Land Suitability Analysis for Potato Crop in the Jucusbamba and Tincas Microwatersheds (Amazonas, NW Peru): AHP and RS–GIS Approach

: Agricultural productivity in the Peruvian region of Amazonas is being jeopardized by conﬂicts and inadequate land use, that are ultimately contributing to environmental degradation. Therefore, our aim is to assess land suitability for potato ( Solanum tuberosum L.) farming in the Jucusbamba and Tincas microwatersheds located in Amazonas, in order to improve land-use planning and enhance the crop productivity of small-scale farmers. The site selection methodology involved a pair-wise comparison matrix (PCM) and a weighted multicriteria analysis using the Analytical Hierarchy Process (AHP) on selected biophysical and socioeconomical drivers. Simultaneously, land cover mapping was conducted using ﬁeld samples, remote sensing (RS), geostatistics and geographic information systems (GIS). The results indicated that for potato crop farming, the most important criteria are climatological (30.14%), edaphological (29.16%), topographical (25.72%) and socioeconomical (14.98%) in nature. The ﬁnal output map indicated that 8.2% (22.91 km 2 ) was highly suitable, 68.5% (190.37 km 2 ) was moderately suitable, 21.6% (60.11 km 2 ) was marginally suitable and 0.0% was not suitable for potato farming. Built-up areas (archaeological sites, urban and road networks) and bodies of water were discarded from this study (4.64 km 2 ). This study intends to promote and guide sustainable agriculture through agricultural land planning.


Introduction
Potatoes (Solanum tuberosum L.) are a vital source of carbohydrates, proteins, vitamins and minerals, thus holding a fundamental role in the food chain [1]. In 2018, potato cultivation areas in China occupied approximately 5 million hectares, representing more than 20% of the world's total potato farming area and making China the world's top potato producer [2,3]. Peru, is the main potato producer within Latin America, and it is home to the largest diversity of potato varieties (about 3000 of the 4000 existing varieties) [4]. Moreover, the gross value of production (GVP) of potatoes represented km 2 ), corn, legumes, wheat and barley [38]. Within Luya, the Jucusbamba and Tincas microwatersheds cover approximately 278.03 km 2 and are the most representative potato-producing areas. However, due to the lack of spatial planning for agricultural activities, the area is threatened by deforestation, land use conflicts and soil degradation, jeopardizing productivity [9,39]. According to the CTCUM, the microwatersheds possess land suitable mainly for pastures (147.97 km 2 ), permanent crops (51.95 km 2 ), silviculture (49.16 km 2 ), protection/conservation (24.81 km 2 ) and clean/intensive crops (2.88 km 2 ) [40]. Additionally, the study area has three microclimates (humid temperate cold, warm temperate slightly humid and warm temperate humid in the upper, middle and lower parts of the basin, respectively) [41], with mean annual temperature and rainfall of 15 °C and 1015 mm, respectively [42].  Figure 2 shows the applied research framework for land suitability analyses with respect to potato (S. tuberosum) cultivation using a AHP and RS-GIS approach in the Jucusbamba and Tincas microwatersheds. To summarize, (i) criteria that could determine potato growth and production were identified and selected; (ii) these criteria were mapped using field and RS data in a GIS environment; and (iii) the criteria were compared to determine their degree of importance using AHP. Then, using GIS, maps were integrated and reclassified according to suitability thresholds using weighted overlay.  Figure 2 shows the applied research framework for land suitability analyses with respect to potato (S. tuberosum) cultivation using a AHP and RS-GIS approach in the Jucusbamba and Tincas microwatersheds. To summarize, (i) criteria that could determine potato growth and production were identified and selected; (ii) these criteria were mapped using field and RS data in a GIS environment; and (iii) the criteria were compared to determine their degree of importance using AHP. Then, using GIS, maps were integrated and reclassified according to suitability thresholds using weighted overlay.

Identification of Criteria and Subcriteria
The identification process of suitable land for potato cultivation is based on favourable criteria with regard to crop growth and production. Criteria selection was carried out using technical manuals [43][44][45] and crop suitability analysis articles [20,21,46,47]. Subsequently, a hierarchy composed of four main criteria (climatological, edaphological, topographical and socioeconomical) and sixteen subcriteria was made, based on the accessibility and relevance of the data within the biophysical and socioeconomic conditions of the study area ( Figure 3).

Identification of Criteria and Subcriteria
The identification process of suitable land for potato cultivation is based on favourable criteria with regard to crop growth and production. Criteria selection was carried out using technical manuals [43][44][45] and crop suitability analysis articles [20,21,46,47]. Subsequently, a hierarchy composed of four main criteria (climatological, edaphological, topographical and socioeconomical) and sixteen subcriteria was made, based on the accessibility and relevance of the data within the biophysical and socioeconomic conditions of the study area ( Figure 3).

Identification of Criteria and Subcriteria
The identification process of suitable land for potato cultivation is based on favourable criteria with regard to crop growth and production. Criteria selection was carried out using technical manuals [43][44][45] and crop suitability analysis articles [20,21,46,47]. Subsequently, a hierarchy composed of four main criteria (climatological, edaphological, topographical and socioeconomical) and sixteen subcriteria was made, based on the accessibility and relevance of the data within the biophysical and socioeconomic conditions of the study area ( Figure 3).

Second hierarchy
Third hierarchy Reclassification based on suitability:

Datasets Sources and Processing
Temperature and precipitation layers were obtained from the WorldClim 2.0 database, which contains the average monthly climate data from 1970 to 2000 with a spatial resolution of 30 s (about 1 km 2 ) [42]. Elevation, slope and aspect terrain layers were generated from the Digital Elevation Model (DEM, Agronomy 2020, 10, 1898 5 of 18 ALOS PALSAR, 12.5 m), downloaded from the Distributed Active Archive Center (DAAC) from the Alaska Satellite Facility (ASF) [48].
Land Use/Land Cover (LULC) layers were obtained using a Sentinel 2A image (Path/Row: 17MRP; 20 June 2020; Bands: 2-8A, [11][12], acquired from the Copernicus Services Data Hub platform operated by the European Spatial Agency (ESA). Image preprocessing (i.e., atmospheric correction and resampling at 10 m) was executed using the Sentinel Application Platform (SNAP). Afterwards, digital processing was done in ENVI 5.3, following the methodology established by Murillo and Ortiz [49] and CORINE Land Cover adapted for Perú [50]. Hence, LULC were organized into five classes: urban area, grasslands, pastures and crops, bodies of water and forests. To assess the thematic accuracy, a 195-point confusion matrix indicated a general precision of 97% and a Kappa coefficient of 0.96.
The road network of the Ministry of Transport and Communications (MTC) was used as a reference for road distribution [51]. This vectoral layer was updated through field explorations. Furthermore, distance to roads and rivers was calculated using the Euclidean distance tool. The base layer of the rivers was obtained from the national charts 13 g and 13 h, downloaded from the Ministry of Education (MINEDU) geoportal [52].
Edaphological layers were generated from data obtained by collecting 72 soil samples at a depth of 30 cm throughout the study area ( Figure 1). These 72 samples were calculated and distributed in accordance with the regulations of the MINAGRI [15], based on spatial data on physiography, life zones, geology and geomorphology obtained from the ZEE-A [53]. Such samples were analysed at the Water and Soil Research Laboratory (LABISAG) of the Research Institute for Sustainable Development of Ceja de Selva (INDES-CES), that belongs to the Toribio Rodriguez de Mendoza National University (UNTRM). The physical (sand, silt and clay content) and chemical (pH, Soil Organic Matter-SOM, Nitrogen-N, Phosphorus-P, Potassium-K, Electrical Conductivity-EC, and Cation Exchange Capacity-CEC) properties of the soil were analyzed. Subsequently, all physical chemical properties were interpolated using the Ordinary Kriging (OK) method (in ArcGIS 10.5) based on five semivariogram models (Spherical, Circular, Gaussian, Linear and Exponential) [21,24,25,27,32]. In order to determine the most suitable semivariogram for each property, the coefficient of determination (R 2 ), mean bias error (MBE), mean absolute bias error (MABE), root-mean-square error (RMSE) and the Student's t-distribution were used (see formulas in Quiñones et al. [54]). For this purpose, of the 72 samples, 14 (20%) were chosen for the validation stage due to their homogenous distribution. In case any property did not result in an acceptable performing model (see statistical interpretation in Quiñones et al. [54]), layers (250 m; average depths 0-5, 5-10 and 10-15 cm) from the SoilGrids global digital soil mapping system were considered [55]. Additionally, according to the classification given by the United States Department of Agriculture (USDA), soil texture data were generated by using sand, silt and clay content layers along with the QGIS raster calculator [56].

Standardization of Subcriteria Layers through Suitability Thresholds
In order to execute weighted overlay, the units of the subcriteria were standardized using the raster reclassification of their maps. [57]. For this purpose, each continuous map (subcriterion) was divided according to land suitability level [17,58]. The five levels which are most commonly used in this process are taken from "A Framework For Land Evaluation" [59], and are as follows: "highly suitable" (territory with minor limitations for agricultural use that do not affect productivity or significantly increase expenses), "moderately suitable" (territory with moderate limitations that reduce productivity or imply slight risks of soil degradation), "marginally suitable" (territory with severe limitations that reduce productivity or increase expenses that will only be marginally justified), "currently unsuitable" (territory whose limitations can be eliminated with technical means or costs, although these modifications are currently unthinkable) and "permanently unsuitable" (territory with serious limitations that are supposed to be insurmountable in the long term). However, in this study, as in many other studies [8,16], the last two levels (N1 and N2) were combined and four levels of land suitability for potato cultivation were determined (1-4), with 1 being "not suitable" and Agronomy 2020, 10, 1898 6 of 18 4 "highly suitable" [60]. Climatological, topographical, socioeconomical and edaphological criteria were standardized according to suitability thresholds reported in previous studies (Table 1) [33,44,46]. Lastly, standardized thematic maps for each subcriterion are shown in Figure 4.

Analytical Hierarchy Process (AHP)
The AHP is an accurate approach for quantifying the importance weight for each criterion and subcriterion; the sum of the importance weight of each hierarchy group (Figure 3) is equal to 1 [12]. Relative importance was estimated using pairwise comparison matrices (PCM) [31,66], completed by a group of experts based on Saaty's nine-point ratio scale [19] (Table 2). This group of experts was made up of researchers in potato cultivation from the UNTRM and the National Institute for Agrarian Innovation (INIA). Although experts' opinions are highly valued, their subjective preferences may lead to inconsistencies [67,68]. Therefore, a consistency ratio (CR) must be calculated (Consistency Index (CI) / Random Index (RI) of the matrix) in order to establish a maximum inconsistency value. RI is

Analytical Hierarchy Process (AHP)
The AHP is an accurate approach for quantifying the importance weight for each criterion and subcriterion; the sum of the importance weight of each hierarchy group (Figure 3) is equal to 1 [12]. Relative importance was estimated using pairwise comparison matrices (PCM) [31,66], completed by a group of experts based on Saaty's nine-point ratio scale [19] (Table 2). This group of experts was made up of researchers in potato cultivation from the UNTRM and the National Institute for Agrarian Innovation (INIA). Although experts' opinions are highly valued, their subjective preferences may lead to inconsistencies [67,68]. Therefore, a consistency ratio (CR) must be calculated (Consistency Index (CI) / Random Index (RI) of the matrix) in order to establish a maximum inconsistency value. RI is calculated for different "n" values (number of factors in the matrix) from a simulation of 100,000 matrixes (Table 3) [69]. Also, CI serves as a measure of logical inconsistency against expert judgment during peer-to-peer comparisons [70]. For this purpose, the matrix with the highest value (λmax) is greater than or equal to the number of rows and columns (n) (Equation (1)) [34]. Moreover, CR must be <0.1 (less than 10% inconsistency) to be acceptable. Matrices with two subcriteria are automatically assumed to be consistent and acceptable [68]. This was the case for the climatological matrix.

Weighted Overlay of Thematic Maps
A weighted overlay analysis using QGIS was carried out with standardized thematic maps (GRIDi) and weight importance (WEIGHTi) for each subcriterion i. Each raster-based layer (Figure 4) was integrated using Equation (2) [68] to generate final output land suitability maps for potato cultivation (GRIDresult). Table 4 shows the descriptive statistics of soil properties in the study area. Moderate variability was shown by pH, while other physical chemical properties displayed a high variability in their results. Furthermore, a negative skewness was observed for pH, Log P, CEC and Sand, and a slight positive skewness for EC, SOM, Log SOM, N, P, K, Silt and Clay. Fifty Ordinary Kriging (OK) spatial interpolation models with five semivariogram models were generated for each soil property. Performance statistics were calculated for all models and the model with the best performance for each physicochemical property was chosen ( Table 5). The Gaussian semivariogram model was better adjusted to different soil properties. Properties with extremely low R 2 (N, SOM and Silt) were excluded. Therefore, the silt layer was calculated by supplementing the Sand + Clay layer to 100, and N and SOM layers were obtained from SoilGrids. Lastly, thematic maps for each edaphological subcriteria are shown in Figure 5.

Subcriteria Importance Weights
Ten experts were consulted and 50 PCMs were constructed, one at the criteria level and four at the subcriteria level, for each expert. The average weighting for each climatological, topographical, socioeconomical and edaphological criterion and subcriterion is shown in Table 6. These were standardized and scored according to their influence on potato cultivation in the Jucusbamba and Tincas microwatersheds.

Suitability Submodels for Potato Cultivation
Suitability submodels were generated with the weighted overlap (Equation (2)) of the subcriteria for each hierarchical group ( Figure 6). The submodels that presented the largest "highly suitable" surface for potato cultivation were socioeconomical (96.41 km 2 ) and topographical (89.65 km 2 ) ( Table 8). In contrast, the submodels with the smallest "highly suitable" area were climatological (34.30 km 2 ) and edaphological (12.72 km 2 ). In order not to overestimate the suitability of the criteria maps, submodels and final model, the road network, the main urban centres, rivers and a 200 m buffer zone of archaeological sites (4.64 km 2 ) were excluded from the analysis.

Land Suitability Model
The final model of land suitability for potato cultivation was generated with the weighted superposition of the submodels (Figure 7). Of the microbasin area, 8.2% (22.91 km 2 ) was "highly suitable" for potato cultivation, "moderately suitable" (190.37 km 2 ) and "marginally suitable" (60.11 km 2 ) areas represented 90.1%, and there were no "not suitable" areas found.

Land Suitability Model
The final model of land suitability for potato cultivation was generated with the weighted superposition of the submodels (Figure 7). Of the microbasin area, 8.2% (22.91 km 2 ) was "highly suitable" for potato cultivation, "moderately suitable" (190.37 km 2 ) and "marginally suitable" (60.11 km 2 ) areas represented 90.1%, and there were no "not suitable" areas found.

Discussion
The coupled AHP and RS-GIS approach is widely used in agricultural land zoning processes. Nevertheless, in comparison to previous studies of the suitability of the territory for potato cultivation [20][21][22][23]46,47,63], for other crops (Brassica napus L. [11,25], Glycine max L. [11], Triticum aestivum L. [23,27], Zea mays L. [27,29], Oryza sativa L. [28], coffee [31], and Vicia faba L. [32]) or for agriculture in general [7,10,[33][34][35], we included a greater number of subcriteria. However, in this type of study, the number of subcriteria depends on the focus of the study and the availability of spatial data. For example, future studies may include economic subcriteria (cost benefit, productivity, crop return rate or others [71]) and social (level of family skills, availability of labour, access to information, poverty rate or others [24]) that were not considered here due to the lack of spatial data [68]. An

Discussion
The coupled AHP and RS-GIS approach is widely used in agricultural land zoning processes. Nevertheless, in comparison to previous studies of the suitability of the territory for potato cultivation [20][21][22][23]46,47,63], for other crops (Brassica napus L. [11,25], Glycine max L. [11], Triticum aestivum L. [23,27], Zea mays L. [27,29], Oryza sativa L. [28], coffee [31], and Vicia faba L. [32]) or for agriculture in general [7,10,[33][34][35], we included a greater number of subcriteria. However, in this type of study, the number of subcriteria depends on the focus of the study and the availability of spatial data. For example, future studies may include economic subcriteria (cost benefit, productivity, crop return rate or others [71]) and social (level of family skills, availability of labour, access to information, poverty rate or others [24]) that were not considered here due to the lack of spatial data [68]. An attempt was made to overcome this difficulty by generating larger-scale soil maps through field sampling and the application of geostatistics. Like other studies on the suitability of the territory for the cultivation of potato [21,22] or other crops [25,27,29,30,32], geostatistics were applied, being among the most suitable interpolation methods for the estimation of soil properties [32,72].
Ordinary Kriging (OK) interpolation was used and the Gaussian semivariogram model was best adjusted to different soil properties. For Tashayo et al. [29,30], Gaussian was also the best fit model, while for Ostovari et al. [25] and Pilevar et al. [27], it was Spherical. Although OK is the most widely used method and several studies [21,27,29,30,72] have reported that it is the best for mapping soil characteristics, Li et al. [73] noted that it has a smoothing effect and tends to neglect local variability. In this regard, the Regression Kriging (RK) and other more advanced methods take advantage of the spatial information of environmental covariates and exceed OK [73,74]. In future studies, it is suggested that more advanced spatial interpolation methods be implemented. In this study, we used models generated with low yields (Sand, K, EC and P) due to the lack of detailed secondary information for P and K, and because field sampling was considered better than the information at a smaller scale which was available for Sand and EC (in SoilGrids [55]).
Additionally, after an extensive literature review, only four studies [20][21][22][23] included an integrated AHP approach in land suitability analyses for potato cultivation. Among these, three [21][22][23] evaluated edaphological subcriteria; the study of Kamau et al. did not [20]. This study indicated that the most important criteria for potato cultivation are climatological (30.14%) and edaphological (29.16%), followed by topographical (25.72%) and socioeconomical (14.98%). Similar results were reported by Kamau et al. [20], who did not consider socioeconomic criteria; for their study, edaphological criteria were the most important, followed by climatological and topographic. The similar level of importance of climatological and edaphological subcriteria was due to the fact that both strictly determine the development of crops [75]. Topographic subcriteria are of similar importance, since in mountainous ecosystems, like our study area, the topography has a great influence on the microclimates and edaphic properties (erosion and nutrient cycling) [7,24]. Of the climatological subcriteria, precipitation (57.92%) was the most important in this study, followed by temperature (42.08%). For Kamau et al. [20], precipitation was also more important than temperature. The great importance of precipitation was due to the high water needs of the potato crop (1000-1200 mm, well distributed during the growing cycle) [76]. The deficit of irrigation technologies in the province of Luya and in Amazonas influences the dependence on precipitation for the cultivation of potatoes and other crops. This dependence leads poorly informed residents to start fires with the purpose of generating rains during the dry season of the year [77], leading to land degradation.
Topographical subcriteria (slope, elevation and aspect) were included in this study, unlike that of Kamau et al. [20], as they were taken into account in other agricultural zoning studies [12,26,33,35]. With regard to terrain aspect, this subcriterion strongly influenced solar light and microclimate temperature variations [78]. Slope (54.75%) was considered the most important topographical subcriterion in this study, since it influenced soil depth distribution, soil moisture, soil erosion, nutrient availability, LULC, etc. [79]. Therefore, Zolekar and Bhagat [7] suggested using a high-resolution DEM, which describes the topographic variations in detail and provides greater precision in determining the suitability of the territory for agriculture, especially in mountainous areas such as those in this study.
With regards to edaphological subcriteria, SOM (16.51%) and pH (14.49%) were the most important, followed by primary macronutrients (N 13.98%, P 13.43% and K 13.42%) and other subcriteria (CEC 9.52%, EC 9.42% and Texture 9.24%). This differs from other land suitability analyses, like that of Kamau et al. [20], who used four edaphological subcriteria, whereby texture and drainage had higher importance values, followed by depth and pH. Furthermore, this study differed from studies carried out by Singha et al. [21] and Keshavarzi et al. [22], who used eight different subcriteria, from which K, texture, Organic Carbon (OC) and pH obtained higher importance values. On the other hand, Bagherzadeh et al. [23] obtained higher importance values for CEC and OC within the ten subcriteria they analyzed. The number and the different edaphological subcriteria used by previous studies, as well as each local reality and the varied experience of the experts, contributed to determining the importance of the edaphological subcriteria used in this study [8]. In fact, in any study on the suitability of land for agriculture, knowledge of the physicochemical properties of the soil is indispensable [29].
Within the socioeconomic subcriteria, LULC (47.64%) stands out on the distance to highways (26.56%) and distance to the water network (25.80%); this may have been due to the conservationist approach of the experts. This is because in the province of Luya, agriculture and livestock have brought about large changes in LULC, which caused a loss of 60.21 km 2 between 2001-2018 (66-88% of the loss occurred in areas <1 ha) [80]. Additionally, the slightly greater importance of proximity to a road compared to proximity to a water network may have been due to the fact that the study area featured many river water resources, while it lacked good road access. However, this difficulty may be overcome with the paved road project that goes from Corral Quemado (exit to markets on the coast), passes through Cumba, Collonce, Cohechán, to Luya, and continues to Chachapoyas (capital of the region Amazonas) or to the markets of the coast (Chiclayo, etc.) and the jungle (Moyobamba, Tarapoto, etc.) [81].
In the Jucusbamba and Tincas microbasins, the area modeled as "highly suitable" (22.91 km 2 ) for potato cultivation was 454.5% greater than the potato area reported in the last national agricultural census of 2012 (4.13 km 2 , adding areas of the districts that include the microwatersheds [35]) and spatially coincided with only 29.7% (6.81 km 2 ) with the area classified as being suitable for crops (54.83 km 2 adding classes clean/intensive crops and permanent crops [37]) of the CTCUM of the microwatersheds.
Our methodology framework, that integrates AHP and RS-GIS in order to assess land suitability for potato cultivation, can be applied in various future studies with necessary adjustments. Likewise, this methodology can be applied to other crops that have nutritional, economic and environmental importance. Unlike CTCUM, a normative technical instrument that determines the general suitability of land, our framework makes it possible to analyze the suitability of land for specific crops, which could improve the zoning process. Furthermore, CTCUM is based on a Peru Life Zones map which is outdated, even at local scales. For its part, although the ZAE will be an advance in agricultural territorial planning in Peru because it can be used to evaluate specific crops, it is recommended that the weighting of the importance of the criteria through the AHP be implemented. This is important, since each agricultural reality (district, province, region or country) is different, and therefore, the criteria must be prioritized by local experts.

Conclusions
This study assessed land suitability for potato cultivation by coupling AHP and RS-GIS techniques and basing them on four main criteria (climatological, edaphological, topographical and socioeconomical) and sixteen subcriteria. A thorough, expert-based analysis showed that climatological and edaphological were the most important criteria. Additionally, the resulting socioeconomical submodel possessed the largest "highly suitable" area in contrast to the edaphological submodel. A final output land suitability map labeled 8.2% (22.91 km 2 ) and 68.5% (190.37 km 2 ) of the Jucusbamba and Tincas microwatersheds as "highly suitable" and "moderately suitable" for potato cultivation, respectively. This research will provide support in the decision-making process for sustainable agricultural production in the Jucusbamba and Tinas microwatersheds by giving local farmers tools to improve agricultural planning. Finally, this study is intended to promote and guide sustainable agriculture.