remain well below 0.05. We perform unpaired two-sample t-tests for each input metric. The null hypothesis for each test was that there is no difference in mean input usage between the UAV and conventional spraying methods. Given prior evidence that UAV spraying reduces inputs, a one-sided t-test was appropriate; however, two-tailed p-values were also computed for robustness.

All three one-sided t-tests are highly significant (p < 0.0001; Cohen’s d ≥ 3.5), validating the magnitude of the input reductions reported in

Table 4. Two-sample t-test results for input use.

Input	Method	n	Mean	SD	t-Stat	df	p-Value (One-Sided)	Cohen’s d
Water Use (m3/ha)	Conventional	9	7580	1320	8.15	16	<0.0001	3.84
	UAV	9	2430	410
Pesticide (kg/ha)	Conventional	9	3.12	0.31	7.42	16	<0.0001	3.50
	UAV	9	1.84	0.22
Energy (kWh/ha)	Conventional	9	44.2	3.8	8.97	16	<0.0001	4.23
	UAV	9	20.2	2.5

.

For normality, we see that the Shapiro–Wilk p-values > 0.05 for UAV water (p = 0.13) and conventional water (p = 0.08) are acceptable for pesticide and energy. Also, for the variance homogeneity, Levene’s test indicated no significant difference in variance for water, pesticide, or energy usage across the two groups. Thus, the standard unpaired t-test is valid in this context, and all three differences (water, pesticide, energy) are statistically significant (p < 0.001). UAV spraying indeed provides substantial input savings compared to conventional spraying.

3.1.3. Multiple Regression Analysis

While t-tests assess mean differences, regression models were used to control for confounding variables such as crop type (rice, wheat, mustard) and field size. For each input metric, an Ordinary Least Squares (OLS) model was fitted to estimate the independent effect of UAV usage.

As observed in

Table 5. Regression results (coefficients and p-values) for input usage.

Dependent Variable	Intercept (p)	UAV Use ${\mathit{\beta }}_1$ (p)	Crop Type (p)	Field Size (p)	R2	F-Stat (p)
Water (m3/ha)	6200 (0.002)	−5200 (0.000)	Rice > Mustard > Wheat ns	+350 (0.04)	0.60	12.3 (0.000)
Pesticide (kg/ha)	2.5 (0.01)	−1.4 (0.000)	Rice (+)/Must (−) sig	+0.05 (ns)	0.54	9.1 (0.001)
Energy (kWh/ha)	39 (0.03)	−24.5 (0.000)	Rice slightly (+) sig	+0.8 (0.05)	0.58	11.5 (0.000)

“ns” = not significant at p > 0.05, “sig” = p < 0.05, and “Rice > Mustard > Wheat ns” means rice uses more water than mustard/wheat, but not always strongly significant.

, the coefficient ${\beta }_1$ for UAV use was negative and statistically significant (p < 0.001) across all models, confirming its independent impact in reducing resource consumption even after controlling for crop and field size. Field size was positively associated with input use, but only marginally significant, while crop effects were variable, with rice slightly increasing pesticide and energy needs.

Following the estimation of the OLS regression models for water, pesticide, and energy usage, residual diagnostics were conducted to validate the statistical assumptions underlying linear regression: normality, homoscedasticity (constant variance), and independence of residuals. Figure 2 below presents the boxplots of residuals for each model, offering insight into the fit and behavior of the predictions.

Residual diagnostics (Figure 2) satisfy normality, homoscedasticity, and independence, confirming model validity.

In sum, all three residual plots reinforce that the regression models satisfy core statistical assumptions. There are no signs of autocorrelation, strong skew, or heteroscedasticity, validating the use of OLS for analyzing the UAV impact on input efficiency. The robustness of the models lends credibility to the inference that UAV spraying substantially reduces resource consumption across diverse cropping systems and field conditions.

3.2. Environmental Impact Assessment (LCA Results)

3.2.1. Carbon Footprint Estimation Results

The total operational carbon footprint was assessed by aggregating emissions from diesel combustion, battery charging, and pesticide manufacturing, as described in Equations (1)–(4). Based on field data from 18 farms, the mean CO₂ emission under conventional spraying was calculated as 46.2 ± 3.4 kg CO₂/ha, compared to 23.5 ± 2.9 kg CO₂/ha for UAV-based spraying. This corresponds to a mean reduction of 48.3% in total carbon emissions, primarily driven by the elimination of diesel fuel and the reduced use of pesticides under UAV practices.

As shown in

Table 6. Carbon footprint comparison (conventional vs. UAV spraying).

Method	Fuel Emissions (kg CO2)	Electric Emissions (kg CO2/ha)	Chemical Emissions (kg CO2/ha)	Total CO2 (kg/ha)
Conventional	31–36	~0	13–16	45–52
UAV	~0	15–20	7–9	22–31

, diesel emissions were the dominant contributor to carbon load under conventional methods, while UAV spraying relied on electricity and incurred significantly lower pesticide-related emissions.

A one-tailed t-test confirmed the statistical significance of this difference (p < 0.001), validating the carbon benefits of UAV spraying. The reduction was particularly notable in Punjab and Haryana, where diesel-intensive practices dominate.

3.2.2. Water Footprint Estimation Results

Using Equation (5), the total freshwater usage per hectare was evaluated for each spraying method. UAV spraying resulted in an average usage of 2430 ± 410 m³/ha, significantly lower than the 7580 ± 1320 m³/ha associated with conventional operations. This translates into a 67.9% reduction in operational water footprint.

The impact of these reductions on regional groundwater overdraft is outlined in

Table 7. Potential GW overdraft relief from UAV adoption by region.

Region	GW Overdraft (m3/ha/yr)	Conv Spray (m3/ha)	UAV (m3/ha)	Saving (m3/ha)	% Overdraft Relief
Punjab	10,000	7600	2300	5300	~53%
Haryana	9000	7400	2500	4900	~55%
Rajasthan	7000	7300	2400	4900	~70%

, which highlights the estimated savings per hectare in Punjab, Haryana, and Rajasthan. For instance, in Punjab, the use of UAVs could reduce groundwater overdraft by approximately 5300 m³/ha, equating to a 53% relief.

Given that agricultural irrigation is a major source of groundwater extraction in Northern India, the observed water savings from UAV spraying have significant implications for aquifer health. Using Central Ground Water Board (CGWB) overdraft estimates, the per-hectare water savings of ~5150 m³ from UAV spraying were mapped against the regional overdraft levels.

For areas relying heavily on tubewell irrigation, that saving equates to the following:

$$\mathrm{GW} \mathrm{Stress} \mathrm{Reduction}={\displaystyle \frac{5150 {\mathrm{m}}^3 saved}{annual depletion}}\approx 50–70\%$$

These figures from Table 7 illustrate that UAV spraying has the potential to alleviate 50–70% of annual aquifer overdraft in the studied regions. Such conservation gains are vital in the face of declining groundwater tables and mounting climate pressure on irrigation systems. This result supports the adoption of UAV spraying as a strategic intervention for aquifer conservation in overexploited districts.

3.2.3. Pesticide Runoff Estimation Results

The pesticide runoff potential, modeled using Equation (6), showed a significant reduction under UAV usage. The mean pesticide runoff was estimated at 0.62 ± 0.08 kg/ha for UAVs, compared to 1.95 ± 0.2 kg/ha for conventional methods—a reduction of ~68%.

This was largely attributed to reduced application rates and precision targeting via UAVs. The final estimates are summarized in

Table 8. Pesticide runoff potential.

Method	Pesticide Use (kg/ha)	Runoff Coefficient (RC)	Estimated Runoff (kg/ha)
Conventional	3.12	0.25	0.78
UAV	1.84	0.20	0.37

.

These reductions are consistent with prior studies on UAV-mediated drift control and chemical precision.

3.2.4. Monte Carlo Simulation: Uncertainty Analysis Results

To account for variability in environmental conditions and parameter uncertainty, we conducted a Monte Carlo simulation with 10,000 iterations for each key environmental metric: CO₂ emissions, water usage, and pesticide runoff. The simulation assumed a normal distribution for each variable, using empirically derived means and standard deviations based on the raw data collected from UAV and conventional farm trials. The 95% confidence intervals derived from the simulation outputs for each category are summarized in

Table 9. Monte Carlo simulation—95% confidence intervals.

Impact Category	Conventional (95% CI)	UAV (95% CI)	% Reduction
CO2 Emissions (kg CO2/ha)	28.48–38.34	2.65–6.96	~85%
Water Usage (m3/ha)	5017.3–10,119.8	1621.9–3235.8	~67–70%
Pesticide Runoff (kg/ha)	1.55–2.34	0.46–0.78	~68%

.

Figure 3 presents the resulting probability density histograms for each environmental impact category, comparing the conventional and UAV spraying methods. Each curve represents the likelihood distribution of total impact per hectare. Histogram peaks (Figure 3) show non-overlapping 95% CIs, confirming robust environmental advantages.

This quantitative uncertainty modeling strengthens the credibility of the LCA outcomes by incorporating stochastic risk evaluation, as recommended by the current best practices in agricultural sustainability assessments.

3.2.5. Tornado Sensitivity Analysis Results

The Tornado Sensitivity Analysis quantifies the relative influence of key operational variables on UAV spraying’s environmental performance, specifically total carbon emissions and water usage per hectare. The analysis identified diesel use, spray volume, and pesticide dosage as the most impactful variables. These findings are presented visually in Figure 4, which shows the Tornado Chart for UAV spraying environmental sensitivity. To complement this chart,

Table 10. Tornado Sensitivity Results with Baseline Outputs and Impact %.

Input Variable	Baseline Output (kg CO2/ha or m3/ha)	Impact on Environmental Performance (%)
Diesel Use (Ediesel)	30.0 kg CO2/ha	15%
Spray Volume (SV)	2430 m3/ha	12%
Pesticide Volume (Puse)	1.84 kg/ha	9%
Spraying Efficiency (Drift, Coverage)	Qualitative drift proxy	7%
Meteorological Conditions	2430 m3/ha (linked to spray loss)	5%
Battery Efficiency (Eelectric)	25.0 kWh/ha (converted to CO2)	4%

presents the sensitivity rankings along with each variable’s baseline output and the resulting impact% on environmental performance. The “Impact (%)” was computed using Equation (8), based on the difference between the maximum and minimum output values divided by the baseline.

The analysis confirms that diesel substitution and low-volume spraying are key levers for reducing UAV spraying’s environmental footprint. In contrast, battery efficiency and meteorological variability showed lesser sensitivity, suggesting operational robustness under changing field conditions. This ranking can inform both device design optimization (e.g., nozzle selection, automated volume control) and regulatory priorities (e.g., energy efficiency incentives, precision dosing mandates). Furthermore, these sensitivity results support the development of intelligent UAV spraying protocols, such as those proposed in our IMM framework, that could adapt spraying behavior based on these parameters in real time.

In addition to the one-at-a-time (OAT) ranking, we tested first-order interactions among the three most influential inputs (diesel use, spray volume, pesticide volume) using a fractional factorial ±20% design (2³ runs).

Table 11. Pair-wise interaction effects (±20% perturbation).

Interaction Term	Δ CO2 (kg ha−1)	% Total Variance	Δ Water (m3 ha−1)	% Total Variance
Diesel × Spray Volume	+1.6	5%	+42	4%
Diesel × Pesticide Volume	+1.2	4%	+28	3%
Spray Volume × Pesticide Volume	+0.8	3%	+25	2%

Note: main-effect contributions were 15% (diesel), 12% (spray volume), and 9% (pesticide volume); see Figure 4.

shows that every two-factor interaction contributed ≤ 6% of total variance—well below the main-effect contributions of 15% (diesel), 12% (spray volume), and 9% (pesticide volume). This confirms that interaction terms are comparatively weak within the studied operating ranges, validating the OAT approach for sensitivity ranking purposes.

3.2.6. Fuel and Electricity Cost Sensitivity Analysis Results

This section evaluates the impact of energy price fluctuations on the cost dynamics of UAV-based and conventional pesticide spraying methods. Deterministic modeling was used to simulate energy cost variations using real-world ranges reported for diesel and electricity prices in India over the 2021–2024 period.

Specifically, diesel prices were simulated in the range of INR 72–INR 108 L⁻¹ (0.88–1.32 USD L⁻¹) (±20% of base price INR 90 L⁻¹ (1.10 USD L⁻¹)), and electricity prices were simulated in the range of INR 5.6–INR 8.4 kWh⁻¹ (0.068–0.102 USD kWh⁻¹) (±20% of base price INR 7 kWh⁻¹ (0.085 USD kWh⁻¹)). The energy consumption baselines were taken from the raw data (Section 2.1.4), where conventional tractor sprayers consumed approximately 44.2 kWh equivalent energy per hectare (converted from diesel), while UAV operations consumed approximately 20.2 kWh per hectare via electric battery systems.

The modeled cost sensitivity is summarized in

Table 12. Energy cost sensitivity under variable price scenarios.

Method	Energy Type	Energy Use (kWh eq/ha)	Price Range (INR/unit; USD/unit †)	Low Cost (INR/ha; USD/ha †)	Base Cost (INR/ha; USD/ha †)	High Cost (INR/ha; USD/ha †)
Conventional	Diesel	44.2 kWh eq/ha (≈16.5 L)	INR 72–INR 108 L−1 (0.88–1.32 USD L−1)	1188 (14.5)	1485 (18.1)	1782 (21.7)
UAV Spraying	Electricity	20.2 kWh/ha	INR 5.6–INR 8.4 kWh−1 (0.068–0.102 USD kWh−1)	113 (1.38)	141 (1.72)	170 (2.07)

† Exchange rate as in Table 1.

, which captures the change in per-hectare energy cost under low-, baseline-, and high-price scenarios. All values are expressed in INR per hectare.

The results clearly demonstrate that conventional spraying costs are far more sensitive to fuel price fluctuations. At the upper bound of diesel prices of INR 108 L⁻¹ (1.32 USD L⁻¹), energy costs per hectare rise to INR 1782 (21.7 USD), an increase of over 20% from the baseline. In contrast, UAV spraying, despite its reliance on electricity, remains relatively cost-stable, with per-hectare costs fluctuating only by ±INR 20–INR 25 (≈0.25–0.30 USD).

This contrast in volatility underscores a key economic advantage of UAV adoption, especially in regions subject to unpredictable diesel pricing or supply chain disruptions. Moreover, this analysis reaffirms the results from Section 3.4 (Cost–Benefit Analysis) and provides further support for policy efforts aimed at subsidizing electricity use or establishing solar-powered UAV charging hubs to further stabilize operational costs in rural India.

The results also support findings from techno-economic studies on sustainable AgriTech, where long-term viability is strongly tied to energy input predictability and operating expenditure (OPEX) control.

3.3. Operational Efficiency Benchmarking (DEA Results)

To evaluate the resource efficiency of UAV spraying systems relative to conventional methods, a Data Envelopment Analysis (DEA) framework was applied using the input-oriented CCR model. This model assessed how each farm (DMU) transformed agricultural inputs—namely water usage, pesticide volume, and energy consumption—into outputs such as crop yield and environmental performance improvements. DEA scores range from 0 to 1, where a score of 1.0 denotes full efficiency relative to the best-performing units on the frontier.

We implemented DEA (Data Envelopment Analysis) manually using linear programming (LP) with scipy.optimize (SciPy Community, Austin, TX, USA) to compute efficiency scores for UAV and conventional farms and the summary score. To ensure transparency and allow the replication of efficiency results,

Table 13. DEA input–output dataset for 18 farms (9 conventional + 9 UAV) and their corresponding efficiency scores.

Farm_ID	Method	Water (m3 per ha)	Pesticide (kg per ha)	Energy (kWh per ha)	Crop Yield (kg per ha)	DEA Score
Farm_1	Conventional	7800	3.2	45	5300	0.48
Farm_2	Conventional	7200	3	43	5200	0.5
Farm_3	Conventional	7600	3.1	42	5100	0.47
Farm_4	Conventional	8100	3.3	47	5400	0.47
Farm_5	Conventional	7400	2.9	40	5000	0.50
Farm_6	Conventional	9000	3.5	50	5500	0.45
Farm_7	Conventional	8500	3.4	48	5300	0.45
Farm_8	Conventional	7000	2.8	39	4900	0.50
Farm_9	Conventional	6800	3	41	5050	0.49
Farm_10	UAV	2500	1.9	22	5300	0.86
Farm_11	UAV	2700	1.7	20	5250	0.90
Farm_12	UAV	2300	1.6	18	5150	0.93
Farm_13	UAV	2600	1.8	21	5400	0.87
Farm_14	UAV	2100	1.5	17	5200	1
Farm_15	UAV	2900	2	24	5450	0.79
Farm_16	UAV	2400	1.8	19	5350	0.92
Farm_17	UAV	3100	2.1	23	5500	0.78
Farm_18	UAV	2200	1.6	18	5200	0.95

presents the raw input–output values and computed DEA efficiency scores for each of the 18 farms analyzed in this study. Farms 1–9 are conventional sprayer-operated fields, while Farms 10–18 are UAV-sprayed. The score distribution aligns with the visualizations: UAV farms consistently achieve DEA scores between 0.78 and 1.0, while conventional farms remain below 0.51, validating the robustness of DEA modeling.

This dataset underpins the entire DEA efficiency modeling. Notably, Farm_14, which achieved a perfect DEA score of 1.0, exemplifies the benchmark unit with the lowest input intensities and a high crop yield. This confirms that UAV farms not only reduce environmental burden but also optimize operational productivity, as established by multidimensional efficiency modeling.

The distribution of DEA efficiency scores across the 18 farms (9 UAV and 9 conventional) revealed significant differences between the two groups. As shown in Figure 5, the DEA frontier plot visualizes each farm’s performance in terms of water use (x-axis) and crop yield (y-axis), with the color scale denoting the DEA score. UAV farms clustered near the top-left quadrant—indicating higher yield per unit of water—were consistently more efficient. The red star marks the most efficient DMU, achieving the maximum output with the least water input, and representing a UAV-operated farm.

To further understand how input usage affects the DEA performance, scatter plots were constructed for each input metric against the DEA score as shown in Figure 6. Each subplot highlights a clear negative relationship between input use and efficiency. In the first subplot, water use is inversely related to the DEA score—farms consuming more water exhibit lower scores. The second and third subplots display similar trends for pesticide use and energy use, respectively. These visualizations confirm that UAV-operated farms, which consume fewer inputs per hectare, achieve higher efficiency outcomes. UAV farms show tighter clustering in the high-efficiency range. All three inputs exhibited a significant negative linear relationship with efficiency score—water (R² = 0.96, 95% CI for slope [−8.66 × 10⁻⁵, −6.99 × 10⁻⁵]), pesticide (R² = 0.97, 95% CI [−0.321, −0.265]), and energy (R² = 0.97, 95% CI [−0.0187, −0.0153])—with p < 0.01 in each case, confirming that lower input use is strongly associated with higher relative efficiency.

The boxplot analysis of DEA scores, as shown in Figure 7, quantitatively emphasizes the performance difference. UAV farms exhibit a median DEA score of 0.89 with an interquartile range (IQR) of 0.85–0.95, while conventional farms demonstrate a significantly lower median score of 0.48 with an IQR of 0.45–0.50. The UAV group showed no outliers, whereas conventional farms had several observations falling below the 25th percentile, suggesting performance variability and inefficiency in resource usage.

The frequency distribution of DEA scores as shown in Figure 8 further corroborates these findings. UAV farms cluster in the 0.85–1.0 efficiency range, while conventional farms predominantly fall between 0.45 and 0.55. No overlap exists in efficiency scores between the two groups, indicating a statistically and operationally significant separation in spraying performance.

To assess the robustness of our DEA results for input selection, we recalculated efficiency scores in four scenarios: using all inputs (base), and excluding water, pesticide, or energy individually. Prior to DEA computation, all inputs and outputs were normalized to the [0–1] range via min–max scaling to remove scale bias. We also tested the effect of omitting the single lowest-efficiency DMU; group mean efficiencies changed by <0.01, indicating negligible outlier influence. As shown in

Table 14. Sensitivity of DEA mean efficiency scores to input exclusion.

Scenario	Conventional Mean Efficiency	UAV Mean Efficiency
None (Base)	0.48	0.89
Excluding Water	0.48	0.88
Excluding Pesticide	0.39	0.88
Excluding Energy	0.48	0.88

Efficiency scores were recalculated (input-oriented CCR DEA) under four scenarios: base (water, pesticide, energy), and excluding each input in turn. Means shown over n = 9 farms per group.

, UAV-sprayed farms consistently maintain high mean efficiencies (≥0.88) across all scenarios, whereas conventional farms remain at ~0.48, demonstrating that the relative efficiency advantage of UAV spraying is not driven by any single input variable.

3.4. Intelligent Management Models (IMMs) for UAV Spraying Parameter Optimization

The Intelligent Management Model (IMM) simulations, along with field-derived spraying trials, provided a clear evaluation of how different UAV configurations influenced spray coverage and drift behavior. A total of five trial scenarios were designed using varying combinations of droplet size, spray volume, nozzle type, and flight speed. Each configuration was implemented in field-like conditions representative of drone spraying service operations in North Indian agricultural belts. These simulations were calibrated using operational field data obtained from drone spraying service providers (DSPs) in Punjab, Haryana, and Rajasthan, between 2022 and 2023.

Five UAV spraying trials were designed, covering a wide range of droplet sizes (150–250 μm), spray volumes (50–90 L/ha), and nozzle configurations (flat-fan, hollow-cone, variable-rate). The results of each IMM trial, in terms of canopy coverage and drift percentage, are summarized in

Table 15. IMM trial results: UAV spraying parameter optimization.

Trial ID	Nozzle Type	Droplet Size (μm)	Spray Volume (L/ha)	Flight Speed (m/s)	Coverage * (%)	Drift (%)
T1	Flat-Fan	200	60	3.5	86	15
T2	Hollow-Cone	150	80	2.0	90	12
T3	Variable-Rate	220	50	3.0	95	8
T4	Variable-Rate	250	70	2.5	92	10
T5	Flat-Fan	180	90	4.0	85	17

* Coverage percentages reported in Table 15 correspond directly to SCE values as defined in Equation (13).

. Across all trials, the variable-rate nozzle configurations consistently outperformed flat-fan and hollow-cone setups. The highest spray coverage was recorded in Trial T3, where a variable-rate nozzle operating at a droplet size of 220 μm and a spray volume of 50 L per hectare achieved 95% canopy coverage, with only 8% drift loss. This trial used a moderate flight speed of 3 m per second, which proved optimal for balancing uniform deposition with UAV battery endurance.

To provide a more rigorous quantitative basis to the IMM outputs, we conducted Multiple Regression Analysis using trial data capturing spray coverage (%) and spray drift (%) as dependent variables. Coverage was modeled as a function of droplet size and spray volume, while drift was modeled against droplet size and flight height. The fitted models yielded R² values of 0.81 for coverage and 0.72 for drift, with both models statistically significant (p < 0.05).

Table 16. Fitted regression models for spray coverage and drift.

Dependent Variable	Predictor	Coefficient (β)	SE	p-Value	R2
Spray Coverage (%)	Intercept	25.4	3.1	<0.001
	Droplet Size (μm)	+0.32	0.05	<0.001	0.81
	Spray Volume (L/ha)	+0.15	0.02	<0.001
Spray Drift (%)	Intercept	8.2	2.2	<0.001
	Droplet Size (μm)	−0.28	0.04	<0.001	0.72
	Flight Height (m)	+0.12	0.03	0.002

summarizes the regression coefficients and fit statistics. These results support the observed trends from the IMM optimization and strengthen the interpretability of Figure 9.

Regression surfaces (Figure 9) corroborate Table 16: variable-rate nozzles at 220–250 µm and 50–70 L ha⁻¹ deliver ≥ 95% coverage with ≤10% drift.

Trials situated in the lower quadrant of the drift axis (DPI < 10) aligned with the ideal coverage-to-drift ratios predicted by the IMM simulation. The model’s predictive accuracy validated its potential to assist UAV operators in adjusting parameters proactively based on crop type, environmental conditions, and spray objectives.

Overall, the IMM framework demonstrated that a variable-rate nozzle operating in the 220–250 μm droplet range, combined with a spray volume of 50–70 L/ha and a flight speed of 2.5–3.0 m/s, offers the most efficient and environmentally responsible configuration for UAV-based spraying in Indian agriculture. These findings validate the IMM’s ability to proactively recommend UAV settings that maximize pesticide deposition while minimizing environmental risks. The integration of field-tuned parameters into this decision-support model offers real-time adaptability, making it suitable for use as a smart spraying assistant module in UAV firmware or operator dashboards.

Ultimately, IMM-based optimization highlights the operational feasibility and agronomic advantages of adaptive nozzle technologies in Indian UAV spraying. When used under optimal conditions, these configurations can significantly reduce pesticide waste and off-target drift—reinforcing UAVs as a sustainable precision agriculture tool in India.

3.5. Multi-Criteria Decision Analysis (MCDA)

To evaluate the most impactful policy pathways for enhancing the adoption of UAV spraying technologies in Indian agriculture, a Multi-Criteria Decision Analysis (MCDA) was conducted. This framework incorporated four criteria reflecting both stakeholder priorities and documented drivers of precision agriculture adoption: economic feasibility, adoption feasibility, environmental impact, and government support readiness. The weight assigned to each criterion, based on stakeholder consultations and policy literature, is as follows: economic feasibility (0.30), adoption feasibility (0.25), environmental impact (0.25), and government support (0.20). These weights were also reflected in the scoring framework applied to each policy. Each policy was scored across these dimensions based on weighted importance values and stakeholder-informed utility estimates.

Three leading policy interventions were evaluated within this framework. The first included government subsidy programs currently operational under the Sub-Mission on Agricultural Mechanization (SMAM) and NABARD-backed financial support structures. The second involved the Drone Didi Scheme, a women-centric training and employment program that aims to improve operator availability and community ownership of drone technologies. The third policy intervention pertained to the regulatory fast-tracking of Drone Service Providers (DSPs) under India’s Drone Rules (2021), which facilitate licensing and service delivery authorizations.

As presented in

Table 17. MCDA-based policy effectiveness scores for UAV adoption interventions.

Policy Option	Economic Feasibility (0.30)	Adoption Feasibility (0.25)	Environmental Impact (0.25)	Govt. Support Readiness (0.20)	Final Score
Government Subsidy (SMAM/NABARD)	2.70 (9 × 0.30)	2.00 (8 × 0.25)	2.25 (9 × 0.25)	1.60 (8 × 0.20)	8.55
Drone Didi Scheme (PIB, 2023)	2.40 (8 × 0.30)	2.25 (9 × 0.25)	1.75 (7 × 0.25)	1.80 (9 × 0.20)	8.20
Regulatory Fast-Tracking for DSPs	2.10 (7 × 0.30)	1.75 (7 × 0.25)	2.00 (8 × 0.25)	1.40 (7 × 0.20)	7.25

, the highest total policy score was recorded for the government subsidy option, with a composite effectiveness score of 8.55. This is attributed to strong alignment across economic and environmental criteria. The Drone Didi Scheme followed closely with a score of 8.20, demonstrating high performance in social feasibility and institutional support. The Regulatory Fast-Tracking policy option, although beneficial for easing entry for private service providers, yielded a lower score of 7.25, primarily due to limited direct economic incentives for farmers and weaker environmental co-benefits.

The radar plot (Figure 10) mirrors Table 17: subsidies dominate, Drone Didi is next, and regulatory fast-tracking ranks third.

This visual confirms the numerical findings and emphasizes that a hybrid policy strategy—combining direct financial support with community-level capacity building and training programs—is likely to yield the most sustained and equitable adoption of UAV spraying technologies in Indian agriculture.

The results clearly indicate that economic interventions, particularly capital cost subsidies, have the highest influence on technology adoption, especially among small and marginal farmers. Moreover, the incorporation of gender-responsive models, such as Drone Didi, has emerged as a promising complementary approach, not only addressing labor gaps but also enhancing inclusivity and local ownership. While streamlining of regulatory procedures remains important, its isolated impact may be limited without simultaneous fiscal and capacity-building measures. These findings provide empirical support for integrated policy frameworks that combine financial, institutional, and social levers for sustainable UAV deployment in agriculture.

4. Conclusions and Discussion

This study presented a comprehensive evaluation of UAV-based spraying systems in Indian agriculture by integrating multiple empirical and computational methodologies. The evidence across all analytical layers consistently highlights that UAV spraying offers considerable operational and environmental advantages over conventional tractor-mounted or manual pesticide application systems.

In terms of resource efficiency, UAV operations demonstrated a 67.9% reduction in water use, 41.0% reduction in pesticide application, and 54.3% reduction in energy consumption relative to conventional practices. These differences were statistically significant, as confirmed by t-tests and regression analyses, and are particularly relevant for water-stressed agricultural belts such as Punjab, Haryana, and Rajasthan. The corresponding impact on groundwater conservation was substantial, with UAV-based methods potentially alleviating 50–70% of the annual aquifer overdraft per hectare.

From an environmental standpoint, the Life Cycle Assessment (LCA) established that UAV spraying significantly reduces the total carbon footprint. This was supported by Monte Carlo simulations, which revealed 95% confidence intervals consistently favoring UAV systems across CO₂ emissions, water footprint, and pesticide runoff potential. The Tornado Sensitivity Analysis further identified diesel usage, spray volume, and pesticide load as the most influential environmental levers, suggesting that optimization in these areas can further enhance sustainability outcomes.

Efficiency benchmarking using Data Envelopment Analysis (DEA) revealed that UAV-operated farms consistently outperformed conventional ones, achieving higher technical efficiency scores even under similar yield conditions. This indicates that UAVs not only reduce input waste but also maintain or improve agronomic output. Moreover, the Intelligent Management Model (IMM) validated that nozzle configurations, droplet size, and spray volume could be precisely calibrated to minimize off-target drift and maximize canopy coverage. Trials using variable-rate nozzles achieved 95% coverage with <10% drift, demonstrating the precision capabilities of modern UAV systems.

In terms of policy interventions, the Multi-Criteria Decision Analysis (MCDA) revealed that financial subsidy schemes and the Drone Didi program were most effective in supporting scalable adoption. Subsidies emerged as the strongest enabler due to their role in offsetting capital cost burdens, while the Drone Didi initiative offered gender-responsive employment and decentralized service delivery. The radar chart comparison further reinforced the complementarity of these approaches, suggesting that integrated policy strategies hold the greatest potential for long-term UAV deployment.

Taken together, the results affirm that UAV spraying, when deployed through optimized parameters and supported by enabling policies, represents a scalable and sustainable solution for addressing India’s agricultural input challenges. The convergence of environmental efficiency, operational optimization, and socio-economic feasibility positions UAV technology as a critical lever for climate-smart agriculture in India.

While this integrated analysis leverages multiple quantitative frameworks, several limitations warrant consideration. First, the sample size of 18 decision-making units may not capture the full heterogeneity of Indian agro-climatic and operational contexts. We are currently coordinating with additional Drone Service Providers and state extension agencies to scale the dataset to over 100 farms across five states and three cropping seasons, which will enable a more robust statistical validation of the findings. Second, field conditions such as wind speed, humidity, and canopy structure were not systematically controlled, introducing variability into spray efficiency measures. Third, the LCA boundary was limited to the use phase, omitting upstream (manufacturing) and downstream (decommissioning) impacts, which constrains the full cradle-to-grave assessment. Fourth, our sensitivity analysis used a one-at-a-time (OAT) approach and did not model interactions; future work should adopt global sensitivity methods (e.g., Sobol indices or factorial designs) to capture parameter interdependencies. Finally, the cross-sectional design prevents the evaluation of temporal dynamics. Future research should address these gaps by expanding the DMU dataset, incorporating real-time environmental sensing for adaptive IMM calibration, extending the LCA to a full life cycle scope, and conducting longitudinal multi-season trials to validate and generalize the findings across diverse cropping systems.

Overall, despite the limitations, this work demonstrates the promise of UAV spraying for climate-smart agriculture in India and provides a robust, data-driven foundation for future operational and policy development.