Introduction

The rural–urban fringe is the boundary zone outside the urban area proper where rural and urban land uses intermix. It is an area of transition from agricultural and other rural land uses to urban use. Located well within the urban sphere of influence, the fringe is characterized by a wide variety of land use including dormitory settlements housing middle-income commuters who work in the main urban area, heterogeneity of land uses, very active social and economic life, instability and uncertainty in space layout, morphological conditions and density of the developed areas, changing social structure, and complex functional relations among traditional rural activities, industry, residential development, and recreational activities (Bryant et al. 1982; Thapa and Murayama 2008; Mitsova et al. 2011). Over time these characteristics of the rural–urban fringe are transformed into those of an urban system. But this transformation leads to significant trends toward urban sprawl in the rural–urban fringe and a resulting decrease in agricultural land area (Helbich and Leitner 2010; Arsanjani et al. 2013). Particularly in recent years, with rapid economic development and accelerated urbanization in developing countries, the obvious conflicts between agricultural land and non-agricultural land in the rural–urban fringe have resulted in serious environmental and social problems, such as a growing demand for construction land and protection of farmland, orchard and woodland, environmental pollution, reduction in biodiversity, fragile ecological environment, and disordered functional layout (Jiang and Yao 2010; Xu et al. 2012). Better planning and understanding of optimum agricultural land utilization while still considering environmental issues should be given priority in the rural–urban fringe, which may facilitate creating safe urban environments for human beings (Eyoh et al. 2012). Therefore, urban planners in developing countries should pay more attention to agricultural land practices in the rural–urban fringe. Such land could assist in improving the urban environment, generating employment, and reducing future food insecurity. Furthermore, a thorough understanding of agricultural land suitability levels and changes in the rural–urban fringe is the key to allowing effective land use planning and management.

Land suitability assessment takes into account the properties (e.g., climate, soils, and land cover) of a parcel of land so that recommendations can be made about realistic alternatives for improving its use (FAO 2007). It is a dilemma with multiple criteria and multiple objectives that produces a land use pattern, minimizes conflicts, and maximizes consensus among the stakeholders (Ettema 2011). According to guidelines issued by the Food and Agriculture Organization of the United Nations (FAO 2007), land suitability assessment is an integral part of land use planning and has been established as one of the preferred methods for supporting sustainable land use management. This assessment is a holistic, multidisciplinary approach to sound development and conservation and combines economic and social principles with environmental, agricultural, and biological sciences.

Today, most researchers are focusing on land suitability assessment. Many approaches have been used to assess the suitability of land, such as GIS-based approaches (Bojórquez-Tapia et al. 2001; Reshmidevi et al. 2009; Elsheikh et al. 2013; Feizizadeh and Blaschke 2013; Appiah et al. 2015), multicriteria decision analysis (MCDA) (Pereira and Dückstein 1993; Joerin et al. 2001; Chen et al. 2010; Rodriguez-Gallego et al. 2012), the analytical hierarchy process (AHP) method (Saaty and Vaargas 1991; Chen et al. 2010; Kumar and Hassan 2013; Akinci et al., 2013; Zhang et al. 2015), fuzzy logic techniques (Boroushaki and Malczewski 2010; Arsanjani et al. 2013), expert systems (Feizizadeh and Blaschke 2013), and artificial neural networks (Wang 1994; Gahegan 2003; Jiao and Liu 2004; Liu and Jiao 2005; He et al. 2007; Li and He, 2008; Yu et al. 2011; Wang and Mountrakis 2011; Egrioglu et al. 2013). It is important to point out that many case studies use a combination of these methods. For example, De la Rosa et al. (2004) designed the Mediterranean land evaluation information system by integrating neural networks, expert systems, and statistics. Xue et al. (2007) used fuzzy back-propagation neural network (BPNN) and fuzzy radial basis function neural network (RBFNN) to evaluate land resources in Guangdong Province. Reshmidevi et al. (2009) used a GIS-integrated fuzzy rule-based inference system for land suitability evaluation in agricultural watersheds. Chen et al. (2010) showed a spatial, simple, and flexible tool for assessing cropland suitability addressing the application of the new GIS-based AHP-SA (sensitivity analysis). Xu et al. (2011) evaluated the geo-environmental suitability of urban construction land using K-means clustering and BP neural network techniques. Eyoh et al. (2012) explored the implementation of a loosely coupled logistic regression model and geographic information systems in modeling and predicting future urban expansion of Lagos from historical remote sensing data based on the ArcGIS and MATLAB software. Feizizadeh and Blaschke (2013) investigated the optimal utilization of land resources for agricultural production in Tabriz County using GIS-based multicriteria decision analysis based on expert knowledge from stakeholders. Zhang et al. (2015) assessed land suitability for tobacco using GIS-based AHP and fuzzy set. All of these studies have been greatly important for land suitability assessment. However, most of these methods are based on knowledge and experiences of experts. And subjectivity is an issue with these technologies. Thus, there is a need to develop a method that keeps the knowledge-driven advantages, while reducing human participation. Therefore, BPNN and GIS methods have been adopted for their advantages in fast computing, self-organization, and unique adaptive capacity for the purposes of evaluation and forecasting in various fields (Paola and Schowengerdt 1997; Gahegan 2003; Yu et al. 2011; Song et al. 2014), including rural–urban fringe suitability assessment for agricultural land.

Hangzhou, the capital of Zhejiang province, is the provincial center in economy, culture, science and education, and the key city of the Yangtze Delta and the Grand Canal. It is located to the south of the Yangtze River Delta, west of Hangzhou Bay, at the lower reaches of the Qiantangjiang River, and the southern end of the Grand Canal; its location makes it an important transportation hub in southeast China. Hangzhou is in the subtropical transition zone, with hot and rainy summers and mild and fairly dry winters. The average temperature and relative humidity are 17.5 °C and 70.3 %, with an annual precipitation and daylight time of 1454 mm and 1765 h, respectively. It covers an area of 3068 km2 and has a population of 8.89 million in 2014. It has expanded rapidly as a metropolis for Hangzhou since the Drive to Reform and Open Up policy. The gross domestic product (GDP) of Hangzhou increased rapidly from 1.69 billion US dollars in 1978 to 150.4 billion US dollars in 2014. This growth has created a greater demand for rural–urban fringe land than in previous decades, leading to significant changes in the landscape, but it has resulted in potentially negative impacts for natural, social, and economic environments. For example, vegetable production in the rural–urban fringe may be threatened by a scarcity of land and labor resources unless alternative production technologies become available and the positive externalities generated by rural–urban fringe agriculture become internalized (Midmore and Jansen 2003). At the same time, a severe loss of agricultural land will bring about significant impacts on further economic development and social stability (Xu et al. 2012). Especially in the last few years, with the rapid growth of population and economy, the land use landscape pattern in the rural–urban fringe of Hangzhou has changed dramatically, which caused problems such as land degradation, soil erosion, environmental pollution, and ecological environment deterioration, which makes the rural–urban fringe of Hangzhou a very unstable fragile area (Huang et al. 2012). In this context, commissioned by the Hangzhou Geological Survey Bureau, suitability was evaluated for the agricultural land of the rural–urban fringe in Hangzhou to allow for effective land planning and use potential.

The goal of this paper was to assess the geo-environmental suitability levels of agricultural land in the rural–urban fringe by using BPNN and GIS on the basis of geo-scientific, geochemical, and geo-environmental assessment factors, along with the land use status of Hangzhou, as well as to consider the rural–urban fringe properties and urban sprawl of Hangzhou. This assessment integrated the following sub-factors: geomorphic type, slope, soil type, soil nutrients, soil pH value, soil pollution index, conditions of irrigation and drainage, degree of road accessibility, salinity of groundwater, and the degree of geological hazards of the Hangzhou. The assessment results could provide important scientific information to improve decision making for agricultural land planning and management, to ensure sustainable use of land resources in the rural–urban fringe in Hangzhou.

Materials and methods

Study area

The study area is the rural–urban fringe of Hangzhou, which has a total area of 2640.88 km2, excluding the developed areas, the Qiantangjiang River, and the West Lake. The northern, southern, and western portions of the study area consist of medium–low mountain terrain, whereas the eastern portion covers land near Hangzhou Bay, where the ground is low, flat, and covered with a dense river network. It has the typical characteristics of “Water Town in Southern Yangtze River” and “Capital of Silk” (Fig. 1).

Fig. 1
figure 1

Map of the study area in Hangzhou, Zhejiang Province, China

Full size image

Data sources

The data sources for this study include maps of Hangzhou (Table 1) and statistical data from the Zhejiang Statistical Yearbook from 1953 to 2014.

Table 1 Data sources for agricultural land suitability assessment in Hangzhou
Full size table

Technology and methodology

Geo-environmental suitability assessment of agricultural land is an important project in the rural–urban fringe. Many kinds of factors can influence the assessment quality of agricultural land, and these factors are also associated with each other. It is hard to adopt classical mathematical models to assess the quality of rural–urban fringe agricultural land because of the ambiguity of the factors and their boundary conditions. Therefore, the following flowchart is designed to objectively evaluate the rural–urban fringe agricultural land suitability level, as seen in Fig. 2.

Fig. 2
figure 2

Flowchart of suitability assessment for agricultural land based on the BPNN model

Full size image

Figure 2 shows a flowchart of suitability assessment for agricultural land based on the BPNN model. BPNN is a multilayer feed-forward perception neural network with the error inverse propagation training algorithm. It can learn and store a large number of input–output mapping relationships without mathematical equations for previously describing the mapping relation. The suitability assessment model for agricultural land based on the BPNN in this study includes an input layer, hidden layer, and an output layer. In the assessment process for agricultural land, the model does not need to illustrate the relationship between input and output sub-factors, as it can automatically extract the “reasonable” solution rules from training samples using its self-learning function, and it can automatically seek the relationships and construct mathematical equations to the suitability assessment according to data from the training samples. Its learning rule is to use the steepest descent method to adjust the weight and threshold of the network, which makes the mean squared error minimal. Current studies have proved that BPNN can achieve any complex nonlinear mapping function and that it was especially suitable for solving the challenge of complex internal mechanisms (Song et al. 2014). As a result, the BPNN model set up in this study can accurately reflect the agricultural land level for the rural–urban fringe in Hangzhou.

Establish geo-environmental suitability assessment factor index for agricultural land

The agricultural land suitability level is the result of the interactions among assessment factors, so proper selection of assessment factors is a premise for ensuring the accuracy of the suitability assessment results for agricultural land (Joerin et al. 2001). According to the uniformity and difference principle of rural–urban fringe agricultural land quality in Hangzhou, four groups comprising ten separate sub-factors of attributes were selected for the agricultural land suitability assessment: geomorphology, soil conditions, basic infrastructure, and geological environment. Among the four factor groups, geomorphology is the natural condition of urban development planning, as well as the most active leading factor in evaluating the suitability level of agricultural land. It includes two sub-factors: geomorphic type and slope (°). Soil condition refers to the characteristics, development, and utilization of the land, and to production levels. Its sub-factors include soil type, nutrients, soil pH, and soil pollution index. Basic infrastructure reflects the human activity, and its sub-factors include irrigation and drainage conditions and degree of road accessibility (m), which refers to the distance of human activity from roads. Geological environment is comprised of geologic hazards and groundwater salinization levels (mg/L). Geologic hazards mainly include landslides, collapse, and ground subsidence in the rural–urban fringe of Hangzhou. These four factor groups comprising ten separate sub-factors of geo-characteristics form a 2-level hierarchical structure according to their subordinate relationships, as seen in Fig. 3.

Fig. 3
figure 3

The 2-level hierarchical structure used in this study

Full size image

At the same time, agricultural land suitability was first classed into five levels: I, II, III, IV, and V, and then, the five levels were employed scores of 9, 7, 5, 3, and 1, respectively. Table 2 describes all data standardized according to agricultural land suitability and reclassified to appropriate categories based on data frequencies.

Table 2 Geo-environmental suitability assessment grading standards for agricultural land
Full size table

Data analysis based on GIS

The rural–urban fringe agricultural land-related assessment factors and sub-factors just mentioned were of different formats, scales, accuracies, and geometries and were almost unusable for direct application to this research. For this reason, these map data were converted into a vector-type spatial database using ArcGIS. At the same time, the corresponding attributes of the data were also collected in the database.

Geomorphic type is the material basis for agricultural land formation, and it plays an important role in water status, soil types, and properties in the rural–urban fringe of Hangzhou. Figure 4a shows that the geomorphic type was reclassified into five classes using ArcGIS from the 1:100,000-scale geomorphology and quaternary geology map: (1) plains region, (2) alluvial plain region, (3) low mountainous region, (4) valley plain area, and (5) valley area of mountains.

Fig. 4
figure 4figure 4

Ten input agricultural land conditioning factors used for the BPNN model: a geomorphic type; b slope; c soil type; d soil nutrients; e soil pH; f soil pollution index; g irrigation and drainage; h road accessibility; i geologic hazard degree; and j groundwater salinization

Full size image

Slope is one of the most important limiting factors for agricultural land use in the rural–urban fringe of Hangzhou, which greatly influences the value of agricultural land use and the level of productivity. A digital elevation model (DEM) describing the terrain is the key to producing slope, which has been used as input data layers for rural–urban fringe agricultural land suitability assessment. The DEM with 10-m interval contours and survey base points revealing elevation values were drawn from the 1:10,000-scale terrain map. From the DEM, slope was reclassified into five classes following the standard classification scheme set by the Urban Geology Survey Bureau of Hangzhou: (1) slope <2°, (2) 2–10°, (3) 11–15°, (4) 16–25°, and (5) >26° (Fig. 4b).

Soil is one of the most important factors for plant growth, and productivity of agricultural land depends on the level of soil fertility.

According to the Food and Agriculture Organization (FAO) soil classification, the dominant soil types are paddy soil, alluvial soil, red soil, calcareous soil, and coastal saline soil from the 1:100,000-scale soil type map in the study area (Fig. 4c). It is the soil nutrients that provide the elements necessary for plant growth, and it is an important comprehensive evaluation index for soil assessment. Based on comprehensive consideration of N, P, K, and organic C, the soil nutrients are divided into 5 levels. Figure 4d shows the soil nutrients for the rural–urban fringe of Hangzhou from the 1:100,000-scale soil type map. At the same time, soil pH value reflects the chemical properties of the soil, suitability of the crops, and the type of fertilizer. As a result, soil pH was reclassified into four classes according to the 1:100,000-scale environmental geochemical map: (1) 6.5 < pH < 7.5, (2) 5.5 < pH < 6.5, (3) 4.5 < pH < 5.5, and (4) 7.5 < pH < 8.5 (Fig. 4e). In addition, the soil pollution index from the 1:100,000-scale environmental geochemical map was reclassified into four groups, such as clean area (I), lightly polluted area (II), medium-polluted area (III), and heavily polluted area (IV) (Fig. 4f).

Irrigation and drainage illustrate the transformation and utilization of agricultural land by human beings, which has a direct impact on the productivity of the agricultural land for the rural–urban fringe of Hangzhou. According to the 1:100,000-scale revision of the hydro-geological map by Hangzhou Urban Geological Survey Bureau, the irrigation and drainage were reclassified into four classes: superior, good, medium, and poor (Fig. 4g).

Road accessibility is a concrete manifestation of the traffic conditions, which plays an important role in development and circulation of agricultural products for the rural–urban fringe. According to the 1:100,000-scale infrastructure map, a 2-km buffer zone is selected in the rural–urban fringe of Hangzhou for agricultural land suitability assessment (Fig. 4h).

Geologic hazard is another important factor responsible for the rural–urban fringe of Hangzhou for agricultural land suitability assessment. The geologic hazard was reclassified into four classes following the 1:100,000-scale zoning map of geological disasters, for instance, scarcely, low, medium, and high (Fig. 4i).

Nowadays, freshwater resources are becoming more and more limited because of increasing demand in the world. Thus, understanding the hydro-chemical characteristics and water quality is becoming critical for groundwater planning and management, to ensure sustainable safe use of the resources for drinking, agricultural and industrial purposes (Zhang et al. 2012). In this paper, according to the 1:100,000-scale revision of the hydro-geological map by Hangzhou Urban Geological Survey Bureau, groundwater salinization was reclassified into five classes: freshwater, freshwater above and saline water below, saline water above and freshwater below, slightly saline water, and saline water (Fig. 4j).

The entire set of agricultural land suitability assessment conditioning factors was converted to a raster grid with 250 m × 250 m units for the application of the BPNN model. The ArcGIS software was adopted as the basic analytical tool for spatial and attribute data management and analysis for the suitability assessment of agricultural land in the rural–urban fringe of Hangzhou.

Set up assessment model based on BPNN

BPNN is successfully used because of its advantages in strong nonlinear mapping ability, associative memory ability for input information, unique adaptive capacity, strong identification and classification capabilities for input samples, and self-organization in evaluating and forecasting purposes in various fields, including urban environmental quality evaluation (Gahegan 2003; Wang and Mountrakis 2011; Xu et al. 2012). The BPNN used in this study is made up of sets of nodes arranged in layers—an input layer, hidden layer, and an output layer—trained by the generalized delta rule (Rumelhart et al. 1986).

The BPNN model of agricultural land suitability assessment was established in this study, according to the rural–urban fringe geo-characteristics of Hangzhou (Fig. 5). Ten sub-factors were chosen for the input layers according to the assessment factors system; five neurons in the hidden layer were identified according to experiments by comparing 3, 4, 5, 6, 7, and 8 as hidden layer neurons; and one output neuron was decided according to the predetermined Hangzhou agricultural land suitability level for the rural–urban fringe. Therefore, there were ten neurons in the input layer, one neuron in the output layer, and five neurons in the hidden layer in this BPNN. At the same time, nodes in the same layer were not connected, nodes in different layers formed a focus-like connection path, and the weight value represented the connection strength between adjacent nodes in this model.

Fig. 5
figure 5

BPNN model of geo-environmental suitability assessment for agricultural land

Full size image

Five hundred and forty-nine training samples from a field investigation of the Hangzhou Urban Geological Survey Bureau were selected randomly, in considering a balanced regional distribution and uniform distribution of assessment factors and levels in order to train the BP network by using MATLAB. The plot of weight value and the score of each sub-factor as the value of the input layer neurons participate in the BPNN model of computing. The actual output value of the network was compared to the expected output value to calculate whether any error was within the tolerance. If the difference was within the tolerance, the network mapping result was considered the final result; if not, the input and output parameters of the model would be normalized as a group of new training samples and added to the original sample set to have network training again until the error was within tolerance. The expected output value was determined by the field survey at the 549 training sites. The optimal assessment BPNN was determined after the network models had repeated the debugging training and testing; at the same time, the convergence standard was the fastest, and prediction error was the smallest.

After repeating debugging training on the network model with the same 549 training samples, a BPNN, with 0.01 learning rate, 0.001 training goal, and maximum number of iterations that were terminated at 1000 epochs, was found to be the best, with the closest approximation to the function, the smallest error, and the shortest network training time. At the same time, the back-propagation algorithm was used to calculate the weights between the input layer and the hidden layer and between the hidden layer and the output layer by adjusting the number of hidden nodes and the learning rate. The final weights between layers obtained during training of the neural network and the contribution of each of the 10 factors were applied to assess the geo-environmental suitability levels for agricultural land in the rural–urban fringe in Hangzhou (Table 3).

Table 3 Weights of each sub-factor estimated by the BPNN considered
Full size table

Table 3 shows that soil type was the most important factor on agricultural land suitability levels in the rural–urban fringe of Hangzhou. Soil type showed the highest weight index, 0.1800. The second, third, and fourth important factors contributing to agricultural land suitability levels were soil nutrients (0.1560), soil pH (0.1380), and soil pollution index (0.1260), respectively. The fifth and sixth important factors contributing to agricultural land suitability levels were slope (0.1002) and irrigation and drainage conditions (0.0983). Geologic hazard (0.0512) and groundwater salinization (0.0512) play the next important roles in agricultural land suitability assessment in the rural–urban fringe of Hangzhou. Analysis of the weights indicated that the less important factors were geomorphic type (0.0501) and road accessibility (0.0491). Finally, the weights were applied to the entire study area and the agricultural land suitability levels for rural–urban fringe in Hangzhou were determined.

Results

The 54,649 assessment grids of Hangzhou agricultural land were input into the well-trained BP network, and their corresponding output parameters were acquired after network mapping. These output parameters could be used to classify the agricultural land suitability assessment results for the rural–urban fringe of Hangzhou. The assortment levels were input into the attribute table of assessment grid layers using ArcGIS. On the basis of the overall agricultural land assessment criterion defined by the Ministry of Housing and Urban–Rural Development of the PRC, ArcGIS would automatically determine the suitability level of an assessment grid with its re-classification function. Finally, agricultural land suitability assessment results were obtained (Fig. 6). Field investigation proved that the suitability assessment results for agricultural land were a good match with the current land use status in the rural–urban fringe of Hangzhou.

Fig. 6
figure 6

Result map of geo-environmental suitability assessment for agricultural land in the rural–urban fringe of Hangzhou based on BPNN and GIS

Full size image

Figure 6 shows the agricultural land suitability divided into four levels. Agricultural land suitability zone I has 560.59 km2, accounting for 21.23 % of the study area. It is sparsely spread outside of the built-up area and mainly spread in the eastern part of the Xiaoshan District, northern part of the Yuhang District, and the western and southern parts of the Hangzhou urban districts. This area has good soil conditions, excellent infrastructure, good geological environment, and very flat terrain.

Agricultural land suitability zone II has 837.66 km2, accounting for 31.72 % of the study area. It is mainly distributed in the eastern part of the Xiaoshan District and in most parts of the Yuhang District. This area has comparatively good soil conditions, good irrigation and drainage conditions, relatively high accessibility to roads, good geological environment, and relatively flat topography. At the same time, zone II is widely distributed and interspersed within the contiguous areas of level I.

Agricultural land suitability zone III has 975.06 km2, accounting for 36.92 % of the study area. This area, the largest, is mainly distributed in the eastern coastal areas of the study area and in the southern and northwestern hills and valley region of Hangzhou. It is located on mountainous land and has average soil conditions, average irrigation and drainage conditions, poor accessibility to roads, and a poor geological environment. The part located in the hills and valley has a steep slope. The eastern coastal areas of the study area, the plains regions, have poor soil conditions, but better irrigation and drainage and higher accessibility to roads.

Agricultural land suitability zone IV has 267.56 km2, accounting for 10.13 % of the study area. This zone is the smallest, and it is sparsely spread in the hilly region and the eastern part of the Xiaoshan District. It has poor soil conditions, poor basic infrastructure, poor geological environment, and steep slopes.

The evaluation result showed that the area of agricultural land in suitability zone III is the largest, while suitability zone IV is the smallest. The higher suitability levels (I and II) of agricultural land are mainly distributed in the plains areas of the eastern and southern areas of Xiaoshan District and in the rest of the Yuhang District except for the area to the northwest of the mountains. However, a lower suitability level (III and IV) of agricultural land is mainly distributed in the eastern coastal area of saline-alkali soil erosion by the seawater of the Xiaoshan District and the hilly areas of Xiaoshan District and Yuhang District.

Overall, the evaluation results proved that most areas of the rural–urban fringe of Hangzhou were suitable for agricultural land and that the current land use was appropriate. At the same time, the validity of the assessment result has been certified by field investigations done by the Hangzhou Geological Survey Bureau.

In addition, our results exposed many problems that one encounters when using a neural network for agricultural land suitability assessment. We eliminated any uncertainties that might originate from scene-dependent factors in a field investigation by the Hangzhou Urban Geological Survey Bureau. Our results indicated that a single hidden layer of the three-layer BPNN was sufficient to complete a given mapping task by approaching any mapping function. Our experiments also demonstrated that the increase in the neuron number of the hidden layer cannot improve the accuracy of BPNN (Paola and Schowengerdt 1997); instead, it needs to spend more training time. Therefore, five neurons of the hidden layer were identified according to experiments by comparing 3, 4, 5, and 6 as hidden layer neurons in this study, and the shortest time and the number of iterations for 1000. At the same time, our results also proved the characteristics of BPNN in agricultural land suitability assessment of rural–urban fringe in Hangzhou: strong nonlinear mapping ability, associative memory ability for input information, unique adaptive capacity, and self-organization.

Discussion

The goal of this study is to highlight the impact of the geo-scientific, geochemical, and geo-environmental assessment factors for agricultural land suitability and to find evidence for rapid urbanization in a target area that is representative of the rural–urban fringe in Hangzhou. The ten sub-factors have a clear influence on the actual land use pattern of the rural–urban fringe in Hangzhou. From these assessments, we find that most of the rural–urban fringe in Hangzhou is highly suitable for agricultural production, mainly because the surrounding area of Hangzhou belongs to the Hangjiahu plain; the elevation of the low-lying plain is only 3–6 m, and there are crisscross surface rivers and lakes which are rich in water resources and fertile soil. Field investigation demonstrated that the suitability assessment results for agricultural land in the rural–urban fringe of Hangzhou were a good match with the current land use status.

Soil condition is one of the most important factors for plant growth, whose weight is 0.5999 (Table 3), and the productivity of land depends on the level of soil type, soil nutrients, pollution index, and physical and chemical properties. It is these properties which determine the characteristics of the land resources and the development and utilization of the direction and level of productivity. There are large differences in the soil properties of Hangzhou because of its complex soil environment, such as paddy soil, alluvial soil, red soil, lime soil, and coastal saline soil.

Paddy soil is mainly distributed in the alluvial lake plain, the small part of the silt plain, such as in the plains region of Yuhang District, east and south of Xiaoshan District, and southwest of the major urban area of Hangzhou, where there are relatively fertile soil, weakly acidic, and a large amount of organic matter. It is suitable for paddy, wheat, cotton, rape, and so on. Red soil, mainly distributed in the hilly regions, western part of Yuhang District, and southern part of Xiaoshan District, has a significant enrichment of Fe–Al and is strongly acidic to acidic, and the surface organic matter content is not high. It is suitable for tea and fruit trees. Alluvial soil is mainly distributed in the plains and on both sides of the river valley, which has deep soil and good irrigation facilities, is nearly neutral in pH, and has low organic matter and nitrogen content. There is an urgent need to improve the soil properties for agricultural land, because it exists in some drought-prone areas with salinization problems. Lime soil, mainly distributed in hilly areas, the western part of Yuhang District, has abundant nutrients and better performance for fertilizers and is nearly neutral in pH. It is suitable for calcium-loving plants, such as ferns, miscanthus, and imperata. Coastal saline soil is mainly located in the coastal alluvial plain region and the estuarine area of the Qiantangjiang River, which has weakly alkaline to alkaline pH and low mineral content in general, and both the extent and degree of soil leaching were relatively low.

Overall, soil conditions are suitable for agricultural land in the rural–urban fringe of Hangzhou. There are significant regional differences in soil conditions and the distribution is scattered; therefore, the distribution of agricultural land suitability level is relatively fragmented and has smaller contiguous areas, especially in agricultural land suitability zones I and II.

Slope is one of the limiting factors for agricultural land use in the rural–urban fringe of Hangzhou, whose weight is 0.1002 (Table 3), and it has a great impact on agricultural land values and productivity. With greater surface fluctuations and steeper slopes, the effect of soil erosion and its severity are more serious, and thus, it is not conducive to agricultural production in this area. On the other hand, the smaller the surface fluctuations, the better the land is for farmland irrigation and mechanization, and it is therefore more conducive to agricultural production in the rural–urban fringe of Hangzhou. From the above assessment, these results can be seen: (1) Agricultural crop production is very suitable in the coastal plain, where the slope of <10° has good conditions; (2) it is very easy to use machinery in most parts of the low mountainous and hilly regions, where slopes between 11 and 15° have suitable conditions for agricultural production by taking soil and water conservation measures, but it is difficult for mechanization and irrigation in these regions, which are suitable for a vegetable and agricultural base, producing food for the city and for the residents in surrounding areas; and (3) other regions in the mountainous areas are suitable for woodland and orchard land only because of the steep slopes.

Another limiting sub-factor is the irrigation and drainage condition, whose weight is 0.0983 (Table 3); it has an important impact on agricultural land use. The bases of the north subtropical crop plantings, such as rice, wheat, oil, and cotton in the Xiaoshan District, Yuhang District, and in the northeast part of the main urban district of Hangzhou, have good irrigation and drainage. On the other hand, the bases of tea, citrus, and other orchards in the low mountains and hilly regions of the rural–urban fringe in Hangzhou have poor irrigation and drainage. Those areas need to improve the conditions of irrigation and drainage to enhance agricultural productivity in the rural–urban fringe of Hangzhou.

In addition, geological environment condition is one of the important limiting factors for the agricultural land suitability assessment in the rural–urban fringe of Hangzhou. It is not suitable for economic crop production, but is suitable only for eco-agricultural development in areas of potential geological disaster. On the other hand, it influences agricultural development by groundwater salinization because it has a greater restrictive effect on crop varieties for field planting. It is the guide for rational distribution of agricultural production through full analysis of groundwater salinization, occurrence, and hydro-geological background and provides the scientific basis for planning of planting and forestry in Hangzhou.

Conclusions

GIS is a powerful tool for land use planners in their effort to make agricultural land development processes more efficient and attractive. Recently, BPNN has become popular for prediction and forecasting purposes in various fields, especially for complex issues in nonlinear fields for its self-adaptive learning in a given environment. Our results certified that BPNN and GIS can commendably capture the complex nonlinear relationship between assessment factors and suitability levels. At the same time, the weight of each sub-factor was calculated by a back-propagation algorithm, which reduced human error in data processing and provided the required data for the suitability assessment of the agricultural land in the rural–urban fringe of Hangzhou.

The results demonstrate the application of BPNN and GIS in evaluating agricultural land suitability and have been very successful based on the geomorphology, soil conditions, basic infrastructure, and geological environment. The model developed as part of this research provides the necessary information for planners and managers of Hangzhou to determine the optimum use of agricultural land. At the same time, this empirical scenario provides a fast, simple, cost-effective, and rapid agricultural land assessment framework for the rural–urban fringe which may help policy makers, urban and regional planners, and researchers who are working in developing countries.

The geo-environmental suitability assessment factors influencing agricultural land are complicated in the rural–urban fringe. Therefore, assessment factors, factor grading, and factor weight are critical to assessment results, whatever mathematical methods are adopted for the geo-environmental suitability assessment of agricultural land. A conclusion for the scientific and objective assessment is possible only when each assessment step and method is carefully evaluated and processed.