Sea and Land Surface Temperature Radiometer (Sentinel 3): Pre-mission development of clear-cloud-aerosol classification

从2013年起，一系列名为海陆表面温度辐射计(SLSTR)的传感器将在欧洲卫星上运行。这些SLSTR将具有对地球表面和大气进行长期观测的独特能力，特别是在气候应用方面。SLSTR将从两个观察方向拍摄每个立交桥上的地球图像，而不是单一图像，这大大增加了可以从图像中推断出的科学信息。SLSTR的观测也将比大多数可比传感器的观测更准确。从SLSTR获得的科学信息的例子有：地表温度、火灾的发生和强度(烧毁森林和草原)、地表反射率(反照率和植被产品)以及大气中的烟尘和矿物粉尘的数量。使用目前的技术，这些数据的准确性将因不充分的“分类”而受到影响。解释：为了获得最好的效果，必须对图像的每个区域进行准确的解释，以确定是否存在烟雾、其他气溶胶或云层。即使对于人类专家来说，这有时也很困难，而且当前的软件技术甚至更不可靠。因此，我们建议为这一分类问题找到一个更好的解决方案，以最大限度地发挥SLSTR在观测陆地表面温度(LST)、火、地表反射率(反照率和植被产品)和大气气溶胶方面的科学效益。如果没有这个项目，SLSTR对这些参数的估计将在气候应用方面受到影响。我们将开发和证明有效的技术，将陆地上的图像分类为晴朗的天空、云层和升高的气溶胶(烟雾和矿物粉尘)。我们将通过建立一种基于物理的、已被证明对云/晴空区分有效的概率方法来实现这一点，并将通过先进的气溶胶模拟和拟合技术来增强这种方法。该项目将开发一种晴云气溶胶条件的多向贝叶斯分类器，满足LST、火灾、地表反射和气溶胶反演的不同需求。我们的目标具有重要的科学意义，因为这些参数在气候系统中的重要性，特别是对地球的辐射平衡和碳循环。在全球范围内进行准确和有代表性的天基观测对于充分了解这些过程并建立模型至关重要。现在也是开展这项工作的合适时机。假设成功，我们将努力确保从第一个SLSTR推出时起就使用新技术。这项工作还可能带来更直接的好处，因为新技术将使用现有的类似传感器的图像进行原型制作。因此，在过去的二十年里，新技术也可以用来改进对这些参数的估计。

From 2013 onwards, a series of sensors called Sea and Land Surface Temperature Radiometers (SLSTRs ) will be operational on European satellites. These SLSTRs will have unique capabilities for long-term observation of Earth's surface and atmosphere, especially for climate applications. SLSTRs will capture images of Earth from each overpass from two viewing directions rather than capturing a single image, which greatly adds to the scientific information that can be deduced from the imagery. SLSTR observations will also be more accurate than those of most comparable sensors. Examples of the scientific information that will be obtained from SLSTRs are land surface temperature (LST), occurrence and intensity of fire (burning of forests and grasslands), surface reflectance (albedo and vegetation products), and the amount of smoke and mineral dust in the atmosphere. Using current techniques, the accuracy of these will be compromised by inadequate 'classification'. To explain: for the best results an accurate interpretation has to be made for each area of the image as to whether there is smoke, other aerosols, or clouds present. This is sometimes difficult even for a human expert, and the current software techniques are even less reliable. So, we propose to find a better solution for this classification problem, to maximize the scientific benefit of SLSTR for observation of land surface temperature (LST), fire, surface reflectance (albedo and vegetation products), and atmospheric aerosol. Without this project, the SLSTR estimates of these parameters will be compromised for climate applications. We will develop and prove effective techniques for the classification of imagery over land into areas of clear sky, cloud-cover and elevated aerosol (smoke and mineral dust). We will do this by building on a physically based, probabilistic approach that has proven effective for cloud/clear sky discrimination , and which will be enhanced with advanced aerosol modelling and fitting techniques. The project will develop a multi-way Bayesian classifier of clear-cloud-aerosol conditions, meeting the different needs of LST, fire, surface reflectance and aerosol retrieval. Our objective is scientifically important because of the importance of these parameters in the climate system, particularly to Earth's radiative balance and carbon cycle. Accurate and representative space-based observations on a global scale are essential to adequate understanding and modelling of these processes. It is also just the right time to undertake this work. Assuming success, we will try to ensure that the new techniques are used right from the time the first SLSTR is launched. The work may also offer more immediate benefits, since the new techniques will be prototyped using images from an existing, similar sensor. So, the new techniques could also be used to improve estimates of these parameters over the last two decades.