"A Method for Creating an Agricultural Sustainability Index Through Artificial Intelligence-Based Fusion of Satellite, Soil, and Meteorological Data"

Abstract

This invention relates to the generation of a Unified Agricultural Sustainability Index (Fusion_Index) using an artificial intelligence-based data fusion method that combines spectral indices derived from multiband satellite images (NDVI, NDWI, SAVI, EVI, etc.), laboratory-based soil analysis parameters (pH, organic matter, macro / micro nutrient elements, electrical conductivity), and meteorological data (temperature, precipitation, relative humidity). It is evaluated in the technical fields of precision agriculture, remote sensing, and environmental data science. In current applications, soil and satellite data are analyzed separately, resulting in fragmented sustainability status and creating uncertainty in input optimization and yield estimation decisions. This invention solves this technical problem using Principal Component Analysis (PCA) with dimensionality reduction and a random forest or multilayer artificial neural network, integrating individual datasets into a highly accurate composite score.The resulting Fusion_Index is classified into "Poor-Medium-Good" categories based on 33% and 66% percentile thresholds and displayed color-coded on a GIS-based map. The system consists of data collection, pre-processing, PCA, machine learning, and visualization layers. This allows for real-time monitoring of sustainability status at the field level, and enables the optimization of variable rate fertilization and irrigation strategies on a scientific basis. Furthermore, annual updates of the Fusion_Index model ensure adaptation to changing field conditions over time.

Description

1 TARIFF AI-BASED ANALYSIS OF SATELLITE, SOIL, AND METEOROLOGICAL DATA. CREATING AN AGRICULTURAL SUSTAINABILITY INDEX THROUGH FUSION METHOD I. Technical Area This discovery is based on satellite imagery, laboratory soil analysis, and meteorological data. By combining observations with AI-based data fusion, it achieves fine spatial resolution. A method for creating a "Combined Index of Agricultural Sustainability" It includes inventions in precision agriculture, remote sensing, environmental data analytics, and decision support. They are included in the systems disciplines. II. State of the Art 1. Individual spectral indices (e.g., NDVI) provide only a superficial assessment of plant health. It shows, but is incomplete because it does not include soil chemistry or physiology. 15 2. Laboratory soil analyses offer high accuracy at specific points, but not continuously. It does not provide spatial coverage. 3. Current data aggregation approaches mostly rely on simple statistical correlations. It is limited and does not utilize the full potential of machine learning. These shortcomings necessitate a holistic approach to fertilization, irrigation, and field management decisions. This makes sustainability assessment more difficult. III. Purposes of the Invention 1. By developing a fused quantitative index (Fusion_Index), the land To accurately define the level of sustainability. 25 2. Agricultural decision-making by providing real-time spatial maps layered as “Poor – Medium – Good”. To strengthen support processes. 3. Combining free satellite data with a limited number of soil analyses for long-term analysis. To reduce monitoring costs. 2 IV. Summary of the Invention The system consists of four main layers: 1. Data Collection Layer i. Soil parameters (pH, organic matter, NPK, microelements, EC). ii. NDVI, NDWI, SAVI, EVI 5 derived from Sentinel-2 / Landsat imagery. etc. indices. iii. Recent meteorological data (temperature, precipitation, relative humidity). 2. Pre-processing Layer i. Point-pixel mapping (GPS at the cm level). ii. Outlier removal, median-based missing data completion. 10 iii. Z-Score normalization. 3. Dimensional Reduction and Fusion Layer i. Principal Component Analysis (PCA) to preserve ≥ 95% of the variance. ii. Fusion vector in Random Forest or Multilayer Neural Network to be given as input to the model. 15 4. Prediction and Classification Layer i. A continuous Fusion_Index output is obtained. ii. Three-tier classification is performed using 33% and 66% percentile thresholds. iii. Results are displayed in a Folium / Leaflet-based interactive map interface. It is visualized. 20 V. Detailed Description of the Units Module Component Basic Function A Soil Data Loader Full productivity in time-space stamped CSV. It imports its parameters. B Satellite Data Uploader Sentinel-2 SR collection cloud mask < 20% It calls it using a filter. C Index Calculator NDVI, NDWI, SAVI, EVI, MSAVI, NDMI etc. indices It produces at m resolution. 3 D Synchronization Module Each ground point is paired with the nearest pixel of the same date. It associates them. The EPCA module reduces 20+ variables to 1-2 main components. FML Module 200-tree Random Forest; alternatively, 3-tiered MLP (ReLU). G Classifier automatically colors the map using quantitative thresholding. H Visualization Interface Value display via interactive map and pop-up window. provides. VI. Best Practice In a pilot trial conducted on 200 hectares of rehabilitated land, PCA component one + MLP The combination of the outputs mapped land productivity with an accuracy of R² = 0.87. VII. Industrial Applicability • Supports variable rate fertilization (VRT) schedules. • It can reduce chemical fertilizer use by approximately 25%. • It provides a data source for carbon credit and ESG certification platforms. VIII. List of Drawings 1. Figure 1 – System block diagram (layers A→H). Figure 2 – PCA and machine learning workflow. Figure 3 – Example Classification Map IX. Industrial Application The system uses cloud-based environments such as Google Earth Engine or large-scale farms. It can be integrated into local servers; Field Management Systems via REST API. Data sharing is possible with (FMS). 20

Claims

4 REQUESTS 1. It is a method; At least one spectral index data point (NDVI) derived from multiband satellite imagery. NDWI, SAVI, EVI or similar), at least one soil analysis from a laboratory 5 parameters (pH, organic matter, macro- / micro nutrient elements, electrical conductivity or similar) and at least one meteorological parameter (temperature, precipitation, relative humidity) (i) the main dataset containing multiple datasets (humidity or similar) for dimensionality reduction purposes (ii) Converting it into at least one component by applying Component Analysis (PCA); then 10 is used as input to the random forest regressor or artificial neural network model. by providing the data, a combined agricultural sustainability score (Fusion_Index) is created. It is a method that includes.

2. The Fusion_Index value obtained according to the method defined in Claim 1, with a 33% margin. and according to 66th percentile thresholds, into three classes: “Poor / Medium / Good” The method by which it separated. 15 3. Method according to claim 1 or 2; using Sentinel-2 or Landsat as satellite data. NDVI, NDWI, SAVI and bandwidths calculated from the bandwidths obtained from the platforms. EVI is characterized by the use of at least two of the spectral indices.

4. Method according to claims 1-3; soil parameters include pH, organic matter, nitrogen, At least two of the following data points: phosphorus, potassium and / or electrical conductivity (20) It is characterized by its use.

5. It is a system consisting of: (i) Data Collection Module, (ii) Pre-processing and Dimensionality Reduction Module, (iii) Machine Learning Module and (iv) Classification and Visualization The module contains and automatically executes the method defined in Claims 1-4. It is a system. 25 6. The system in claim 5; the classified Fusion_Index score is layered on the map. Showing color-coded indicators (Weak-Red, Medium-Orange, Good-Green), web-based system where the user can query values ​​by clicking on a specific location It is characterized by an interactive map interface.

7. The method or system specified in Claims 1-6; the resulting class is classified as agricultural 30 in management decisions (variable rate fertilization, irrigation optimization, crop (crop rotation planning) characterized by a decision-support function that directly utilizes it. It is done.

8. The method described in Claims 1-7 is designed to improve model performance. Audited using historical performance records associated with Fusion_Index. conducting the re-training step periodically (at least once a year) It is characterized by including an update routine. 10 20

Images