Land Suitability Evaluation for Agricultural Cropland in Mongolia Using the Spatial MCDM Method and AHP Based GIS

The purpose of this study was to prepare a cropland suitability map of Mongolia based on comprehensive landscape principles, including topography, soil properties, vegetation, climate and socio-economic factors. The primary goal was to create a more accurate map to estimate vegetation criteria (above ground biomass AGB), soil organic matter, soil texture, and the hydrothermal coefficient using Landsat 8 satellite imagery. The analysis used Landsat 8 imagery from the 2016 summer season with a resolution of 30 meters, time series MODIS vegetation products (MOD13, MOD15, MOD17) averaged over 16 days from June to August 2000-2016, an SRTM DEM with a resolution of 30 meters, and a field survey of measured biomass and soil data. In total, 6 main factors were classified and quality evaluation criteria were developed for 17 criteria, each with 5 levels. In this research the spatial MCDM (multi-criteria decision-making) method and AHP based GIS were applied. This was developed for each criteria layer’s value by multiplying parameters for each factor obtained from the pair comparison matrix by the weight addition, and by the suitable evaluation of several criteria factors affecting cropland. General accuracy was 88%, while PLS and RF regressions were 82.3% and 92.8%, respectively.


Introduction
Science-based agricultural production has been developing intensively in Mongolia since 1960 [1]. Between 1960 and 1989 the total sown area increased from 267.1 to 846.1 thousand hectares. From 1989 the total sown area fell, reaching 165.0 thousand hectares in 2006 [2]. The sown areas rose steadily by 440.6 thousand hectares between the years 2006 and 2016. However, cropland remains 405.5 thousand hectares less than in 1989. In this same time period, the total population increased 3.19 times while the amount of sown area declined by half as compared with the population growth. There is a significant difference in vegetable consumption between the urban and rural population. Urban population vegetable consumption is double that of the rural population [3].
In 1960, 40.2% of the total population lived in settled areas. This increased to 66.4% by 2016. Population increase coupled with consumption increase resulted in an intensified demand for food. On the other hand, agricultural products, especially wheat and potato production, increased as a result of the national government crop development program. Nowadays, potato and wheat consumption needs can be fulfilled by domestic production. However, of the total vegetable consumption (not including potato), 40% -45% were imported [4].
The main vegetables imports (onion, garlic, cabbage, turnips and other root seed vegetables) increased from 5438.4 tons in 1995 to 64,107 tons in 2016, an increase of 11.7 times. Of these, 96% -99% were imported from China. Mongolia remains strongly dependent on food security from neighboring countries. In addition, soils of currently cultivated areas are degrading. The country is facing challenges (especially local governments and community groups) to identify new crop areas with enough capacity for cultivation.
We have previously studied this topic: "Land suitability evaluation for cropland based on GIS between 2014 and 2016", was funded by the Mongolian Agency of Administration of Land Affairs, Geodesy and Cartography. In our preliminary study we used small and medium scale digital thematic maps to analyze and assess land suitability for cropland. During the study it was recognized that there was a need to improve the accuracy of input data using high-resolution satellite imagery for future research [5].
Geographic information system (GIS) and remote sensing (RS) techniques have been broadly used in agricultural studies. Remote sensing can provide a timely and accurate picture of the agricultural sector, as it is very suitable for gathering information over large areas with frequency and regularity [6]. The derived information is used for qualitative and quantitative analysis within near real-time production forecasts as well as for the anticipation of food security problems within the framework of monitoring agriculture [7].

Objectives
The purpose of this study is to identify new crop areas with enough capacity for cultivation across the entirety of Mongolia. The specific objectives are as follows: • Prepare more accurate input data using high-resolution satellite imagery.
• Use the spatial MCDM method and the AHP GIS for land suitability evaluation for agricultural cropland.

Methods
A combination of Boolean and Fuzzy logic theory, the spatial multi-criteria decision-making method, the analytical hierarchical process (AHP), expert knowledge analysis, random forest (RF) and partial least square (PLS) regression were used.
The study's general procedure for land suitability evaluation had several phases ( Figure 1). The first phase was to define the objectives. The second phase was to select criteria, for which there are two kinds of factors and constraints [8].
The third phase was standardization of the criteria; the fourth phase was assessing the ranking and weights of the criteria; the fifth phase was to overlap the map layers; the sixth phase was accuracy assessment.

Creation of Constraint Map Using Boolean Logic Theory
Constraints can be expressed in the form of a Boolean (logical) [8]. Boolean logic can have only two outcomes, true (1) or false (0). A constraint factor is a discrete   [9]. Zero values are prohibited conditions, and 1 values are permitted conditions. Constraints in this particular study often include legal restrictions. These are current land-use policy restrictions. Condition assessments and prohibitions can be factors as well.
The Boolean logic method must assume there is a definite cut-off point, because there is no flexibility for assessing real uncertainty [10]. Boolean logic can't be used when environmental and socio-economic factors are imprecise and incomplete. Under uncertain situations, fuzzy (probabilistic) logic comes in handy [11].

Creation of Factor Map Using Spatial Multi-Criteria Decision-Making (MCDM) Method
A factor is a criterion that can determine the suitability of specific outcomes for activities under consideration [8]. In this study, the spatial MCDM method was used in the creation of factor maps. Suitability levels for each of the factors were defined; these levels were used as a base to generate the factor maps (one for each factor [12]. Land suitability evaluation is expressed by qualitative and quantitative parameters.
In this section a combination of the spatial MCDM-, and the Fuzzy method was used. The main objective of land suitability analysis is to select the most optimal areas for a specific purpose. Land suitability analysis is a multi-criteria decision-making process [11]. Land suitability analysis is an interdisciplinary approach that includes information from different factors such as environmental and socio-economic. A main advantage of the MCDM procedure is the decision rule relationship between the input and output map. The MCDM method is divided into 4 groups and 7 classes [13].
• Multi-attribute and multi-objective decision making methods based on an objective or attribute.
• Individual and group decision making methods based on the number of people involved in the decision making process.
• Decision making under certainty and uncertainty methods based on the situation under which decision-making is being done and the nature of the criteria.
• Spatial MCDM based on spatial data.
From these, multi-attribute, multi-objective and spatial multi-criteria decision-making methods have been widely used in land-use suitability analysis. The multi-objective methods are based on mathematical programming models, and the multi-attribute methods are data oriented [14]. Spatial MCDM is a process where geographical data can be combined and transformed into a decision [11].
The main purpose of the spatial MCDM is to solve spatial decision-making problems originating from multiple criteria. The integration of spatial MCDM techniques with GIS has considerably advanced conventional map overlay approaches with regard to land-use suitability analysis [11] [24]. The fuzzy set theory technique is one of the most commonly used techniques for improving upon imprecise, incomplete and vague information [25]. Fuzzy logic is like Boolean logic but more fuzzy. Mathematician Lofti Zadeh presented fuzzy set theory in 1965, illustrating a mathematically meaningful method to quantify the degree of uncertainty and imprecision of non-discrete data [26]. The main point was that fuzzy data are obtained using an array of fuzzy membership functions with values that range from "0" to "1" [27].

Standardization of Criteria
All criteria used in the analysis were measured with different measurement values. Different values of criteria needed to be transformed into common values [28]. In order to implement this objective, we used a criteria standardization procedure. We used a simple linear scaling equation based on the fuzzy set me- where: i E is value of standardized in pixel i, min X is the minimum value criteria, max X is the maximum value.

Assessing Ranking and Weights of Criteria
In land suitability analysis there must be an evaluation that ranks the relative importance of the criteria. In this evaluation many different factors such as geophysical, biophysical, climate, and socio-economic were ranked. We ranked each criterion based on conclusions from literature from professional experts. Next, came the important step of determining the weighting values for each criterion.
There are many different approaches for assessing the weight of criteria based on MCDM techniques such as ELECTRE-TRI [29], ordered weighted averaging [30], compromise programming [31], analytical hierarchy process (AHP) [32] [33] [34] and Fuzzy AHP [11] [24]. Sensitivity analysis [35] includes 3 different approaches such as one-dimensional weights, random weights and selected weights [36]. From these, the most widely used method in spatial multi-criteria where: X ij -normalized value of a pairwise comparison matrix; n-the order of the matrix; ij W -weight of the criteria. The consistency ratio (CR) indicates the probability, and that the matrix ratings were randomly generated. The consistency of the pairwise comparison matrix is expressed by the consistency ration index. When the CR exceed 0.1 the weighting value is disagreeable, and when the index value is estimated below 0.1, the weighting value is agreeable.
Herein, calculating the consistency index was applied to the following com- where: CI-consistency index; max λ -maximum eigen value, and n is the order of the matrix

Overlap of Map Layers
After describing weights values of the criteria concerning their importance for land suitability analysis, all criteria maps have been overlaid using suitability index. The formula used for calculating the suitability index of each layer was as follows: where, X i -values of the each criterion, W i -weight values of the each criterion, S i -suitability index.

Accuracy Assessment
Accuracy assessments for random forest (RF) and partial least square (PLS) regression were calculated and compared with field survey biomass and soil archive data obtained from the Information and Research Institute of Hydrology, Meteorology and Environment. The Institute is authorized to provide qualified nationwide data sets.

Study Area
The study area covers the entirety (1566.6 × 10 3 square kilometers) of Mongolia

Data Used
The main goal of the study was to create more accurate input maps using satellite imagery and ground measurement data such as soil humus, soil texture, soil permeability, agro-climate condition, and hydrothermal in land suitability evaluation for agricultural cropland. In order to implement this three different datasets were used; satellite data, biomass data from field surveys, and field survey soil data (Table 1). In the subsequent analysis Random forest (RF) and Partial least square (PLS) regressions were used.

Data Pre-Processing
The first step in processing the Landsat 8 satellite imagery was to calibrate the radiometric and atmospheric correction. Radiometric calibration is used to calibrate radiance, reflectance or brightness temperature in imagery analysis. Atmospheric correction was applied to eliminate the impact of the atmosphere, such as the amount of water vapor, distribution of aerosols, and scene visibility.
In other words, eliminating the impact of the atmosphere is a pre-processing step for analyzing images of surface reflectance. Atmospheric correction was implemented in the QGIS 2.18 SCP plugin, parameterized with a tropical atmos- RStudio. Data validations accuracy assessment RF and PLS were calculated to compare with field survey soil and biomass data. RF regression was chosen because RF is a statistical algorithm that is capable of synthesizing regression or classification functions based on discrete or continuous datasets [37]. RF and CDT regression analyses were performed in Salford predictive Modeler 8.0 software. We also used PLS regression because the main goal of PLS regression is to predict or analyze a set of dependent variables from a set of independent variables or predictors [38]. PLS can easily treat data from a large number of variables in each factor that is identified [39]. Finally, all vector data were converted to raster format and then, all raster format data were transformed to the same geographical coordinate system and spatial resolution (30 m). Thereafter, each criterion map was classified into five suitability classes applying the classification threshold values of each criteria and standard scores for the corresponding class obtained in Table 2.

Analysis
The analysis comprised of three phases; the development of criteria parameters in land suitability evaluation for agricultural cropland; the preparation of more accurate input data using high-resolution satellite image, and an integrated evaluation.

Develop Criteria Parameters for Land Suitability
Evaluation for Agricultural Cropland 6 main factors and 17 criteria for land suitability evaluation for agricultural cropland were selected. A criteria evaluation schema was then developed based on our own, and other countries practices, literature and expert knowledge ( Table 2, Table 3). The criteria evaluation were divided into two types, multi-variables (factor) and constraint criteria parameters. A constraint is restraint criteria and it serves to limit the alternative. The constraint can also be often represented the legal restriction. That will be the decision based on the current land-use policy. It can apply for land use constraints condition assessment such as determined by the sum of factors prohibiting the use. In this study, 9 constraints have been chosen and there are obtained range values 0 and 1. The land use constraints condition assessment determined by the sum of factors prohibiting the use. The constraint factor assessment of land use is true or false condition represent. Zero value is impossible, and 1 value is possible.   Be near to water reserve, but not in buffer zone (buffer zone with 500 m radius)

Prepare More Accurate Input Data Using High-Resolution Satellite Image
Complex natural factors are nearly impossible to express by quantitative and qualitative values with 100 percent conviction. In order to improve accuracy, various analytical methods and satellite images were used. In this section we attempted to estimate vegetation parameters, soil organic matter, soil texture, soil moisture, and agro-climatic conditions for the hydrothermal coefficient using  (Table 4).

Topography Factor Analysis
Topography is important for maintaining slope stability and is critical to the distribution of other variables at a regional and local scale (e.g. a steep terrain should not be tilled to prevent soil erosion). The factors of slope and elevation were chosen for analysis of the contribution of topography to land suitability.
The analysis used STRM DEM with a spatial resolution of 30 meters, which can be inverted from the remote sensing data. This was then classified into five map classes for slope and elevation by land suitability level (

Vegetation Factor Analysis
Stable natural vegetation growing areas can be represented as a habitat in areas with crop vegetation. Natural vegetation parameters can provide an additional source of information for regional agro-production use [43]. In the four abovementioned analyses, we obtained the 6 most important variables to evaluate AGB, the Cl green, simple ratio, NDVI, EV 2 , WDRVI and MSAVI2. We then calculated six vegetation indices using Landsat 8 to estimate AGB across the study area. The results are shown in Table 5. We explored the relationship between estimated AGB and MODIS time-series vegetation products (NDVI, LAI, GPP), to understand the major controls of estimated AGB. Our country on average, has a 5-month natural growing season (April to August). At about the end of April and the start of May the grass turns green. June is the primary period of grass growth. The growth slows down toward the end of August, then the grass begins to fade. Therefore, in this study the MODIS vegetation products (NDVI, LAI, and GPP) covering the period from the beginning of June to the end of August was used, ranging from the year 2000 to 2016. The performed regression analyses were used to evaluate the relationship between estimated AGB and the 17-year average MODIS vegetation products.

Soil Factor Analysis
Parameters of soil properties mirror the land suitability evaluation for agricultural cropland. The spectral response of soil is influenced by a number of soil related properties such as surface condition, particle size (texture), organic matter, soil color, moisture content, iron and iron oxide content and mineralogy [57]. It is also possible to obtain soil property estimations from remotely sensed images [58]. Several studies attempted to demonstrate the relationship between soil properties and reflectance data from satellite imagery [ ( ) SOC is the surface organic C; a, b, c and d are curve fit parameters ); R, G and B are wavelength ranges.

Agro-Climatic Factor Analysis
Agro-climatic factors establish a quantitative connection between vegetative processes of specific plants and their in situ atmospheric environment [63].
Mongolia has an extreme continental climate with great variation between the four seasons. It has long, cold winters and short summers, with more than 65% of its annual precipitation falling in the summer season. In the summer season,

Results
In this study a combination of constraint and factor analysis methods were used.
There were nine constraint factors and 17 criteria factors. All constraints can be represented with values of 0 or 1. Suitability levels between 0 and 5 were obtained for each of the factors. The levels were 5-highly suitable, 4-suitable, 3-moderately suitable, 2-unsuitable and l-highly unsuitable ( Table 2, Table 3).

Result of Constraint Factor Analysis Based on Boolean Logic Theory
Assessment

Result of Factor Analysis Based on the Spatial MCDM Method
A comprehensive analysis of the study area used six major factors (topography,    Table 6 shows the ranking of 17 factors based on a literature review and expert consultations, with the weights calculated using AHP based GIS. In this study we have estimated a CR = 0.089, suggesting that there was a reasonable level of consistency in judgement.

Result of Map Layer Overlay Analysis Based on Suitability Index
After weighing the importance of different criteria for land suitability analysis, seventeen criteria maps were overlaid using the suitability index.  The results of the integrated assessment of constraint and factor analysis are shown in Figure 6, and

Accuracy Assessment
Accuracy assessments used were the random forest (RF) and partial least square       Table 8.

Conclusion
Since 1960, the method of wholesale selection was used for cropland area. This was conducted based on a few parameters such as the general condition of the