|
کلیدواژهها
|
Land Use and Land Cover (LULC), Remote Sensing (RS), Geographic Information System (GIS),Change detection, Convolutional Neural Network (CNN), Machine learning algorithms, Random Forest(RF), Deep Learning, Supervised Classification, Python Language, Google Colaboratory, Google Earth Engine(GEE), Ranya district, Kurdistan Region of Iraq (KRI).
|
|
چکیده
|
The demand for timely and accurate information on land resources and natural ecosystems has increased due to rapid urbanization and its impact on climate change. Urban heat islands, a result of urbanization, require precise land cover classification to mitigate their effect in both developed and developing cities. Such classification enables the identification of changes in built-up areas. The complex dynamics of land use and land cover (LULC) changes have sparked extensive scientific debate due to their diverse environmental natural and anthropogenic impacts. Land use represents how land is utilized includes categories such as agricultural land, industrial land, wildlife management areas, urban and recreational areas. On the other hand, land cover refers to natural or man-made elements such as water, snow, grassland, deciduous forest, vegetation, and bare ground. Understanding LULC is essential in many geospatial applications, including urban planning, regional management, and environmental management. The relationship between land use and land cover can serve as a model for ecosystems. An ecosystem imbalance, such as a higher proportion of industrial land compared to agricultural land, reflects the impacts stemming from industrial activities. To maintain a sustainable ecosystem, it is crucial to monitor and identify changes in land use and land cover (LU/LC) as it is an essential part of understanding how human activities interact with the environment. This study highlights the significant changes in LULC observed in Ranya district of Kurdistan Region. The main objective is to integrate and apply various remote sensing (RS) and geographic information system (GIS) methodologies, along with deep learning and Machine learning techniques particularly CNN and RF Respectively, to create land use and land cover maps and detect changes. In this study, deep learning architecture (CNN) and machine learning (RF) will help to identify LULC changes from 2022 to 2023 and their impact on sustainable food and ecosystem production. The research methodology is based on spatial modeling using (RS, RF, GIS, CNN and Earth Engine) and it consists of four tasks: a) pre-processing sentinel-2 image and Collecting data using Arc GIS, Arc GIS Pro, and Google Earth Engine. b) Utilizing various Python code in the Google Colab environment, as well as trained RF classified image to classify Buildings and Non-Buildings. c) Preparing Data and Designing the CNN model to generate the LULC map and detecting changes. d) Evaluation of the classification results of both CNN and RF models and investigate the effects of land use and land cover (LULC) changes on ecosystems and food production. This thesis presents both algorithms (CNN, RF) to achieve accurate building detection from remote sensing images with optimal resolution, which is very important for urban development and digital mapping of urban areas. Experimental evaluations on various datasets within the Ranya district demonstrate the superiority of CNN over existing methods, especially RF. This model has the potential to predict future urban expansion, assisting government bodies and public welfare departments in urban planning efforts. Informed decisions can be made using this model to ensure a fair distribution of the population and facilitate the strategic planning of industrial projects. Additionally, initiatives such as green city planning with rooftop gardens can be implemented to mitigate the effects of climate change and reduce the urban heat island effect. The results demonstrate that CNN and RF are the most effective classification methods both achieving their precision, recall, and F1Score were 0.76, 0.97, and 0.85, respectively. These results were expected because the CNN model was trained on the RF model data. This approach highlights the potential of CNN and RF in facilitating efficient urban monitoring and urban planning processes. later We evaluated the CNN model in over 12 countries worldwide, obtaining the precision, recall, and F1 Score of 0.97, 0.99 and 0.98, respectively. This indicates that CNN outperforms RF, especially when dealing with large geographic areas. Over the past two years Buildings emerged as the dominant LULC covering (0.7331 km2 or 7.11%) of the area as reported in the CNN model. Challenges identified in the study area include expansion of built-up areas, conversion of agricultural land and decline of agricultural land, all of which contribute to reduced food production, regional income, drought, and environmental pollution.
|