Site suitability analysis for potential agricultural land with spatial fuzzy multi-criteria decision analysis in regional scale under semi-arid terrestrial ecosystem

The study addresses sustainable agricultural production under increasing land-use pressure in Turkey, particularly Central Anatolia, where semi-arid conditions and anthropogenic factors (e.g., erosion, urbanization, industrialization) threaten fertile farmlands. With about 24 million ha of agricultural land nationally and significant risk of degradation, there is a need to evaluate land according to its potential using robust, data-driven methods. Land suitability assessment is a multi-criteria problem; while simulation-based quantitative models are data- and cost-intensive, qualitative approaches can encode expert knowledge but face uncertainty in criteria weighting. Integrating MCDA with GIS and fuzzy logic (FAHP) can handle uncertainties and multiple criteria. The research question is to delineate and map potential agricultural suitability classes across Central Anatolia using a fuzzy MCDA framework, supporting sustainable land use planning and soil conservation in a semi-arid ecosystem.

Land evaluation methods include qualitative expert-based and quantitative simulation models. MCDA methods, especially AHP, are widely used to assign weights via pairwise comparisons but suffer from imprecision in linguistic judgments. Fuzzy set theory (Zadeh, 1965) and FAHP (e.g., Buckley, 1985) integrate uncertainty in decision-making and have been applied in various domains (e.g., environmental assessments, risk ranking). Land suitability workflows commonly involve indicator selection, categorization/scoring, weighting, and aggregation. No universal land quality index exists due to geographic variability; thus, context-specific indices are required. Prior studies support the relevance of soil texture, organic matter, bulk density, pH, lime content, and terrain factors (slope, depth, erosion, parent material) for agricultural suitability. Natural Breaks (Jenks) is appropriate for classifying unevenly distributed suitability scores.

Study area: Central Anatolia Region, Turkey (30°01′07″–38°43′19″ E; 36°18′08″–41°07′11″ N), ≈195,013 km², encompassing multiple provinces. The region is semi-arid to semi-continental with low rainfall, steppe vegetation, and varied geology (Central Anatolian Crystalline Complex; metamorphic, magmatic, sedimentary units). Topography includes extensive plains and volcanic mountains; average slopes around 9.6%. Data collection and soil analyses: 4517 georeferenced surface soil samples (0–30 cm) were collected. Laboratory analyses included particle size distribution (hydrometer), pH and EC (1:2.5 soil:water), CaCO3 (volumetric), and organic matter (Walkley–Black). Bulk density was measured; summary statistics were computed (min, max, mean, SD, skewness, kurtosis, CV). Spatial data: 10 m DEM (for elevation, slope), digital geology and soil maps, CORINE 2012 LULC, hydrology, and long-term meteorological datasets. Interpolation: Spatial prediction of soil criteria (texture components, pH, bulk density, CaCO3, OM) was performed using multiple methods in ArcGIS 10.2.2: IDW (powers 1–3), RBF (thin plate spline, completely regularized spline, spline with tension), and kriging (simple, ordinary, universal) with spherical, exponential, and Gaussian variograms. Model selection used RMSE: the lowest RMSE indicated best performance. Criteria and classification: Nine criteria were selected—terrain: slope, soil depth, erosion (RUSLE-based), parent material; soil: organic matter, bulk density, texture, pH, CaCO3. Each criterion was categorized and scored 1–4 (1 = optimal suitability; 4 = low suitability) based on agronomic thresholds (Tables described in paper). Suitability classes were defined using Natural Breaks (Jenks) into five classes with index ranges: S1 (1.33–1.91), S2 (1.92–2.24), S3 (2.25–2.52), N1 (2.53–2.79), N2 (2.80–3.95). FAHP weighting: Three experts conducted pairwise comparisons using a linguistic scale converted to triangular fuzzy numbers (TFNs). Buckley’s geometric mean method computed fuzzy weights: r̃i = (⊗j ãij)^(1/n); w̃i = r̃i ⊗ (⊕i r̃i)^(−1). Defuzzification used Center of Area (COA) to obtain Best Non-fuzzy Performance (BNP) values, then normalized to weights. Reported normalized weights: slope 0.245; depth 0.217; erosion 0.155; texture 0.118; organic matter 0.091; bulk density 0.061; pH 0.048; parent material 0.041; lime (CaCO3) 0.024. Aggregation: Weighted Linear Combination (WLC) computed suitability: Si = (1/Σk wik) Σk wik aik, where aik is the standardized score for criterion k at location i. The resulting suitability index map was classified into S1–N2. Validation and mapping: Spatial distributions of criteria and final suitability were mapped; provincial summaries were derived.

- Soil property ranges (0–30 cm, n=4517): pH 6.40–9.47 (mean 7.60); bulk density 0.72–1.84 g/cm³ (mean 1.37); CaCO3 0.01–95.24% (mean 16.72%); OM 0.86–13.0% (mean 1.70%); clay 0.43–80.94% (mean 29.03%); silt 1.01–81.16% (mean 25.90%); sand 1.38–93.84% (mean 43.98%). Lowest CV: bulk density 1.11%; highest CV: CaCO3 95.23%. - Best interpolation models (lowest RMSE): OM—Simple Kriging (Spherical); bulk density—RBF Spline with Tension; pH—Simple Kriging (Gaussian); CaCO3 and texture—RBF Completely Regularized Spline. - FAHP weights (normalized importance): slope 0.245; soil depth 0.217; erosion 0.155; texture 0.118; organic matter 0.091; bulk density 0.061; pH 0.048; parent material 0.041; lime (CaCO3) 0.024. Slope is the most influential criterion. - Criterion spatial distributions (high-level): 83.6% of the region has very low erosion (0–5 t/ha/yr); 7% experiences severe to very severe erosion. Depth: 38.8% very shallow (0–20 cm), 28.1% shallow (20–50 cm), 13.5% moderately deep (50–90 cm), 19.6% deep (>90 cm). Slopes: 64.3% <12% (machinery threshold), 35.7% >12%. OM is mostly low to very low (≈84% ≤2%). Textures are predominantly medium to fine (≈97%). Over half the soils are slightly to moderately alkaline; lime content is generally moderate to very high (>30% area with ≥10% CaCO3). - Suitability outcomes (total 195,012.7 km²): S1 21,045.7 km² (10.8%); S2 38,876.1 km² (19.9%); S3 51,845.3 km² (26.6%); N1 58,722.3 km² (30.1%); N2 24,523.3 km² (12.6%). Thus, S1+S2 = 59,921.8 km² (30.7%); N1+N2 = 42.7%. - Spatial patterns: High suitability (S1–S2) concentrated in Konya, Aksaray, Nevşehir, Kayseri, Yozgat, Kırşehir. Low suitability (N1–N2) widespread in Sivas, Niğde, Çorum, Kırıkkale, and scattered southern areas. Key limiting factors: steep slopes and shallow soil depth. - Comparison with existing assessments: Relative to Turkey’s GDRS LCC (I–III suitable 33.4%, IV slightly suitable 12.4%, non-suitable 54.2%), this study finds slightly lower suitable area (30.7%), markedly higher slightly suitable area (26.6%), and lower non-suitable area (by ~11.2%), attributed to updated datasets and fuzzy-MCDA methodology. CORINE 2012 indicates ~40% agricultural land use; about 12% of current agricultural use occurs on marginal or unsuitable lands per this assessment.

The fuzzy MCDA framework effectively integrates multiple biophysical indicators and expert judgment under uncertainty to delineate agricultural suitability in a large semi-arid region. The dominance of slope, depth, and erosion in the weights aligns with their control over mechanization, erosion risk, and rooting/soil water storage. The mapping identifies substantial areas suitable for cultivation but also highlights extensive marginal and non-suitable zones where agricultural expansion would risk degradation. Discrepancies with CORINE LULC suggest that significant portions of current agriculture are occurring on marginal or unsuitable lands, underscoring potential sustainability risks. Compared to traditional LCC, the fuzzy weighting and WLC aggregation offer more nuanced gradations, increasing the slightly suitable category and reducing non-suitable extent, likely reflecting higher-resolution data and improved modeling of uncertainty. These results can guide land-use planning, prioritizing conservation on steep, shallow soils and targeting management (e.g., erosion control, OM enhancement) to upgrade marginal lands where feasible.

This study integrates FAHP-based MCDA with GIS and geostatistics to map potential agricultural land suitability in Central Anatolia. Approximately 30.7% of the region is suitable (S1–S2), 26.6% marginally suitable (S3), and 42.7% not suitable (N1–N2). Slope, soil depth, and erosion are the most influential constraints. The approach reduces uncertainty in expert-based weighting and provides actionable suitability maps for regional planning. Future work should integrate additional thematic layers (e.g., climate variability, water availability, socio-economic factors, market access, yields) to refine suitability and support sustainable intensification and land protection policies.

The assessment focuses on permanent biophysical characteristics and does not incorporate economic viability, infrastructure or market access, social and policy constraints, or detailed climatic variability beyond general regional characterization. Suitability thresholds and weights, while expert-informed, remain context-specific. Interpolation uncertainties and inconsistencies in some descriptive statistics may affect local accuracy. Integrating socio-economic data, finer-scale climate and management information, and validation against yield/production outcomes would improve generalizability and decision relevance.