Leguminous forage yield estimation method based on Meta analysis and unmanned aerial vehicle multispectral remote sensing

Abstract

The invention relates to the technical field of grassland agricultural vegetation, and provides a leguminous forage grass yield estimation method based on Meta analysis and unmanned aerial vehicle multispectral remote sensing, which comprises the following steps: step 1, Meta analysis data integration and gradient design: retrieving and screening alfalfa field test data through a literature database, extracting irrigation amount, fertilization amount and yield data, and carrying out Meta analysis data integration and gradient design; a Meta analysis method is adopted, a response ratio (RR) and a confidence interval are calculated, an influence rule of different water and fertilizer gradients on the yield is determined, an irrigation gradient and a fertilization gradient are divided based on a Meta analysis result, and a water and fertilizer cross treatment scheme is designed. 20-year literature data are integrated through Meta analysis, irrigation and fertilization gradients are standardized, regional heterogeneity interference is solved, a water and fertilizer threshold effect is defined and combined with a nonlinear optimization algorithm, a water and fertilizer-yield response model is established, the optimal irrigation amount and fertilization amount are accurately predicted based on an unmanned aerial vehicle multispectral remote sensing technology, and production capital investment is reduced by 10%-15%.

Description

Technical Field

[0001] The present invention relates to the technical field of grassland agricultural vegetation, and in particular to a method for estimating leguminous forage yield based on meta-analysis and unmanned aerial vehicle multispectral remote sensing. Background Art

[0002] Alfalfa (Medicago sativa) is an important global legume forage, and its yield directly affects livestock development and food security. However, traditional water and fertilizer management models have significant limitations:

[0003] Data dispersion and regional differences: Existing research is often based on single field trials or small-scale experiments, lacking systematic integration of multi-regional and multi-climate conditions. This results in limited universal applicability of water and fertilizer management solutions. For example, different studies have reached different and even contradictory conclusions on the relationship between irrigation volume, fertilization type, and yield (e.g., nitrogen fertilizer has a significant yield-increasing effect in some studies, but not in others).

[0004] Resource waste and environmental pressure: While blindly increasing water and fertilizer inputs can boost yields in the short term, it can easily lead to soil degradation, nutrient loss, and greenhouse gas emissions. Studies have shown that excessive irrigation (e.g., WH gradients) increases yield but reduces water use efficiency (IWUE), while excessive fertilization (e.g., NH gradients) can inhibit alfalfa nutritional quality.

[0005] Monitoring technology is lagging behind: Traditional yield estimation relies on manual sampling, which is time-sensitive and costly. Although remote sensing technology has been applied to agriculture, it is mostly limited to single spectral index analysis and lacks dynamic coupling with multi-source data (such as water and fertilizer gradients and soil properties).

[0006] While existing meta-analysis techniques can integrate multiple research results, they haven't yet been fully integrated with multispectral remote sensing using drones. Furthermore, drone technology is primarily used for vegetation coverage monitoring, and no specific prediction models for water and fertilizer responses in legumes have been developed. Therefore, a technology that integrates global data analysis with real-time monitoring is urgently needed to achieve precise and sustainable forage production management. Summary of the Invention

[0007] In response to the shortcomings of the existing technology, the present invention provides a legume forage yield estimation method based on meta-analysis and unmanned aerial vehicle multispectral remote sensing. It solves the problem in the existing technology that although meta-analysis can integrate multiple research results, it has not yet been deeply integrated with unmanned aerial vehicle multispectral remote sensing; and unmanned aerial vehicle technology is mostly used for vegetation coverage monitoring, and no water and fertilizer response prediction model for legume forage has been formed.

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions: a method for estimating legume forage yield based on meta-analysis and unmanned aerial vehicle multispectral remote sensing, comprising the following steps:

[0009] Step 1: Meta-analysis data integration and gradient design:

[0010] We searched and screened alfalfa field trial data from literature databases, extracted irrigation, fertilization, and yield data, and used a meta-analysis method to calculate response ratios (RRs) and confidence intervals to determine the effects of different water-fertilizer gradients on yield. Based on the meta-analysis results, we divided irrigation gradients into fertilization gradients and designed a water-fertilizer crossover treatment plan.

[0011] Step 2: Field experiment and multispectral data acquisition:

[0012] Irrigation, fertilization, and water-fertilizer cross-plots were set up in the experimental area. The speed and angle of the sprinkler were controlled to achieve gradient processing. A multispectral drone was used to take aerial photos of alfalfa during the greening, branching, and early flowering stages, collecting data in blue, green, red, red edge, and near-infrared bands. The drone images were radiometrically corrected using a radiation correction plate and spatially matched with the measured yield data on the ground.

[0013] Step 3: Construction and optimization of yield forecast model:

[0014] Combining the water and fertilizer response patterns from the meta-analysis with multispectral data, vegetation indices were extracted and an alfalfa yield prediction model was constructed. A nonlinear optimization algorithm was used to adjust the model parameters, verify the model accuracy, and screen the optimal water and fertilizer management model.

[0015] Step 4: Practical application and dynamic monitoring:

[0016] Apply the optimized model to actual production, monitor grass growth in real time, dynamically adjust water and fertilizer strategies, and achieve efficient resource utilization.

[0017] Preferably, the calculation formula of the response ratio (RR) in step 1 is:

[0018]

[0019] in, is the mean yield of the experimental group, is the mean yield of the control group;

[0020] The response ratio variance (v) is calculated as:

[0021]

[0022] The weighted mean effect size (RR ++ ) and 95% confidence intervals were used to judge the significance of water and fertilizer treatments.

[0023] Preferably, the irrigation gradient division in step 1 is based on the dryness ratio method:

[0024]

[0025] Among them, W ck / mi is the irrigation amount, P is the annual precipitation, and PET is the annual potential evapotranspiration;

[0026] Press r ck / mi The irrigation gradient was divided into: WL (≤0.25), WM (0.25-0.3), WMH (0.3-0.35), and WH (>0.35).

[0027] Preferably, the water-fertilizer cross treatment in step 1 includes 9 combinations: WLFL, WLFM, WLFH, WMFL, WMFM, WMFH, WMHFL, WMHFM, and WMHFH, wherein WL, WM, and WMH are irrigation gradients, and FL, FM, and FH are fertilization gradients.

[0028] Preferably, the drone aerial photography parameters in step 2 include: an altitude of 20 meters, a speed of 5 meters per second, a heading and lateral overlap of 75%, and an image ground resolution of 1 centimeter per pixel.

[0029] Preferably, the nonlinear optimization in step three adopts a genetic algorithm or a particle swarm optimization algorithm, and the objective function is to minimize the root mean square error between the predicted yield and the measured yield.

[0030] Preferably, the vegetation index in step 3 includes:

[0031] Normalized Difference Vegetation Index (NDVI), the calculation formula is:

[0032] The calculation formula of Enhanced Vegetation Index (EVI) is:

[0033] The red edge chlorophyll index (RECI) is calculated as follows:

[0034] Preferably, the heterogeneity test in the meta-analysis in step 3 adopts the Q test, and if P < 0.05, a random effects model is used, otherwise a fixed effects model is used.

[0035] Preferably, when dynamically adjusting the water and fertilizer strategy in step 4, real-time drone multispectral data and meteorological data are combined to predict the optimal combination of irrigation and fertilization amounts through a model.

[0036] This invention provides a method for estimating legume forage yield based on meta-analysis and unmanned aerial vehicle multispectral remote sensing. It has the following beneficial effects:

[0037] 1. This invention integrates 20 years of literature data through meta-analysis, standardizes irrigation and fertilization gradients, addresses regional heterogeneity interference, clarifies the water and fertilizer threshold effects and combines them with nonlinear optimization algorithms to establish a water-fertilizer-yield response model. Based on drone multispectral remote sensing technology, it accurately predicts the optimal irrigation and fertilization amounts, reducing production capital investment by 10% to 15%.

[0038] 2. This invention uses multispectral drones to collect data during the critical growth period and combines it with vegetation indices to invert alfalfa growth in real time. The model prediction accuracy is improved to RMSE ≤ 0.8 Mg / ha. Dynamic coupling with meteorological data enables real-time adjustment of water and fertilizer strategies to ensure stable alfalfa yields in arid areas.

[0039] 3. The optimized water and fertilizer management model of the present invention reduces nitrogen and phosphorus loss by 20% to 30%, and reduces the risk of soil salinization. After enterprises apply this technology, alfalfa hay yields increase by 18% to 25%, while saving water and fertilizer costs by 15% to 20%, contributing to green agriculture and carbon emission reduction goals. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 Flowchart of the present invention. DETAILED DESCRIPTION

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0042] Please see the attached Figure 1 The embodiment of the present invention provides a method for estimating legume forage yield based on meta-analysis and unmanned aerial vehicle multispectral remote sensing, comprising the following steps:

[0043] Step 1: Meta-analysis data integration and gradient design:

[0044] Alfalfa field trial data were retrieved and screened through literature databases. Irrigation, fertilization, and yield data were extracted. Meta-analysis was used to calculate response ratios (RRs) and confidence intervals. The effects of different water-fertilizer gradients on yield were determined. Based on the meta-analysis results, irrigation gradients and fertilization gradients were divided, and a water-fertilizer crossover treatment plan was designed. The response ratio (RR) was calculated as follows:

[0045]

[0046] in, is the mean yield of the experimental group, is the mean yield of the control group;

[0047] The response ratio variance (v) is calculated as:

[0048]

[0049] The weighted mean effect size (RR ++ ) and 95% confidence interval to judge the significance of water and fertilizer treatments;

[0050] In addition, the water-fertilizer crossover treatment includes 9 combinations: WLFL, WLFM, WLFH, WMFL, WMFM, WMFH, WMHFL, WMHFM, and WMHFH. WL, WM, and WMH are irrigation gradients, and FL, FM, and FH are fertilization gradients. The irrigation gradient division is based on the dryness ratio method:

[0051]

[0052] Among them, W ck / mi is the irrigation amount, P is the annual precipitation, and PET is the annual potential evapotranspiration;

[0053] Press r ck / mi The values ​​divided the irrigation gradient into: WL (≤0.25), WM (0.25-0.3), WMH (0.3-0.35), and WH (>0.35);

[0054] Step 2: Field experiment and multispectral data acquisition:

[0055] Irrigation, fertilization, and cross-fertilization plots were set up in the experimental area. The speed and angle of the sprinkler were controlled to achieve gradient processing. A multispectral drone was used to conduct aerial photography of alfalfa during the greening, branching, and early flowering stages. The drone photography parameters included an altitude of 20 meters, a speed of 5 meters per second, a heading and lateral overlap of 75%, and an image ground resolution of 1 cm / pixel. Data in the blue, green, red, red edge, and near-infrared bands were collected. The drone images were radiometrically corrected using a radiation correction plate and spatially matched with the measured yield data on the ground.

[0056] Step 3: Construction and optimization of yield forecast model:

[0057] Combining the water and fertilizer response patterns with multispectral data in the meta-analysis, the Q test was used for heterogeneity testing in the meta-analysis. If P < 0.05, a random effects model was used; otherwise, a fixed effects model was used. Vegetation indices were extracted and an alfalfa yield prediction model was constructed. The model parameters were adjusted using a nonlinear optimization algorithm to verify the model accuracy and screen the optimal water and fertilizer management mode. The nonlinear optimization used a genetic algorithm or a particle swarm optimization algorithm. The objective function was to minimize the root mean square error between the predicted yield and the measured yield. Vegetation indices included:

[0058] Normalized Difference Vegetation Index (NDVI), the calculation formula is:

[0059] The calculation formula of Enhanced Vegetation Index (EVI) is:

[0060] The red edge chlorophyll index (RECI) is calculated as follows:

[0061] Step 4: Practical application and dynamic monitoring:

[0062] The optimized model is applied to actual production to monitor grass growth in real time, dynamically adjust water and fertilizer strategies, and achieve efficient resource utilization. When dynamically adjusting water and fertilizer strategies, real-time drone multispectral data and meteorological data are combined to predict the optimal combination of irrigation and fertilization amounts through the model.

[0063] The following is an introduction with reference to specific embodiments:

[0064] Example:

[0065] 1. Experimental area and conditions

[0066] The experimental area has an arid desert climate with an average annual precipitation of 59.8 mm and evaporation of 2560 mm. The soil is sandy loam. Pre-experimental soil physical and chemical properties at the 0-40 cm layer were as follows: organic matter 0.54%-0.87%, available phosphorus 13.12-41.45 mg / kg, available potassium 113.55-167.21 mg / kg, and pH 8.03-8.15. Data on 1431 items, including irrigation (W), fertilizer application (N, P₂O₅, K₂O), and hay production, were collected and used to construct an Excel database.

[0067] 1. Meta-analysis response ratio calculation:

[0068] The response ratio (RR) and confidence interval (CI) were calculated according to the formula:

[0069]

[0070] in is the mean yield of the experimental group, is the mean yield of the control group, and the weighted average effect value (RR ++ ) showed that medium-high water (WMH) irrigation increased yield by 42.5% (95% CI 38.2%-46.8%), and the combination of medium nitrogen (NM) and medium phosphorus (PM) increased yield by 55.6%.

[0071] 2. Water and fertilizer gradient division:

[0072] Irrigation gradient: based on the aridity ratio method Set 4 gradients: WL (480mm), WM (600mm), WMH (720mm), WH (840mm);

[0073] Fertilization gradient: compound fertilizer (N-P2O5-K2O=15-15-15) was set at four gradients: FL (466.5 kg / ha), FM (606 kg / ha), FMH (757.5 kg / ha), and FH (909 kg / ha);

[0074] Water and fertilizer cross-treatment: 9 combinations were designed (such as WLFL, WMFM, WMHFH, etc.).

[0075] 2. Field Experiment and Data Collection

[0076] 1. Sample site setting:

[0077] The experimental area was divided into three circular plots (radius 270-300 m), which were used for irrigation, fertilization and water-fertilizer cross-treatment respectively.

[0078] Each treatment was set up with 6 3m×3m sample plots, repeated 6 times, for a total of 1200 sample plots.

[0079] 2. UAV multispectral data acquisition:

[0080] A DJI P4 Multispectral drone was used, equipped with blue (450nm), green (560nm), red (650nm), red-edge (730nm), and near-infrared (840nm) sensors.

[0081] Aerial photography parameters: altitude 20m, speed 5m / s, overlap 75%, ground resolution 1cm / pixel.

[0082] Data collection time: greening period (April 26), branching period (June 21), and early flowering period (August 31). Before each flight, 25%, 50%, and 75% reflectivity calibration plates were laid for radiation correction.

[0083] 3. Ground-based measured data:

[0084] Hay yield (Mg / ha) was measured after each mowing, and soil moisture content (gravimetric method), chlorophyll content (SPAD-502 instrument), and meteorological data (precipitation and evapotranspiration) were recorded simultaneously.

[0085] 3. Construction and Verification of Yield Forecast Model

[0086] Vegetation index extraction:

[0087] 1. Calculate NDVI, EVI, and RECI:

[0088]

[0089] Extract the mean vegetation index of the area corresponding to the sample plot.

[0090] 2. Model construction:

[0091] Input variables: vegetation index (NDVI, EVI, RECI), water and fertilizer gradients (WL, WM, etc.), meteorological data (evapotranspiration).

[0092] Model selection: Random Forest Regression algorithm, the target variable is hay production.

[0093] Parameter optimization: Genetic algorithm is used to adjust hyperparameters (tree depth, number of leaf nodes), and the objective function is to minimize RMSE.

[0094] 3. Model Validation

[0095] The ratio of training set to test set is 8:2.

[0096] Results: The model prediction RMSE was 0.72 Mg / ha, and the coefficient of determination (R 2 The optimal water-fertilizer system was WMH+FMH (720 mm irrigation + 757.5 kg / ha fertilizer), which increased yield by 18.3%, saved 7.7% in water, and 10.1% in fertilizer compared to the company's original system (780 mm irrigation + 681.75 kg / ha).

[0097] IV. Practical Application and Benefit Analysis

[0098] 1. Dynamic control:

[0099] By combining real-time drone data with weather forecasts, irrigation plans are updated every 15 days. For example, if evapotranspiration is predicted to increase by 10% in July 2024, the system will automatically adjust irrigation rates from 720mm to 690mm, avoiding water waste.

[0100] 2. Environmental and economic benefits:

[0101] 3. Resource conservation: water saving 12.5% ​​(annual irrigation water saving 1.2×10 5 m 3 ), saving 15.2% of fertilizer (reducing nitrogen and phosphorus loss by 28%).

[0102] 4. Yield increase: The average annual hay yield increased from 9.8 Mg / ha to 11.6 Mg / ha, and the company's annual revenue increased by approximately RMB 2.3 million.

[0103] 5. Carbon emission reduction: Reduce N2O emissions by approximately 1.2t CO2-eq / ha through optimized fertilization.

[0104] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for estimating legume forage yield based on meta-analysis and unmanned aerial vehicle multispectral remote sensing, characterized in that: The following steps are involved: Step 1: Meta-analysis data integration and gradient design: We searched and screened alfalfa field trial data from literature databases, extracted irrigation, fertilization, and yield data, and used a meta-analysis method to calculate response ratios (RRs) and confidence intervals to determine the effects of different water-fertilizer gradients on yield. Based on the meta-analysis results, we divided irrigation gradients into fertilization gradients and designed a water-fertilizer crossover treatment plan. Step 2: Field experiment and multispectral data acquisition: Irrigation, fertilization, and water-fertilizer cross-plots were set up in the experimental area. The speed and angle of the sprinkler were controlled to achieve gradient processing. A multispectral drone was used to take aerial photos of alfalfa during the greening, branching, and early flowering stages, collecting data in blue, green, red, red edge, and near-infrared bands. The drone images were radiometrically corrected using a radiation correction plate and spatially matched with the measured yield data on the ground. Step 3: Construction and optimization of yield forecast model: Combining the water and fertilizer response patterns from the meta-analysis with multispectral data, vegetation indices were extracted and an alfalfa yield prediction model was constructed. A nonlinear optimization algorithm was used to adjust the model parameters, verify the model accuracy, and screen the optimal water and fertilizer management model. Step 4: Practical application and dynamic monitoring: Apply the optimized model to actual production, monitor grass growth in real time, dynamically adjust water and fertilizer strategies, and achieve efficient resource utilization.

2. The method for estimating legume forage yield based on meta-analysis and unmanned aerial vehicle multispectral remote sensing according to claim 1, characterized in that: The calculation formula of the response ratio (RR) in the step 1 is: in, is the mean yield of the experimental group, is the mean yield of the control group; The response ratio variance (v) is calculated as: The weighted mean effect size (RR ++ ) and 95% confidence intervals were used to judge the significance of water and fertilizer treatments.

3. The method for estimating legume forage yield based on meta-analysis and unmanned aerial vehicle multispectral remote sensing according to claim 1, characterized in that: The irrigation gradient division in step 1 is based on the aridity ratio method: Among them, W ck / mi is the irrigation amount, P is the annual precipitation, and PET is the annual potential evapotranspiration; Press r ck / mi The irrigation gradient was divided into: WL (≤0.25), WM (0.25-0.3), WMH (0.3-0.35), and WH (>0.35).

4. The method for estimating legume forage yield based on meta-analysis and unmanned aerial vehicle multispectral remote sensing according to claim 1, characterized in that: The water-fertilizer crossover treatment in step 1 includes 9 combinations: WLFL, WLFM, WLFH, WMFL, WMFM, WMFH, WMHFL, WMHFM, and WMHFH, where WL, WM, and WMH are irrigation gradients, and FL, FM, and FH are fertilization gradients.

5. The method for estimating legume forage yield based on meta-analysis and unmanned aerial vehicle multispectral remote sensing according to claim 1, characterized in that: The drone aerial photography parameters in step 2 include: an altitude of 20 meters, a speed of 5 meters per second, a heading and lateral overlap of 75%, and an image ground resolution of 1 centimeter per pixel.

6. The method for estimating legume forage yield based on meta-analysis and unmanned aerial vehicle multispectral remote sensing according to claim 1, characterized in that: The nonlinear optimization in step three adopts a genetic algorithm or a particle swarm optimization algorithm, and the objective function is to minimize the root mean square error between the predicted yield and the measured yield.

7. The method for estimating legume forage yield based on meta-analysis and unmanned aerial vehicle multispectral remote sensing according to claim 1, characterized in that: The vegetation index in step 3 includes: Normalized Difference Vegetation Index (NDVI), the calculation formula is: The calculation formula of Enhanced Vegetation Index (EVI) is: The red edge chlorophyll index (RECI) is calculated as follows:

8. The method for estimating legume forage yield based on meta-analysis and unmanned aerial vehicle multispectral remote sensing according to claim 1, characterized in that: In the meta-analysis in step 3, the Q test was used for heterogeneity test. If P < 0.05, the random effect model was used; otherwise, the fixed effect model was used.

9. The method for estimating legume forage yield based on meta-analysis and unmanned aerial vehicle multispectral remote sensing according to claim 1, characterized in that: When dynamically adjusting the water and fertilizer strategy in step 4, the optimal combination of irrigation and fertilization amounts is predicted through a model by combining real-time drone multispectral data and meteorological data.

Images