Suitable area identification for mulberry plantation using query-based prescriptive analytics and microclimatic parameters

ABSTRACT

Precision farming plays a vital role in suitable land location identification. Precision farming never considers microclimatic condition parameters for suitable land location identification. Farmers need know suitable crops and land area for cultivation based on current soil condition and microclimatic data. Farmers need yield predictions before cultivation of any crop. In this paper, a suitable area for Mulberry plantation cultivation is identified using current soil, microclimatic conditions and query using prescriptive analytics for higher yield predictions. Proposed Query-based Prescriptive analytics (QPA) for mulberry is performed through descriptive and predictive analysis. QPA recommend farmers for suitable area for cultivation of mulberry plants based on query such as soil, microclimatic and previous crop yield data. Descriptive analysis is performed through hybrid machine learning algorithms such as PCA-enabled GPR (PG) and Bayesian-optimized GPR (BG) for identification of data patterns and trends. Predictive analysis is performed using Decision Tree ID3 (DT) algorithm and Pelican optimized LSTM (PL) for land suitability analysis. QPA based on BG-PL, combination of descriptive and predictive analysis, provides 99% accuracy in suitable land identification and crop yield prediction before cultivation. Proposed BG-DT, PG-PL and PG-DT methods of QPA provide suitable land identification with accuracy of 90%, 85% and 80%, respectively.

KEYWORDS:

• Query-based analysis
• land suitability
• machine learning
• decision tree ID3 algorithm
• mulberry plantation

Introduction

Site suitability analysis is a prerequisite in farming based on current soil property and for rotation cropping. Existing land suitability methods never consider the actual mineral requirements for a particular crop and the current mineral level in the soil. Mineral contents in the soil are identified through the inherent properties and constraints of land parcels (Kahsay et al., 2018). The site suitability analysis identifies the problems in farming. Site suitability analysis helps land managers for rotational crop management. Land suitability is performed for the crop yield assessment based on rotational crop cultivation. Land suitability analysis helps the farmers for maximum income through maximum crop yield. In site suitability analysis, parameters such as soil, topography and weather conditions are used for a particular crop and region. The soil, topography and climate of a region are the main factors for site suitability of any agricultural plantation (Gholizadeh et al. 2020; Jafari & Narges Zaredar, 2010; Lupia, 2014; Tadesse & Negese, 2020 Tashayo et al., 2020). Existing methods consider past crop and soil data for land suitability analysis. For crop production, crop management and soil sustainability analysis, major parameters, such as soil depth, texture, pH, drainage, and electrical conductivity, are considered. Topography parameter, such as land slope, plays a major role in site suitability due to soil erosion. Soil erosion removes the soil nutrients. A qualitative approach is used for analysis of the land suitability (Halder, 2013), Land systems, Landforms, soils, hydro-geomorphology and land uses are considered for land suitability. The basic parameters, such as soil, topography and climate, play an important role in site suitability analysis. The above parameters vary for the growth and yield of each crop. So, the weightage of parameters is more important for lite suitability assessment. The Analytic Hierarchy method is used for the identification of the weightage level for each crop, during land suitability analysis (Duc, 2006). Land suitability analysis of a crop is performed using a fuzzy algorithm and with weight level for each parameter based on the crop (Kurtener et al., 2008). Integrated land suitability potential index is based on soil, terrain and land use parameters (Bandyopadhyay et al., 2009). GIS-based multi-criteria land suitability analysis is performed based on 15 parameters, such as soil pH, soil type, soil drainage, soil depth, impermeable layer, phase, organic carbon, texture classes, obstacle to root, land use/land cover, slope and distance from the river outlets. The Weighted Overlay tool determines the weight level and combines the weight of different parameters for suitability analysis (Hussien et al., 2019). The Rule-Based System (RBS) plant suitability analysis considers the parameters, such as rainfall, altitude, drainage, soil type, pH, flood risk, fertility of the soils and soil depth. The parameters, such as rainfall, temperature, slope percentage, soil types and water well distribution, are considered for rain-fed and irrigated land-based site suitability analysis (Al-Taani et al., 2021). Soil depth, soil type and soil fertility parameters are obtained from soil map slope and aspect ratio and elevation parameters are obtained from shuffle radar topography data (Sohaib et al., 2022). The data are applied in geographic information systems for mapping and analysis, generating land suitability maps. Land suitability analysis is performed with parametric and AHP methods for different crop cultivations, such as oak, pine, ground nut, sorghum millet and maize. The land suitability analysis for crops is performed with machine learning algorithms, such as SVM and RF (Congzhen Xiao et al., 2022; Damtew Tsige, 2022; Hümeyra, 2021; Juncheng Wang et al., 2021). Land suitability analysis needs timely and reliable information about parameters, such as soil, topography and climate (Kahsay et al., 2018). Proper site selection enhances agriculture efficiency and reduces environmental impacts (Ghobadi et al., 2021).

Microclimatic conditions refer to changes in climate conditions in small pieces of land. Microclimatic conditions have a high impact on crop growth. Microclimatic factors, such as sunlight, temperature, humidity, moisture and wind, affect crop growth. Understanding the microclimatic conditions helps farmers for efficient decisions-making for crop selection/rotational cropping (Singh et al., 2016). Agrivoltaic systems are used for crop growth and development, which are never suitable for all crops. Microclimatic conditions vary across different regions and climates due to other factors such as Geography, Vegetation, Soil type and Climate change. Climate change can also affect microclimatic conditions (Ahemd et al., 2016).

Inferences from literature survey

Researchers consider various parameters for land suitability and crop yields and never consider the microclimatic parameter for land suitability analysis. Microclimatic conditions have a significant impact on soil properties. Microclimate refers to the local climate conditions which influence the parameters, such as topography, vegetation and soil type. Microclimatic conditions, such as solar radiation and shading, affect the soil temperature, which affects soil biological activity and nutrient cycling. Microclimatic conditions, such as rainfall, evapotranspiration and topography, affect soil moisture levels. Soil moisture is important for nutrient availability, plant growth and soil structure. Soil moisture deficits lead to soil compaction and reduces nutrient level in the land. Microclimatic conditions such as wind speed and direction affect soil erosion, leading to the loss of topsoil and nutrients. Wind erosion changes the texture and structure of the soil. Microclimatic conditions, such as shading, affect the growth and development of crops, which affects the soil organic matter and nutrient cycling. Shaded areas reduce crop growth and lower the nutrient cycling rates in the land. Microclimatic conditions affect the soil properties nutrient cycling and soil structure. Microclimatic conditions have complex and interrelated effects on soil parameters. Understanding the local microclimate condition and soil properties helps farmers and land managers make decisions on rotational cropping. Precision farming (PF) helps farmers for high average yields of crops compared to traditional methods of farming. PF focuses on sustainable agriculture and organic crop cultivation, increases profit, production and economic efficiency and reduces the use of fertilizers on the land, avoiding land pollution. In this paper, microclimatic conditions are considered in land suitability analysis, for mulberry plantations as rotation cropping based on farmers’ queries.

Problem statement

In this study, wastelands are used in site suitability analysis for mulberry plant rotational cropping and yield prediction. To meet the requirement of food demand, the farming community needs to increase the productivity of food grains from the limited agricultural lands. Therefore, intensive farming uses hybrid crops, more pesticides and inorganic fertilizers, which leads to disease in humans and causes environmental issues. Precision farming and organic farming technique plays a major role in minimizing environmental impact and health issues due to inorganic crop yield. Precision farming techniques need proper crop selection and suitable land analysis for a particular crop and yield prediction, which are achieved by site suitability analysis. Present technological advancements in Remote Sensing, GIS and Machine learning domains are used for land suitability analysis. However, microclimatic climate condition-based land suitability analysis needs to be developed considering various current land parameters such as soil, topography and climate. However, land suitability prediction must include micro-climatic parameter data for accurate land suitability (Figure 1).

Figure 1. Workflow of analysis of the QPA method for mulberry cultivation.

Research gap and motivation

Land suitability for a particular crop and yield prediction before cultivation are the major requirements for sustainable farming. Land suitability for any crop needs to be evaluated, frequently, due to changes in microclimatic conditions and cyclic integral agriculture i.e. sunlight, CO2, soil and water quality, etc. Land suitability changes from one season to another season based on the ambient environment conditions i.e. micro-climatic conditions. The motivation of this paper is to identify mulberry plant suitability for cultivation in the forthcoming season based on the input parameter given by the farmers about their land in the user query portal of the proposed QPA system.

Contributions

From the above discussion, the parameters which influence the land suitability analysis for any crop are soil properties and micro-climatic conditions. In this paper, micro-climatic parameters and slope characteristics are used for site suitability analysis for mulberry plants and yield prediction. In this paper, soil, climate and topography parameters are used in descriptive analysis and predictive analysis in the proposed QPA model. Machine learning is used for accurate, reliable information and identification of land suitability analysis than conventional methods. The QPA method is tested for mulberry plantations in wastelands in Krishnagiri district, Tamil Nadu, India.

1. To develop a QPA model for more accurate prediction of land suitability analysis with high productivity based on microclimatic conditions through a hybrid algorithm, which included current soil parameters and land temperature variations for land.

2. To develop a query-based system for farmers, when farmers feed the current input land parameters, such as temperature, soil and water and rainfall data of a particular land area, the proposed QPA identifies the productivity based on the spatial and temporal datasets of the previous year and suggests the suitable land area for high productivity for mulberry plants.

3. To develop a land suitability response output map is provided to farmers based on the input parameters fed to QPA, which is implemented using the GIS Map overlay method.

4. To propose a QPA model, which is the hybridization of Descriptive Analysis such as (i) Principal component analysis-enabled Gaussian process regression (PG) (ii) Bayesian Optimized GPR (BG) and Predictive Analyses, such as (i) Decision Tree ID3 (DT) algorithm and Pelican optimized LSTM (PL).

In this paper, the second section explains the study area. The third section describes the methods and materials in QPA. Finally, the fourth section addresses the proposed QPA method recommended in wasteland areas. The main purpose of this study is to identify the suitable land area for mulberry plantations and yield prediction in Krishnagiri district, Tamil Nadu. QPA is analysed and verified with ground truth verification.

Materials and methods

Krishnagiri district is located in Tamil Nadu, India. The total geographical area of Krishnagiri district is about 5143 sq. kms, with an elevation between 300m and 1400 m above Mean Sea Level. Krishnagiri district has a hot climate from March to September and a cold climate from December to February. The average rainfall is about 830 mm per annum. Krishnagiri district consists of hills and hillocks. The people in Krishnagiri are known for innovative farming, which is about 62%. Krishnagiri district consists of 63% of agricultural land, 8% wasteland, 4% built-up area, 21% forest and 3% water body (Buvan portal). The study area is shown in Figure 2. Land suitability for mulberry cultivation matches current land quality parameters and microclimatic conditions. Mulberry plantation considers soil, topographical, climate properties and water availability parameters in the proposed QPA-based land suitability analysis. In this study, two methods are applied for mulberry plantation site suitability analysis using the proposed QPA model: descriptive analysis and predictive analysis. In the descriptive analysis, Hybrid Machine Learning methods such as (i) PCA-enabled GPR and (ii) Bayesian Optimized GPR are used. In the predictive analysis, the machine learning approach, the decision tree ID3 algorithm is used. The decision tree structure builds nodes and edges from the dataset. Each node is used to make a decision or represent an outcome. The QPA-based land suitability analysis and traditional methods are compared with ground truth verification. Figure 1 shows the workflow methodology of QPA.

Figure 2. SmartArt for the literature survey.

Query-based prescriptive analytics

QPA is performed through Descriptive and predictive analysis. The descriptive analysis consists of (i) PCA-enabled GPR (PG) and (ii) Bayesian Optimized GPR (BG) methods. For the descriptive analysis, the major types of parameters are included such as soil properties, topographical characteristics and microclimate parameters. The sources of the parameters are listed in Table 1. Numerical rating based on limitations is explained in this section. The criteria for land/site suitability analysis are soil depth, soil pH, soil EC, Soil drainage, soil texture, slope, Temperature, Rainfall and Groundwater availability. The four types of degree of limitations and suitability classes (S) are used in QPA Degree of limitation: [1]-indicates no or slight limitation and highly suitable (S1), [2]- indicates moderate limitation and moderately suitable (S2), [3]-indicates severe limitation and marginally suitable (S3) and [4]-indicates very severe limitation and never suitable represented as (S4). The degree of limitations is listed in Table 2.

Table 1. Microclimate data for descriptive analysis of mulberry plantation in Krishnagiri.

S.No	Parameters	Data collection	Data Source
1	Soil	Depth, pH, EC, Texture, Drainage,	Tamil Nadu Agricultural University (TNAU), Coimbatore
2	Water availability	Groundwater availability	Institute of Remote Sensing, Anna University.
3	Topography	Slope	CartoSat DEM Satellite data
4	Micro- Climate	Temperature, Rainfall and Relative Humidity	Measured from agricultural fields

Table 2. Limitation criteria for land suitability class for mulberry cultivation.

Parameters (micro-climate)	Units	Degree of limitations (DL)/suitability class (S)
		1 (DL)	2 (DL)	3 (DL)	4 (DL)
		S1	S2	S3	S4
Soil Depth	Cm	>100	70–100	25–70	<25
Soil pH	–	6.5–7.5	5.5–6.5 7.5–8.5	4.5–5.5 8.5 – 9.5	<4.5 >9.5
Soil Texture	Class	Clay loam – gravelly clay	Fine loamy	Course loamy	Sandy fragmental
Soil drainage		Well drained	Moderately well drained	Imperfectly drained	Excessive drained
EC		<2000	2000−3000	3000–4000	>4000
Groundwater		Good	Fair	Fair to moderate	Poor
Slope	%	0–3	3–5	5–10	>10
Temperature	°C	20–30	30 −37	30–37	<15 >37
Rainfall	mm	500–750	750–2000	2000–3400	<500 >3400

In land suitability analysis, evaluation criteria are represented as ordinal values, which indicates the degree of limitation for mulberry crop cultivation. In the QPA model, the microclimatic parameters and water resources are included. Mulberry cultivation needs soil depth, categorized into >100 cm (deep), 50–100 cm (moderate), 25–50 cm (shallow) and <25%. Furthermore, the slope of land value is obtained from a digital elevation model downloaded from the SRTM-DEM with a spatial resolution of 30 m and a vertical accuracy of 14 m. The slope is categorized as gentle, gentle, moderate, steep and very steep. Soil values are obtained from various physical and chemical parameters. Physical properties are soil depth, texture and drainage. The chemical parameter is the pH value and is considered for land suitability analysis. The microclimate parameters are obtained from the meteorological department. The various individual thematic layers are generated in GIS for individual land characteristic analyses, namely soil depth, soil pH, soil texture, soil drainage, soil EC, groundwater availability, slope, Rainfall and temperature. The degree of limitation is from 1 (no or slight limitation) to 4 (very severe limitation), which are assigned to each land characteristic; limitation maps are prepared as in Figure 3. Figure 3 shows the collected data as a map using GIS software according to numerical rating, limitations and overlay thematic layers for visualizing the Land area. In this paper, weightages are never assigned and all the factors or layers are considered equally important. The limitation maps of land characteristics are spatially overlaid in the GIS environment and the resultant suitability layer is obtained. In GIS, the resultant polygon layer has nine limitation values. The suitability classes are estimated based on the number and intensity of the degree of limitation. The suitability classes are classified as highly suitable (S1), moderately suitable (S2), marginally suitable (S3) and never suitable (N), as shown in Table 3.

Figure 3. Sample elevation data-based mapping using GIS software according to numerical rating limitations and overlay thematic layers for visualizing land area.

Table 3. Criteria for mulberry cultivation land suitability classes.

Suitability class for mulberry cultivation (micro-climatic-based)	Criteria
Highly suitable (S1)	The land area has no or only four slight limitations
Moderately suitable (S2)	The lands area has more than four slight limitations and/or no more than three moderate limitations.
Marginally suitable (S3)	The lands area has more than three moderate limitations and/or one or more severe limitations.
Not suitable (N)	The land area has no very severe limitation

Descriptive analysis – Bayesian optimized GPR

Gaussian process regression (GPR) is a kernel-based probabilistic method used for understanding the uncertainty in land suitability for mulberry cultivation and improves prediction accuracy. Gaussian processes are an extension of multivariate Gaussians. The Gaussian process performs through mean and covariance functions. GPR estimation needs more samples for training, decreases confidence size and reflects uncertainty in the suitability of the mulberry plantation model. For the above problem, the Regression tree is a nonlinear regression method, where sub-dividing data into small sizes fits the models. However, machine learning algorithms such as SVM and ANN have hyper-parameters tuning to increase the accuracy and need nonconvex function evaluations. In this paper, Bayesian optimization optimizes any objective function and decreases the number of actual functions. Bayesian optimization and Gaussian processes are used in objective functions for sampled data. Bayesian Optimized GPR has a probabilistic proxy as objective, which consists of past data as training data. The Bayesian Optimized GPR model provides adequate information and evaluates the true objective function for better land suitability prediction. In this paper, hyper-parameters are tuned for a set of training paradigms. Exploration refers to the search for uncertainty, which is high, finds a set of parameters and improves model accuracy. Exploitation of region search is near to previously calculated high estimated values i.e. regression performance scores.

Descriptive analysis – PCA-enabled GPR

The PCA-enabled GPR method consists of Gaussian process mappings among latent space and observed data space. In this paper, PCA in output dimensions is assumed to be linear, independent and identically distributed and assumptions infringed for PCA-enabled GPR models. PCA-enabled GPR allows arbitrary rotation of the data and breaks the identically distributed through different covariance functions at the output dimension. PCA-enabled GPR focuses on the assumption and linearity through the replacement of inner product kernel, which consists of a covariance function. Non-linear functions are from non-linear latent variable methods. PCA-enabled GPR has a close relationship with the linear model and is interpreted as a non-linear probabilistic PCA. The PCA-enabled GPR model is applied to land suitability analysis for mulberry cultivation. For a query with less data needs the prediction of land suitability and the number of observations available will vary from the user end. So, an efficient algorithm is needed for the prediction of the land suitability through simple preprocessing and training patterns using a Predictive Analysis. In this paper, the Decision Tree is the proposed algorithm.

Predictive analysis – decision tree

A Decision Tree is a supervised algorithm, where data are split according to parameters. This algorithm consists of decision nodes and leaf nodes. The leaf nodes are the outcomes. The decision-making starts from decision nodes and leaf nodes. Each decision node has data attributes and they are classified. The leaf node is the result. Each path is a classification rule, and the set of classification rules is the set of decision tree expressions. In this paper, decision tree methods extract the classification rules among huge area of land area sites influencing parameters and land suitability. Figure 4 depicts a schematic diagram of the decision tree, which is used as a decision rule for site suitability analysis.

Figure 4. Schematic diagram of decision tree for land suitability prediction.

Initially, influencing parameters and suitability index are constructed and input for the decision tree method is derived. Furthermore, tree suitability is performed, and the output of the decision tree method is obtained. The decision tree method is selected as a knowledge rule for the extraction of land suitability, and extracted rules are the results. Influencing factors and suitability index are identified using the decision tree method. In this study, site conditions are with nine influencing factors, including soil parameters, such as Texture, depth, drainage, pH, EC and slope, micro-climate parameters such as Rainfall and Temperature and groundwater availability. The nine factors are input variables for the decision tree method. In the decision tree, the decision tree for the classification of land suitability of mulberry plantations is established. The decision tree was established for mulberry plantations for 13 layers and 18 leaf nodes were generated, among which five were non-suitable seven were marginally suitable, two were moderately suitable and four were highly suitable. Groundwater availability, texture, slope, drainage, depth and pH factors are used for non-suitable categories. The groundwater availability, soil texture properties, soil drainage properties, soil pH, terrain slope and soil depth properties play a major role in defining the unsuitable categories.

Predictive analysis – Pelican Optimization Algorithm (POA) – LSTM

The Pelican Optimization Algorithm (POA) simulates the hunting behaviour of pelicans. The POA optimizes the problems through different objective functions such as unimodal and multimodal. Unimodal functions have a high exploitation ability for optimal solutions. Multimodal functions have a high ability exploration to find the optimal search space. POA solves real-world problems, such as jamming resource allocation, and optimal demand for distribution networks. POA shows a proportional balance between exploration and exploitation, compared to existing optimization algorithms and provides optimal solutions. POA is used for tuning the hyperparameters of the LSTM model. The tuning parameters are the number of LSTM units, Number of layers, Learning rate, Dropout rate, Activation function, Batch size, Number of epochs and Optimizer. Optimizing the number of units improves the performance, increases in units and leads to overfitting. The number of layers shows the depth of the model. Deep model analysis of the complex patterns in the data increases in the depth leads to overfitting. Learning rate: The learning rate determines the step size of the optimization algorithm during training. A high learning rate can cause the optimization algorithm to overshoot the minimum, while a low learning rate can cause the optimization algorithm to converge slowly. Dropout rate: Dropout is a regularization technique that randomly drops out some of the LSTM units during training to prevent overfitting. The dropout rate determines the probability of dropping of a unit. The activation functions such as sigmoid, tanh and ReLU decide the output of each LSTM unit. Batch size decides the number of samples to be iterated for the optimization. More batch size improves algorithm performance. The number of epochs decides the times for iterates and innumerable epochs improve the performance. An increase in batch size leads to overfitting. Optimizer updates weights during the training process using different optimizers such as stochastic gradient descent (SGD), Adam and RMSprop. These hyperparameters of LSTM are tuned with POA and achieve optimal performance.

Results of land suitability analysis

In this paper, land suitability analysis for mulberry cultivation is based on soil properties, topographic parameters, groundwater availability and micro-climate parameters. The thematic maps for all the factors and their limitations are mapped. The descriptive statistics of the soil properties, topographic parameters, groundwater availability and microclimate parameters are influencing factors in this study, which is based on different areas of soil parameters, as shown in Table 4. The area of Krishnagiri district is 5125.85 sq. km. The land available for suitability analysis is 4803.142 sq. km, which excludes the water body, settlements and rocky areas. The soil depth varies from 18 cm to 203 cm and the average depth in this region is 93 cm. The mulberry plantation requires more than 50 cm soil depth and covers 83% area in the Krishnagiri district. The soil depth-based prediction of land suitability for different samples is tested such as deep, moderate, shallow and very shallow.

Table 4. Sample single query based on the soil parameters.

Sample query area to QPA	Input parameters based on soil	Area (Sq.km)	Random forest regression prediction % (Srikantaiah K. C. et al.)	Hunt's algorithm prediction % (Ramya M. C. et al.)	QPA method prediction % (BG-PL)
Query 1	Deep (>100 cm)	2688.040	75.17	81.78	98.75
Query 2	Moderate (70–100 cm)	1303.095	74.14	85.47	97.14
Query 3	Shallow (50–70 cm)	166.302	79.12	88.97	97.99
Query 4	Very shallow (<50 cm)	645.705	74.12	89.14	98.14
		4803.142	79.23	87.12	98.77

The pH is an important factor for plant growth. In this paper, soil pH is between 4.2 and 8.3 and the mean value is 6.97. The soil pH is categorized into seven classes such as >9.5 (strong alkaline), 8.5–9.5 (moderate alkaline), 7.5–8.5 (slight alkaline), 6.5–7.5 (neutral), 5.5–6.5 (slight acid), 5.5 −4.5 (moderate acid) and <4.5 (strong acid). The neutral, slight acid and slight alkaline are suitable for mulberry plantation, which covers 87% of the area. Table 5 shows the area occupied by these categories.

Table 5. Sample single query based on pH parameters.

Sample query to QPA	Input parameters based on pH	Area (Sq.km)	QPA method prediction % (BG-DT)	Random forest regression prediction % (Srikantaiah K. C. et al.)	Hunt's algorithm prediction % (Ramya M. C. et al.)
Query 1	6.5–7.5	2198.218	99.24	87.45	89.12
Query 2	5.5–6.5 & 7.5–8.5	1950.655	97.14	84.17	87.14
Query 3	4.5–5.5 & 8.5–9.5	547.194	95.12	89.47	88.47
Query 4	<4.5 & >9.5	112.075	95.45	78.85	86.47

Soil texture is a very important parameter for soil suitability which determines the quality of the soil. The clay loam, fine and coarse loamy and sandy fragmental types are commonly available in this region. The clay loam and fine loamy soils are most suitable for mulberry plantation, which covers 43% of the area. Table 6 shows a sample single query-based soil texture.

Table 6. Sample single query based on soil texture parameters.

Sample query to QPA	Input parameters based on soil texture	Area (Sq.km)	Random forest regression prediction % (Srikantaiah K. C. et al.)	Hunt's algorithm prediction % (Ramya M. C. et al.)	QPA method prediction % (PG-PL)
Query 1	Clay loam	1181.419	76.24	85.14	98.47
Query 2	Fine loamy	864.947	79.25	87.25	97.45
Query 3	Coarse Loamy	1722.327	75.78	84.17	95.23
Query 4	Sandy fragmental	1034.450	74.89	83.45	94.78

The soil drainage determines the types of crops that grow better in the soil and plants need good drainage for high yield. The study area is occupied with 61% of the soil, well and moderately-drained soils, which is suitable for the mulberry plantation. The land capability and land suitability are based on the topography of the land. Based on the percentage of slope, limitation categories are derived. The slope up to 5% is suitable for mulberry plantations and covers 59% of the study area. Table 7 shows samples based on soil drainage.

Table 7. Sample single query based on soil drainage parameters.

Sample query to QPA	Input parameters based on soil drainage	Area in Sq.km	Random forest regression prediction % (Srikantaiah K. C. et al.)	Hunt's algorithm prediction % (Ramya M. C. et al.)	QPA method prediction % (PG-PL)
Query 1	0–3% (Level)	2225.02	78.12	85.47	96.26
Query 2	3–5% (Gentle)	622.02	79.89	79.15	92.93
Query 3	5–10% (Moderate)	688.69	75.14	89.54	84.32
Query 4	>10% (Steep)	1273.62	74.48	85.42	96.48

Electrical conductivity is an important parameter for soil health, which influences mulberry plant health and crop yield. The whole area is suitable for mulberry plantations for electrical conductivity. Table 8 shows a sample single query based on rainfall parameters.

Table 8. Sample single query based on rainfall parameters.

Sample query to QPA	Input parameters based on rainfall	Area in Sq.km	Random forest regression prediction % (Srikantaiah K. C. et al.)	Hunt's algorithm prediction % (Ramya M. C. et al.)	QPA method prediction % (BG-PL)
Query 1	Very good	1124.400	74.23	87.23	98.24
Query2	Fair	3678.742	71.24	84.78	98.21

The groundwater availability is categorized into four groups: good, fair, moderate and poor. The mulberry plantation needs abundant water availability. The good and fair categories are suitable for mulberry plantation and cover 23% of the study area. Table 9 shows a sample single query based on groundwater parameters.

Table 9. Sample single query based on groundwater parameters.

Sample query to QPA	Input parameters based on groundwater	Area in Sq.km	Random forest regression prediction % (Srikantaiah K. C. et al.)	Hunt's algorithm prediction % (Ramya M. C. et al.)	QPA method prediction % (BG-PL)
Query 1	Good	239.29	75.21	85.24	96.98
Query 2	Fair	844.43	79.24	80.47	98.65
Query 3	Moderate	35.226	80.24	79.24	97.78
Query 4	Poor	3684.196	74.18	86.17	95.78

Table 10 depicts a single query based on rainfall parameters. The average rainfall in Krishnagiri district is 830 mm. The rainfall ranges from 750 to 850 mm, which comes under the moderately suitable class. The entire area is under this category.

Table 10. Sample single query based on rainfall parameters.

Sample Query to QPA	Input parameters based on rainfall range	Area in Sq.km	Random forest regression prediction % (Srikantaiah K. C. et al.)	Hunt's algorithm prediction % (Ramya M. C. et al.)	QPA method prediction % (BG-PL)
Query 1	750–850 mm	4803.14	71.25	78.23	98.56

The average temperature of Krishnagiri district is 35°C. The minimum and maximum temperature are about 20°C and 37°C, respectively. The entire area is highly and moderately suitable for mulberry plantations. Table 11 shows a sample single query based on temperature parameters.

Table 11. Sample single query based on temperature parameters.

Sample Query to QPA	Input parameters based on temperature	Area in Sq.km	Random forest regression prediction % (Srikantaiah K. C. et al.)	Hunt's algorithm prediction % (Ramya M. C. et al.)	QPA method prediction % (BG-PL)
Query 1	20–30 °C	4735.62	85.24	78.45	98.59
Query 2	30–37 °C	67.52	87.21	81.47	95.23

The advantages of Gaussian regression processes are usability, flexibility and powerful algorithm for regression and classification. BG-PL performs better due to the reliable estimation of uncertainty. BG-PL estimates of their uncertainty through direct consequence of Gaussian regression roots in probability and Bayesian inference. BG-DT can find accurate predictions for small data and single queries. BG-DT provides uncertainty for single queries. BG-DT adds prior knowledge about the shape of the model by selecting different kernel functions. Figure 5 shows the comparison of proposed QPA methods for land suitability analysis.

Figure 5. Comparison of proposed QPA methods for land suitability analysis.

This allows for more flexible and customizable modelling. BG-PL is non-parametric, which means more suitable for complex and non-linear relationships between input and output variables. BG-PL provides a unifying framework for many regression methods and allows for a more comprehensive understanding and analysis of these methods. BG-PL can learn long-term dependencies and handle long-term dependencies in data. BG-PL addressed the vanishing gradient problem. Gradient vanishing is the loss of information, which occurs during neural networks as connections recur over a longer period. BG_PL manages the gradient vanishing by ignoring the unwanted data/information in the network. BG-PL improves the method of backpropagation and allows the user to train models using sequences with several hundreds of time steps. BG-PL is more efficient at learning sequential data, captures complex patterns in data and leads to more accurate predictions for small amounts of data. BG_PL adjusts the layers and the size of the hidden state to improve the accuracy and is less prone to overfitting. Figure 6 shows the comparison of the QPA-based Proposed BG-PL method with existing methods.

Figure 6. Comparison of the QPA-based proposed BG-PL method with existing methods.

The limitation maps were overlaid in the GIS environment and the resultant polygon was graded based on suitability criteria. The statistics of suitability grades are in the table and the final output map of suitability grades is obtained. 79% of the area is not suitable for mulberry plantation, as shown in Table 12.

Table 12. Sample multiple queries based on the QPA algorithm for land suitability for mulberry plantations.

Multiple Query to QPA	Input parameters based on pH/water/soil/temperature/rainfall (Micro-climate)	Area in Sq.km	Random forest regression prediction % (Srikantaiah K. C. et al.)	Hunt's algorithm prediction % (Ramya M. C. et al.)	QPA method prediction % (BG-PL)
Query 1	Highly suitable	231.487	72.14	84.23	96.24
Query 2	Moderately suitable	78.703	79.21	81.48	98.14
Query 3	Marginally suitable	713.485	70.35	79.56	98.89
Query 4	Not suitable	3779.467	79.04	75.17	95.29

The proposed QPA method helps farmers for the prediction of the yield of a particular crop before planting. There are various methods for predicting yield on agricultural land such as Soil Testing, satellite imagery-based crop health, Crop Modelling, Yield Monitoring using sensors and Expert Opinion from agricultural specialists. However, the most effective method for predicting yield in agricultural land will depend on various factors such as the specific crop, soil type, weather conditions and the resources available to the farmer. A combination of the soil/water/pH and climatic data improves the accuracy of yield predictions through QPA-based land suitability prediction before cultivation. Table 13 shows the comparison of land suitability based on parameters for mulberry cultivation. Table 14 shows the Ground truth cost prediction on land based on the QPA method.

Table 13. Comparison of land suitability based on parameters for mulberry cultivation.

Input parameters	Reference	Area in Sq.km	Land suitability	Expert opinion from agricultural specialist (ground truth)	Prediction for yield analysis through land suitability (ground truth verification)
pH/water/soil/	Asgarian A et al. (16)	251.25	S2	S1	Medium
pH/water/soil/	Bodaghabadi et al. (15)	68.45	S3	S1	Low
pH/water/soil/	Safari et al. (14)	523.78	S1	S3	Low
pH/water/soil/	Asgarian A et al.	128.45	S2	S1	Medium
pH/water/soil/	Bodaghabadi et al.	845.12	S3	S1	Low
Micro-climate data	Proposed-QPA (BG-PL)	1457.28	N	N	High
Micro-climate data	Proposed-QPA (BG-PL)	143.25	S1	S1	High
Micro-climate data	Proposed-QPA (BG-PL)	987.25	S2	S1	Medium
Micro-climate data	Proposed-QPA (BG-PL)	2547.45	S3	S3	High

# Highly suitable (S1)/Moderately suitable (S2)/Marginally suitable (S3)/Not suitable (N).

Table 14. Ground truth cost prediction on land based on the QPA method for societal benefits.

Sl. No.	Particulars	Value (Rs.)	Proposed QPA method based predictions (2020–2021) (BG-PL)
A.	Operational costs
1	Farmyard manure (8 Tones)	5000.00	3500.00
2	Fertilizer cost @ (600 kg Ammonium sulphate; 300 kg Single super phosphate and 80 kg Murate of potash)	6538.00	3570.00
3	Manure and fertilizer application	2200.00	1000.00
4	Irrigation water cost	6000.00	3000.00
5	Irrigation	4600.00	1500.00
6	Weeding	4400.00	4400.00
7	Shoot harvest	8200.00	9500.00
8	Pruning and cleaning of plants	700.00	900.00
9	Land revenue	40.00	50.00
10	Miscellaneous	1000.00	1000.00
11	Interest in working capital	721.00	721.00
	Total leaf production cost (cost/kg of leaf Rs1 .64)	38999.00	29141.00

Discussions

Land suitability analysis helps the farmers in making decisions for the selection of the crop to grow in their piece of land and get maximum yield. The QPA-based BG-PL method suggests the farmers select the suitable crop for their area of land for growth based on the current climatic conditions and predicts the yield before the cultivation. The QPA-based BG-PL method is useful for farmers for maximizing crop productivity, reducing costs, improving land-use planning and enhancing farming efficiency. The QPA-based BG-PL method helps the farmers tailor their farming practices based on the selected crop for their piece of land, leading to increased efficiency and productivity. The QPA-based BG-PL method helps the farmers to optimize their land use. The major difference between the proposed method and the existing method is the dataset used for the land suitability analysis. In the QPA-based BG-PL method, micro-climatic data are used for prediction, whereas traditional methods use past data and datasets obtained from the district-wise are used for land suitability analysis. Moreover, current climatic condition-based data are used in the proposed QPA-based BG-PL method. In the proposed QPA-based BG-PL method, the descriptive analysis provides the data pattern and trends for current and past micro-climatic data. The predictive analysis is performed for land suitability analysis.

Conclusion

The influencing factors of land suitability for mulberry plantations in the Krishnagiri district and the advantages of the machine learning approach in site suitability analysis are analyzed. Based on land limitation, the study area is grouped into four categories: Slight, Moderate, Severe and Very Severe limitations for nine influencing parameters of soil properties, namely pH, Depth, Texture, Drainage, Groundwater availability, Topographical characteristics, namely Slope and micro-climate parameters, such as Temperature and Rainfall. Criteria-based suitability index maps and machine learning-based land suitability index maps are compared with field data. The QPA method provides 98% accuracy. The combination of the Decision tree algorithm, PCA-enabled GPR and Optimized GPR has more accuracy and is faster than the criteria-based method and Multiple regression method for the determination of land suitability index. QPA is based on BG-PL, which is the combination of descriptive and predictive analysis used for suitable land identification and crop yield prediction before starting cultivation. BG-DT, PG-PL and PG-DT methods of QPA provide suitable land identification with an accuracy of 90%, 85% and 80%, respectively. The obtained accuracy is higher compared to the traditional method of land suitability algorithms.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was funded by the North Eastern Space Application Center (NESAC), Meghalaya on ‘Application of Remote Sensing and GIS in Sericulture Development’.