نوع مقاله : پژوهشی
نویسندگان
1 دانشیار بخش تحقیقات حفاظت خاک و آبخیزداری، مرکز تحقیقات و آموزش کشاورزی و منابع طبیعی فارس، سازمان تحقیقات، آموزش و ترویج کشاورزی، شیراز، ایران
2 استادیار بخش تحقیقات آبیاری تحت فشار، مؤسسه تحقیقات فنی مهندسی کشاورزی، سازمان تحقیقات، آموزش و ترویج کشاورزی، کرج، ایران
3 مربی مرکز تحقیقات و آموزش کشاورزی و منابع طبیعی فارس، سازمان تحقیقات، آموزش و ترویج کشاورزی، شیراز، ایران
4 کارشناس ارشد سازمان جهاد کشاورزی، شیراز، ایران
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Introduction and Goal
Given the vastness, variability, and dynamism of agricultural sector information in the country, and especially in Fars Province, updating and refining information related to agricultural activities is of particular importance. Agricultural land use in watersheds are among the most important forms of uses, playing a crucial role in both positive and negative aspects (destruction or protection) of the watershed. Numerous studies have been conducted on the classification of agricultural land using satellite images. Remote sensing is an efficient method for identifying crops to estimate potential harvest and managing agricultural fields. A wide range of image types has been used for variety applications in classification. Some image processing methods are more suitable than others for distinguishing land use and land cover categories, especially when images are classified with high-resolution. Additionally, a review of previous research has shown that, under certain conditions, the accuracy of object-based classification is greater than pixel-based classification in separating selected land use and land cover categories. In this study, to evaluate these two methods over an area as vast as Fars Province, GIS-based and remote sensing models were developed that can be used both for current watershed farm management and for updating information in the coming years. The accuracy of the mentioned methods was determined, and the possibility of integrating various stages into a user-friendly model was provided.
Materials and Methods
In this study, all areas under agricultural and horticultural cultivation in the watersheds of Fars Province were examined. Since the dates of maximum greening are different in different climates, in order to study the climatic composition of each county, the climate map of the province was merged with the boundaries of the counties. The climate map in this study was prepared based on the Demarton method and using data from climatological and synoptic stations of Fars Province. To investigate the crop patterns, statistics on the area under cultivation of crops that were planted simultaneously with wheat were also collected for most of the counties where wheat was grown. In this study, based on the goal of distinguishing agricultural land use and measuring the average size of agricultural plots (around 1 hectare in Fars Province), Landsat 8 and Sentinel 2 images, with a ground resolution of 30 and 10 m, respectively, were sufficient. This was because each hectare of land contains 9 and 100 pixels, respectively. In a comprehensive field activity, the map of rainfed agriculture and rainfed orchards was drawn by delineating boundaries on the ground. Then four stages of headquarters and field verification were conducted with the highest spatial accuracy. Using relevant software, first, geometric and then atmospheric corrections were performed on the Sentinel and Landsat images, converting their DN values to reflectance. Next, various vegetation indices were generated and evaluated to select the best one for delineating green cover, which served as the basis for field data collection, including polygons within agricultural land use. Then, supervised classification methods including single-date, multi-temporal, object-based, and pixel-based, were utilized to distinguish irrigated croplands and classify them. Rainfed croplands and orchards were separated through repeated field surveys Irrigated orchards were extracted from cadastral data of agricultural lands and refined using updated images. Based on theKappa coefficient and overall accuracy, the results of each category of agricultural land use were compared with the ground truth.
Results and Discussion
The results of atmospheric corrections on the image indicate a significant improvement in image metrics and visual clarity. In two land uses with fully green cover and fallow land uses, the values of all vegetation indices increased after atmospheric correction, while in the barren land and water catchment areas, the values decreased, bringing them closer to the expected range found in reliable sources. The highest accuracy of the green area map obtained from various vegetation indices was related to EVI and mSAVI, while mNDVI and GBNDVI were in the next ranks. The results of object-oriented classification showed the clear superiority of this method in separating agricultural lands from rangeland and barren lands. Furthermore, this method also allowed for the separation of two types of rangeland, weak and strong. The highest accuracy in map production was related to the vegetation indices EVI and mSAVI. The highest Kappa coefficient and overall accuracy were related to the support vector machine classification. The results of object-oriented classification showed the superiority of this method in separating agricultural lands from rangeland and barren lands. Additionally, the method provided the possibility of distinguishing two types of rangeland, weak and strong.
Conclusion and Suggestions
Among the various vegetation indices examined for distinguishing the green cover in agricultural lands, the results of using indices such as EVI and mSAVI, which tilized the green band in their equations, were more suitable. From the perspective of pixel-based algorithms, the best algorithm was SVM, followed by the Decision Tree, which ranked second. Until now, the spatial distribution and actual extent of rainfed agricultural and orchard land uses had not been determined by any traditional method or remote sensing due to its complexity. In this study for the first time, and at least in the Fars Province , it was carried out with satisfactory accuracy based on a combination of remote sensing and field surveys. The scale of these layers, based on Sentinel base images with a resolution of 10 m, is sufficient and suitable for use in watershed studies up to a detailed-executive scales Based on the findings of this study, it is recommended that the produced layers be used as a model for examining land use changes and illegal land grabbing of natural resources. Additionally, to achieve this goal, it is suggested that the digital boundary of national lands be integrated with the agricultural land use layers, and the boundaries of encroachment be extracted from them.
کلیدواژهها [English]
)