Enhanced Management Zones for Precision Agriculture

Abstract

The present invention is a system and method for delineating agricultural management zones over a broad geographic area without overly local, field-specific data. The current innovation guides precision agriculture sampling and management by delineating enhanced management zones based on remote sensing and artificial intelligence and combining the two with data derived from existing national soil survey databases. In an embodiment, the current innovation uses artificial intelligence from multiple sources to provide granular area details. The output of this innovation can be aggregated to produce management zone sizes with an uncertainty level that is adaptable to the needs of the customer farmer and implementable given the capabilities of the available equipment.

Description

[0001] Priority Claim

[0002] This non - provisional application claims the benefit of partial continuation of non - provisional application 16 / 699,292, entitled "Enhanced Management Zones for Precision Agriculture", filed on November 29, 2019, which is hereby incorporated by reference in its entirety.

[0003] Copyright Notice

[0004] A portion of the disclosure of this patent document contains material which is subject to copyright protection. The copyright owner has no objection to the facsimile reproduction by anyone of the patent document or the patent disclosure, as it appears in the Patent and Trademark Office patent file or records, but otherwise reserves all copyright rights. Background Art

[0005] In order to obtain increased yields from the agricultural land under their control, in recent years, farmers have turned to "precision agriculture" to distinguish areas within a single field that have different levels of fertility or other properties, and to determine best practices for fertilizer application and other management on a site - specific basis. This site - specific application is achieved through variable rate technology (VRT): precision agriculture equipment applies inputs such as fertilizers and soil amendments at different rates within a field based on the spatial variability of soil, crops, etc. One approach to precision agriculture is to delineate "management zones" within a single field. A management zone is an area within a field that has similar properties, is subject to common management, and is different from other similarly delineated areas within that field.

[0006] The "precision agriculture" commonly employed makes use of the Soil Survey Geodatabase (SSURGO) of the Natural Resources Conservation Service of the United States Department of Agriculture (USDA). Although the data resident in SSURGO provides a useful approximation of soil conditions, its originators never intended it to be used to guide precision agriculture or to develop agricultural management zones within a field.

[0007] Other management zone delineation techniques tend to be local and field - specific, relying on older technologies such as proximal sensing (of soil conductivity), visual pattern recognition based on satellite imagery with small samples (sometimes only one sample size), topographic analysis, grid soil sampling, crop yield monitoring data, and farmer and / or advisor knowledge. Brief Description of the Drawings

[0008] The organization and method of the illustrative operation, together with certain illustrative embodiments of purpose and advantages, may best be understood by reference to the specific embodiments set forth below in conjunction with the accompanying drawings, in which:

[0009] Figure 1 is a first process - flow diagram showing a reduction of Zone Attributes and Prescription Recommendation consistent with certain embodiments of the present invention.

[0010] Figure 2 is a second process - flow diagram showing a reduction of Zone Attributes and Prescription Recommendation consistent with certain embodiments of the present invention.

[0011] Figure 3 is a third process - flow diagram showing a reduction of Zone Attributes and Prescription Recommendation consistent with certain embodiments of the present invention.

[0012] Figure 4 is a view of the data stack reduced to a machine - learning - ready data cube consistent with certain embodiments of the present invention. Detailed Description of the Invention

[0013] Although the present invention is susceptible to many different forms of embodiments, specific embodiments are shown in the drawings and will be described in detail herein, and it is understood that this disclosure of such embodiments will be considered an example of the principles and is not intended to limit the present invention to the specific embodiments shown and described. In the following description, like reference numerals are used to describe the same, similar, or corresponding parts in several views of the drawings.

[0014] As used herein, the term "a" or "an" is defined as one or more than one. As used herein, the term "plural" is defined as two or more than two. As used herein, the term "another" is defined as at least a second or more. As used herein, the terms "comprising" and / or "having" are defined as including (i.e., open language).

[0015] References throughout this document to "one embodiment", "certain embodiments", "an embodiment", or similar terms mean that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the present invention. Thus, appearances of such phrases throughout the specification are not necessarily all referring to the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments without limitation.

[0016] References throughout this document to "SSURGO" represent the Soil Survey Geographic Database of the Natural Resources Conservation Service of the United States Department of Agriculture.

[0017] References to "DEM" or its plural "DEMs" in this document represent "Digital Elevation Model" or "multiple Digital Elevation Models".

[0018] References to "STM" in this document represent "Soil Terrain Model".

[0019] References to "remote sensing" in this document represent the acquisition of information by humans and machines without physical contact, including aerial and satellite imagery and light detection and ranging (hereinafter referred to as "LIDAR").

[0020] References to learning systems such as "artificial intelligence" and / or "deep learning" in this document represent data analysis using statistics, classification algorithms, artificial neural networks, machine learning, and / or feature recognition or pattern recognition.

[0021] References to "area" or "areas" in this document represent one or more geospatial areas that meet pre-established agricultural criteria. In non-limiting examples, such criteria can be as simple as the latitude and longitude coordinates that describe the scale and boundaries of the area, or as complex as conforming to specific yield and soil variability conditions.

[0022] References to "variable rate technology" in this document represent technologies that enable growers to apply different or different rates of agricultural inputs and management to various spatially defined areas (as non-limiting examples, management zones) within a single field.

[0023] References to "ground truth" in this document represent the collection of past data on the crop varieties planted, soil test results, fertilizers and soil amendments applied in the past, and the actual yields in each crop planting area, which present a true picture of the management and yields of a specific geographic area.

[0024] For thousands of years, in order to maximize crop yields, farmers have not only tried to better understand the conditions in which their plants perform best, but also the conditions that the fields in which they sow the seeds of these plants provide for their crops. Clearly, farmers who best match field conditions with crop needs can be more confident in maximizing or at least increasing crop yields.

[0025] Recent precision agriculture technologies have used techniques to develop the identification and delineation of crop management zones that are too local and specific to fields. The soil-landscape paradigm of traditional soil mapping recognizes five factors that influence the spatial distribution of soil: climate, organism, relief, parent material, and time. The most detailed source of information about soil parent material or surficial geology is the soil survey maps published by the USDA-NRCS (SSURGO). While useful tools in their time, the soil survey maps themselves do not contain the level of detail needed to ensure farmers achieve state-of-the-art yields.

[0026] Accordingly, there is a need for an agricultural technology that allows for the delineation of management zones over a wide geographic area in the absence of overly local field-specific data. While in embodiments, the current innovation utilizes the parent material information from SSURGO as an input data layer, it also uses other data sources to account for other yield-related factors. In further embodiments, the current innovation uses an iterative artificial intelligence process with data from multiple sources to provide granular zone detail.

[0027] The current innovation delineates enhanced management zones over a broad geographic area through remote sensing, pattern recognition, and artificial intelligence capabilities consisting of machine learning and statistical analysis, and combines all of them with data sourced from existing national soil survey databases to guide precision agriculture sampling and management. By delineating enhanced management zones, farmers can significantly reduce the number of soil samples required to formulate soil amendments such as lime and fertilizers. These zones are equally well-suited for selecting crop varieties developed for different soil environments. Enhanced management zones can also be useful in guiding measures to control weeds, pests, and crop diseases. Additionally, such zones may prove useful in non-agricultural land management such as forestry and natural resource conservation.

[0028] In an embodiment, the current innovation combines the utilization of topographic and spectral information. Based on aerial LIDAR, multi-scale topographic derivatives are generated from high-resolution DEMs. These digital topographic derivatives include, but are not limited to, slope gradient, relative elevation, profile curvature, and plan curvature. Additionally, multi-temporal, multi- / hyperspectral, satellite, and aerial images are compiled. Existing vegetation and soil indices are calculated based on the spectral bands of these images. These indices are mathematical combinations of imagery bands that have historically been shown to be useful in characterizing vegetation and soil. During the process of identifying management zones and predicting soil properties, multiple topographic derivatives created at different analysis scales are used as geospatial layers parallel to the spectral layers. Soil properties include all identified co-variants.

[0029] In an embodiment, a hyperdimensional data cube can be formed to provide inputs to the analysis and machine learning processes of the prediction system. The hyperdimensional data cube can be formed in different aspects based on sensors, images, ground truth, analysis indices, soil parent material, and digital terrain analysis data used in forming the hyperdimensional data cube. In one or more non-limiting examples, the hyperdimensional data cube can be formed by using sensor data from remote sensing of field conditions to create multi-scale digital terrain analysis data and combining it with soil parent material data input from SSURGO. Alternatively, the hyperdimensional data cube can be formed by combining image band data from satellites and other imaging systems with the calculated vegetation and soil indices. Another hyperdimensional data cube can be formed by using sensor data from remote sensing of field conditions to create multi-scale digital terrain analysis data, combining it with soil parent material data input from SSURGO, and combining it with creating a multi-spectral data cube using image band data from satellites and other imaging systems and the calculated vegetation and soil indices. The system can use each data cube to predict soil area properties and provide recommendations for crop planting, area management, and soil maintenance.

[0030] In a non-limiting example, the data cube can be used as an input to ISODATA used by the system and be used to train artificial intelligence algorithms to predict and / or estimate certain agronomic parameters. The predicted and / or estimated agronomic parameters are combined with the ground truth to delineate a set of optimized management zones.

[0031] In an embodiment, an unsupervised classification algorithm will be used for zone delineation: given multi-scale topography and multi-spectral layers (e.g., data cubes) of some agricultural fields of interest stratified with parent material data, the algorithm will delineate different agricultural zones based on common in-zone data characteristics. These zones can be characterized by relative yield and yield stability and / or variations in georeferenced soil sample test data. Since the input data is inherently continuous, the prediction of soil properties and / or yield is treated as a regression problem and a fully connected neural network is used.

[0032] In an embodiment, artificial intelligence (AI) (in non-limiting examples, such as pattern recognition) is applied to georeferenced crop yield and soil test data to delineate preliminary management zones based on data cubes, regardless of the data cube aspects employed by the AI process. Zone delineation is optimized based on variable zone sizes. In non-limiting examples, the optimized zone size can be the minimum zone size desired by the grower. The zone size for each individual grower can be calculated based on the minimum area that the grower can or intends to manage using variable rate technology. Variable rate technology enables growers to apply different types or rates of agricultural inputs and soil management techniques to various spatially defined zones within individual fields, which can be referred to as "management zones". Thus, based on the combination of the preferred zone size desired by the grower and the management zones predicted and recommended by the artificial intelligence algorithm, the zone delineation for each grower can be different. This optimization further delineates the minimum variations within zones and the maximum variations between and within zones.

[0033] Python tools have become very popular in machine learning, and current innovations will utilize these tools during neural network development to create zone attributes and soil management recommendations. Tools such as Pandas, Numpy, TensorFlow, and Keras are several specific examples of libraries that fully support the parallel mathematical operations required to build neural networks and execute output models that provide optimized outputs for growers.

[0034] A neural network is a supervised artificial intelligence (as opposed to unsupervised) that uses labeled data: for each input sample, there is a corresponding output value, also known as the "ground truth". The inputs and outputs, referred to as the training set, are related by a mathematical equation with unknown coefficients, and the network in any embodiment of the current innovation must learn these mathematical equations. A human trainer supervises the operation of the neural network and provides the necessary guidance to the neural network to identify and learn the coefficients of the mathematical equations to produce useful results.

[0035] Human supervision is used to determine how well the network is learning, as it iteratively inputs during training and corrects itself based on the output. During this phase, architectural changes are made to the network to improve its performance. When the human network designer determines that the accuracy for the training set is sufficient, the network is then tested against a development set (a small subset of data separated from the previous training set before any training activity). If the output accuracy for the development set is determined to be insufficient, its tuning parameters (referred to as hyperparameters) can be adjusted, and the network is iteratively re-run against the development set until the accuracy improves. Finally, the network is run against "test set" data, where overfitting and underfitting of the model to the development set can be determined, and another set of tuning adjustments is performed to better generalize or better specialize the model, respectively.

[0036] In an embodiment, the current innovation will use fully connected neural networks for regression, one network per soil property. Regardless of the data cube aspect, the training set is derived from the data cube. Values will be scaled to normalize the data values, and each input sample will correspond to a soil test point. The corresponding ground truth for each soil test point can be the soil property value taken from the same location. All data layers will share the same geographic coordinate system and projection. During development, the corresponding ground truth labels will be based on an examination of actual yield monitoring and test data from georeferenced soil samples collected from the same location, if such data is available from the customer. Finally, the training, development, and test sets will be allocated. The target accuracy of the network can be set based on customer feedback. From there, the iterative process described previously can be carried out. The final product, the RxMaker model, will consist of the final network architecture, learning parameters, learning rate, hyperparameter values, and any heuristics (such as regularization) that may need to be applied to reduce bias and variance errors.

[0037] In such an embodiment, predictions of soil properties can be made at infinitely high resolutions; however, as the resolution increases, the uncertainty about the prediction accuracy also increases. In a non-limiting example, as the size of the management zone increases, the heterogeneity of the zone may increase, but the increased prediction range reduces the uncertainty of the respective predictions of the soil properties. In a commercial setting, the model outputs will be aggregated to produce a management zone size with an uncertainty level that is adapted to the farmer's needs and can be implemented through the capabilities of the farmer's equipment.

[0038] The aggregation of model outputs will be achieved at least in part using spatial generalization techniques. These include, but are not limited to, algorithms that dissolve regional boundaries. Dissolving regional boundaries helps smooth rough regional edges and merge small regional patches and / or inclusions with larger surrounding regions. Smoothing rough regional edges and incorporating small regions or inclusions into larger surrounding regions can allow the system to correlate soil property predictions for the larger region over a combined region consisting of the larger region and one or more smoothed rough regions or regional patches and inclusions. In this way, the dissolved zone boundaries can create an aggregated zone consisting of the large region and one or more regional patches and / or inclusions.

[0039] In a non-limiting example, the generalization tools provided by the ArcGIS toolset offered by ESRI can be used to achieve such regional smoothing. The generalization tools can include, but are not limited to, Aggregate, Boundary Clean, Expand, Majority Filter, Nibble, RegionGroup, Shrink, and Thin tools.

[0040] Optimization of agricultural regions can be created by using each of the data cube aspects discussed previously. A first scenario can be achieved by applying a first hyperdimensional data cube as input to one or more learning algorithms, the data cube including sensor data from remote sensing of field conditions creating multi-scale digital terrain analysis data combined with soil parent material data input from SSURGO, where the learning algorithm is supervised or unsupervised. Alternatively, a second scenario can be achieved by applying a hyperdimensional data cube formed by combining image band data from satellites and other imaging systems with calculated vegetation and soil indices again as input to one or more supervised or unsupervised learning algorithms. A third scenario can be achieved by combining sensor data from remote sensing of field conditions creating multi-scale digital terrain analysis data with soil parent material data input from SSURGO, and combining image band data from satellites and other imaging systems with calculated vegetation and soil indices to create a multi-spectral data cube as input to one or more supervised or unsupervised learning algorithms. The system can be required to use each aspect of the hyperdimensional data cube as input to create regional attributes, soil predictions, and management recommendations. Then, the system can produce three scenarios, one from each sub-process using the first, second, or third data cube aspect. By employing three sub-processes to delineate optimized agricultural regions, the current innovation internally examines its results to achieve maximum optimization. The results of three parallel region delineation strategies using the three aspects of the data cube as input are compared to each other, and the best enhanced management zones and recommended data can be provided to the client. Optimization is achieved by determining the data cube that best leads to predicting any specific parameter of interest, where the input parameter of interest is provided by the grower or farm management entity.

[0041] In a non-limiting example, a system and method for optimizing agricultural zone attributes, zone prediction, and recommendations includes at least a data processor communicating with a data server that is in data communication with a user device capable of displaying data representations to a user. Using any one of three data cube aspects, the system can use remote sensing and digital analysis to build a first data set, use collected image bands and calculated indices to build a second data set, and / or use a combination of inputs from the first and second data sets to build a third data set. The system can then apply an artificial intelligence algorithm to the first data set and output a first set of zone attributes, apply the artificial intelligence algorithm to the second data set and output a second set of zone attributes, and apply the artificial intelligence algorithm to the third data set and output a third set of zone attributes. To determine the optimal set of zone attributes, the system can compare the first, second, and third sets of zone attributes based on one or more parameters input by the user as a guide. The system can then deliver to the user an optimized set of zone attributes and one or more zone management recommendations, where the set of zone attributes presents zone attributes and / or recommendations for zone management that are optimized for the one or more parameters input by the grower or farm management entity.

[0042] Turning now to Figure 1 , a simplified first process-flow diagram is shown that depicts zone attributes and formulation recommendations consistent with certain embodiments of the present invention. Figure 1The sub - process begins at 100. In a non - limiting example, remote sensing of field conditions is performed via aerial LIDAR at 102. At 104, a digital elevation model (DEM) based on the LIDAR data undergoes multi - scale digital terrain analysis (DTA), which generates multiple attributes, including as non - limiting examples, slope curvature. These attributes will be used in subsequent steps. The DTA data, along with the input soil parent material data 108, is compiled in a data cube (A) at 106. At 110, the data compilation is used to train an artificial intelligence deep - learning algorithm to predict or estimate georeferenced crop yields and soil test parameters. At 114, the application of the deep - learning algorithm (at 110) generates prescription recommendations, typically in a non - limiting format example such as a table or a map. The application of the deep - learning algorithm at 110 also generates zonal attributes (A) to aggregate the zones and the attributes of the zones at 122. The data compiled in the data cube (A) 106, along with the soil parent material data 108, also undergoes an artificial intelligence unsupervised classification learning algorithm 112 and is processed to delineate preliminary zones at 116. The effectiveness of the zone delineation at 118 is evaluated based on the degree to which the zones capture the spatial variability in georeferenced soil test data and crop yields. The effectiveness of the zone delineation at 118 can be calculated using ground - truth empirical - provable data. Alternatively, the effectiveness of the zone delineation at 118 can be calculated using the deep - learning output from the application of the deep - learning algorithm at 110. The zone delineation based on SSURGO map units is also evaluated. Among all candidate delineations, the best ones are: 1) maximize the total number of statistically distinct zones for each soil or crop parameter under discussion (as a non - limiting example, the current innovation can be used to create maps of specific parameters, such as a map of phosphorus, different from a map of potassium, which in turn is different from a map of organic matter, etc.); 2) maximize the interval differences of these parameters; and 3) minimize the sum of the weighted variances within the zones. The data for these evaluations is actual georeferenced soil test and yield data (when available). When not available, predictions from the artificial intelligence prediction algorithm are used. The "virtual agronomic effectiveness" of the delineation is based on the degree to which the interval differences are large enough to warrant variable - rate management. These judgments are made based on: 1) the likelihood of responding to differential management within the capabilities of the grower's variable - rate application equipment; and 2) grower preferences. At 120, the system incorporates management "rules" and at 122, exports the aggregated zonal attributes (A) of the smoothed and aggregated zones that are consistent with pre - established management rules. The sub - process ends at 124.

[0043] Now turning to Figure 2 , a second process - flow diagram is shown that describes reducing input images and index data to zonal attributes and prescription recommendations consistent with certain embodiments of the present invention. Figure 2 The sub - process, possibly associated with Figure 1The sub-process parallel operation starts at 202. At 204, multi-temporal, multi-spectral satellite and aerial image bands are collected. At 206, existing vegetation and soil indices are calculated from the collected spectral bands. These indices are mathematical combinations of the image bands and have been shown to be useful in characterizing vegetation and soil. At 208, the spectral image bands and the calculated indices are combined into a data cube (B) 210. The compiled data of the data cube (B) 210 is input into at least a deep learning algorithm 212 and an unsupervised learning algorithm 214.

[0044] At 212, the data compilation is used to train an artificial intelligence deep learning algorithm to predict or estimate georeferenced crop yields and soil test parameters. At 218, the application of the deep learning algorithm (at 212) generates prescription recommendations, typically in the form of non-limiting examples of tables or maps. The application of the deep learning algorithm at 212 also provides an input for creating regional attributes (B) to aggregate the attributes of regions and areas at 224.

[0045] The compiled data of the data cube (B) 210 is also subjected to an artificial intelligence unsupervised classification learning algorithm 214 and processed to demarcate preliminary regions at 216. The regional demarcation effectiveness at 220 is evaluated based on the extent to which the regions capture the spatial variations in the georeferenced soil test data and crop yields. The regional demarcation effectiveness at 220 can be calculated using empirically verifiable data of real ground truth. Alternatively, the regional effectiveness demarcation at 220 can be calculated using the deep learning output from the application of the deep learning algorithm at 212. The demarcated "virtual agronomic effectiveness" is based on the degree to which the intervals are different enough to warrant variable rate management of the differences. These judgments are made based on: 1) the likelihood of responding to differential management within the capabilities of the grower's variable rate application equipment; and 2) grower preferences. At 222, the system incorporates management "rules" and exports aggregated regional attributes (B) of smoothed and aggregated regions consistent with pre-established management rules at 224. The sub-process ends at 226.

[0046] Now turning to Figure 3 FIG. shows a simplified third process - flow chart showing regional attributes and prescription recommendations consistent with certain embodiments of the present invention. Figure 3The sub - process begins at 300. At 310, LIDAR is used to provide remote sensing of the field conditions. The multi - scale digital terrain analysis derivatives at 312 and the soil parent material data enhancement at 314, and the data cube (C) at 308 are used to train an artificial intelligence deep - learning prediction algorithm at 326 and to demarcate a third set of optimized agricultural regions at 324. The deep - learning algorithm at 326 generates recipe recommendations at 328. The data cube (C) 308 inputs data into an unsupervised learning algorithm at 316, resulting in area demarcation at 318. At 320, the system evaluates the area effectiveness using the input from the deep - learning algorithm. At 322, the system incorporates management "rules" to determine the aggregated area attributes (C) of the smoothed and aggregated areas at 324. At 330, the sub - process ends.

[0047] The predicted values of data cube (A), data cube (B), and data cube (C) are statistically compared in order to select the data cube (from the set of data cubes A, B, and C) that best leads to the prediction of any particular parameter of interest.

[0048] Now turning to Figure 4 , a view is shown of the data stack reduced to a machine - learning - ready data cube consistent with certain embodiments of the present invention. Each layer in the data - layer stack at 402 quantifies a single attribute. The attributes include: 1) terrain as captured in a digital elevation model (DEM); 2) DEM derivatives such as DTA; 3) individual spectral bands, e.g., in a non - limiting example, red, green, blue, near - infrared, hyperspectral, or any other provided image spectral band from satellite imagery; and 4) their derived vegetation and soil indices. The data - layer stack 402 is combined to form a hyper - dimensional data cube 404. The hyper - dimensional data cube 404 is used to train an artificial intelligence prediction algorithm at 406 and to demarcate regions via an artificial intelligence unsupervised classification algorithm.

[0049] Although certain illustrative embodiments have been described, it is evident that many alternatives, modifications, permutations, and variations will become apparent to those skilled in the art in light of the foregoing description.

Claims

1. A system for optimizing the regional properties of agricultural soil, comprising: A data processor that communicates with a data server; A user device that can display data representations to a user; A first data set constructed from remote sensing and digital analysis; A second data set constructed from collected image bands and calculated indices; A third data set constructed from a combination of inputs of the first data set and the second data set; wherein the system is configured to: Apply an artificial intelligence algorithm to the first data set and output a first set of soil regional properties; Apply an artificial intelligence algorithm to the second data set and output a second set of soil regional properties; Apply an artificial intelligence algorithm to the third data set and output a third set of soil regional properties; Compare the first set of soil regional properties, the second set of soil regional properties, and the third set of soil regional properties; Create preliminary regional boundaries from the first set of soil regional properties, the second set of soil regional properties, and the third set of soil regional properties by the following steps: (1) maximize the total number of statistically different soil regional properties, (2) maximize the inter-regional differences between soil regional properties, and (3) minimize the within-region variance; Fuse the regional boundaries of regions with at least one set of soil regional properties to create an aggregated region consisting of a larger region and one or more regional blocks and / or inclusions; and Deliver to the user an optimized set of soil regional properties of the aggregated region and one or more regional management recommendations based on the optimized set of soil regional properties, wherein fusing the regional boundaries enables smoothing of rough regional edges and fusing of small regional blocks and / or inclusions with the larger surrounding region, thereby allowing the system to associate soil property predictions of the larger region on a combined region consisting of the larger region and one or more smoothed rough regions or regional blocks and inclusions.

2. The system according to claim 1, wherein, The remote sensing is implemented using LIDAR.

3. The system according to claim 1, wherein The first data set, the second data set, and the third data set are supplemented with SSURGO soil parent material data.

4. The system according to claim 1, wherein, The artificial intelligence algorithm is a deep learning algorithm, an unsupervised learning algorithm, or a combination of a deep learning algorithm and an unsupervised learning algorithm.

5. The system according to claim 1, wherein, The evaluation of regional effectiveness includes ground truth or the output of a deep learning algorithm.

6. A method for optimizing the regional properties of agricultural soil, comprising: Construct a first data set using remote sensing and digital analysis; Construct a second data set using collected image bands and calculated indices; Construct a third data set using a combination of inputs of the first data set and the second data set; Apply an artificial intelligence algorithm to the first data set and output a first set of soil regional properties; Apply an artificial intelligence algorithm to the second data set and output a second set of soil regional properties; Apply an artificial intelligence algorithm to the third data set and output a third set of soil regional properties; Compare the first set of soil regional properties, the second set of soil regional properties, and the third set of soil regional properties; Create a preliminary regional boundary from the first soil area attribute set, the second soil area attribute set, and the third soil area attribute set by the following steps: (1) Maximize the total number of statistically different soil area attributes, (2) Maximize the inter-regional differences between soil area attributes, and (3) Minimize the within-region variance; Fuse the regional boundaries of regions with at least one soil area attribute set to create an aggregated region consisting of a larger region and one or more regional blocks and / or inclusions; And Deliver to the user an optimized soil area attribute set of the aggregated region and one or more regional management suggestions based on the optimized soil area attribute set, wherein fusing the regional boundaries enables smoothing of rough regional edges and fusing of small regional blocks and / or inclusions with the larger surrounding region, thereby allowing the prediction of soil properties of the larger region on a combined region consisting of the larger region and one or more smoothed rough regions or regional blocks and inclusions to be correlated.

7. The method according to claim 6, wherein, The remote sensing is implemented using LIDAR.

8. The method according to claim 6, wherein, The first data set, the second data set, and the third data set are supplemented with SSURGO soil parent material data.

9. The method according to claim 6, wherein The artificial intelligence algorithm is a deep learning algorithm, an unsupervised learning algorithm, or a combination of a deep learning algorithm and an unsupervised learning algorithm.

10. The method according to claim 6, wherein The evaluation of regional effectiveness includes ground truth or the output of a deep learning algorithm.

Images