Document Type : Research Article
Authors
- zohreh mosleh 1
- Mohammad hasan Salehi
- azam jafari 2
- Abdolmohammad Mehnatkesh 3
- Isa Esfandiarpoor Borujeni 4
1 shahrekord university
2 soil science department shahrekord university, shahrekord, iran
3 Agricultural and Natural Resources Research Center
4 Vali-e-Asr University of Rafsanjan
Abstract
Introduction: There is a concern with assessment of land performance when used for specific purposes. Land evaluation analysis is considered as an interface between land resources and land use planning and management. However, the conventional soil surveys are usually not useful for providing quantitative information about the spatial distribution of soil properties that are used in many environmental studies. Development of the computers and technology lead to digital and quantitative approaches have been developed. These new techniques rely on finding the relationships between soil and the auxiliary information that explain the soil forming factors or processes and finally predict soil patterns on the landscape. Different types of the machine learning approaches have been applied for digital soil mapping of soil classes, such as the logistic and multinomial logistic regressions, neural networks and classification trees. To our knowledge, most of the previous studiesapplied land suitability evaluation based on the conventional approach. Therefore, the main objective of this study was to assess the performance of digital mapping approaches for the qualitative land suitability evaluation in the Shahrekord plain of Chaharmahal-Va- Bakhtiari province.
Materials and Methods: An area in the Shahrekord plain of Chaharmahal-Va-Bakhtiari Province, Iran, across 32º13′ and 32º 23′N, and 50º 47′ and 51º 00′E was chosen. The soils in the study area have been formed on Quaternary shale and foliated clayey limestone deposits. Irrigated crops such as wheat, potato, maize and alfalfa are the main land uses in the area. According to the semi-detailed soil survey, 120 pedons with approximate distance of 750 m were excavated and soil samples were taken from different soil horizons. Soil physicochemical properties were determined. The average of soil properties was determined by considering the depth weighted coefficient up to 100 and 150 centimeters for annual and perennial crops, respectively. Qualitative land suitability evaluation for main crops of the area including wheat, maize, alfalfa and potato was determined by matching the site conditions (climatic, hydrology, vegetation and soil properties) with studied crop requirement tables presented by Givi (5). Land suitability classes were determined using parametric method. Land suitability classes reflect degree of suitability as S1 (suitable), S2 (moderately suitable), S3 (marginally suitable) and N (unsuitable). Different machine learning techniques, namely artificial neural networks (ANNs), boosted regression tree (BRT), random forest (RF) and multinomial logistic regression (MLR) were used to test the predictive power for mapping the land suitability evaluation. Terrain attributes, normalized difference vegetation index (NDVI), clay index, carbonate index, perpendicular vegetation index (PVI), geology map, existing soil map (1:50000 scale) and geomorphology map were used as auxiliary information. Finally, all of the environmental covariates were projected onto the same reference system (WGS 84 UTM 39 N) and resampled to 50 * 50 m since the soil samples were collected with approximate distance of 750 m (1:50,000 scale). According to the suggested resolutions for digital soil maps, the pixel size 50 *50 m fits to a 1:50,000 cartographic scale. Training the models was done with 80% of the data (i.e., 96 pedons) and their validation was tested by the remaining 20% of the dataset (i.e., 24 pedons) that were split randomly. The accuracy of the predicted soil classes was determined using error matrices and overall accuracy.
Results and Discussion: The results showed that climatic conditions are suitable (S1) for wheat and potato whereas the most important limiting factors for maize and alfalfa were the average of minimum temperature and average temperature, respectively. Results demonstratedthat among the studied models, random forest showed the highest performance to predict the land suitability classes and subclasses. However, different models had the same ability for prediction. In addition, the overall accuracy decreased from class to subclass for all of the crops. The terrain attributes and remote sensing indices (normalized difference vegetation index and perpendicular vegetation index) were the most important auxiliary information to predict the land suitability classes and subclasses.
Conclusion: Results suggest that the DSM approaches have enough accuracy for prediction of the land suitability classes that affecting land use management. Although digital mapping approaches increase our knowledgeabout the variation of soil properties, integrating the management of the sparse lands with different owners should be considered as the first step for optimum soil and land use management.
Keywords
)
Send comment about this article