| The properties and characters of clouds not only seriously affect the Earth's radiation budget and water cycle,but also demonstrate an important significance over air safety,air military activities and detection operations.So far,cloud height is the most important factor for useful use of aircrafts.The existing methods for obtaining the cloud base height have several disadvantages such as limited acquisition range,low cloud classification accuracy,and being difficult to truly estimate the cloud base height data under a multi-layer cloud case.This thesis studies basic properties and methods including cloud change rules,cloud classification,and cloud bottom height inversion techniques,so as to provide a technical support for practical applications such as flight safety,high-speed aircraft detection,imaging guidance guidance,and so on.The main contents are as follows.Firstly,the key influence of clouds on aircraft detection and imaging guidance is analyzed.Under the condition of existing cloud cover,the target radiation will be greatly disturbed because of the variance of lightwave transmission behaviors.If the spectral transmittance is very low,the infrared imaging system will be unable to accurately detect or identify and position target.In this thesis,a South China Sea region is as an example for statistically analyzing the cloud situations of the target area and then studying the variance characters of cloud layers through using active remote sensing satellite data.The researches demonstrate that the morning period in both spring and autumn are the best for aircraft,and the low cloud proportion will be decreased slightly with the increase of the multi-layer cloud proportion,and also the cloud base height increased slightly with the decrease of the geometric thickness.To solve the problem of the cloud categories being inconsistent and the low accuracy of cloud classification based on the existing extrapolation technology of the same cloud base height,a machine learning algorithms is used to train cloud classification models based on both the active and passive remote sensing satellite data.A cloud classification model is trained efficiently through utilizing the active and passive satellite remote sensing data and the XGBoost algorithm.By expanding the feature dimension,the accuracy of the cloud classification is improved remarkably on the basis of ensuring the consistency of the cloud categories,and then a 81.2% probability of the correct classification of the cloud layers is obtained effectively.In addition,the cloud base height of each regional cloud type can be counted based on the proposed method.The research results show that the height of all types of clouds is relatively uniform,and the height of the cloud base can be estimated directly based on the cloud type obtained according to the inversion operation,so as to remove the operation of finding a known cloud base height observation point with the same cloud type of the measured point.Finally,a model for predicting the lowest cloud base height based on machine learning algorithms is constructed so as to solve the defect that the cloud-based height inversion method for space-based active and passive remote sensing is difficult to estimate the height of the cloud base of multi-layer clouds.Considering the influence of several geographical factors and the non-negligible relationship between the cloud base height and the cloud layer number,an input optimization about the number of cloud layers is performed.The test results show that the proposed algorithm demonstrates an average absolute error of 817.76 m and an average relative error of 10.15% corresponding to the height of the cloud base under a single-layer cloud system,and continuously an average absolute error of 1245.64 m and an average relative error of 16.72% to the cloud base height in a multi-layer cloud system. |
PDF Full Text Request