彭家驹卫星的种类和用途（续）-农险论坛

彭家驹卫星的种类和用途（续）-农险论坛

彭家驹作者：Henri Douche
编译：王笑男、王克
责编：左璇
备注：王笑男为中国农业科学院农业信息研究所在读硕士研究生；王克博士为中国农业科学院农业信息研究所副研究员；左璇为中国农业科学院农业信息研究所在读博士研究生。
出处：法国再保险农业保险系列技术简报
本文有作者版权保护，转载请务必注明出处。
五、气象卫星是如何应用的？
5.1气象卫星的历史
在农业生产保险中，绝大部分的风险事故都是天气原因造成的。传统的损失评估方法——实地调查，能够在农田中对损失做出一个直接的评判，但这需要大量的基础设施。指数保险不需要进行实地考察，它依据的是能够代表生产情况的指数。实际损失和指数之间的相关性及其质量将决定基差风险的水平。想要获得一个较好的气象指数，初始气象数据的质量十分重要（参见“基差风险”）。
气象站是气象数据的重要来源之一。建于地面、用于观测特点的气象参数、观测频率较高，因为具有以上特点，气象站应该是天气保险的最佳数据来源。但是有些问题不得不提，第一个是站点的数量和位置，气象站通常建在远离农田的地方，或者建在不具有代表性的地段（如机场或其他建筑区）。这是一个需要重视的问题，因为如果被保农田和气象站的距离过远会使得指数产生严重的偏差。
第二个是数据的质量问题。要想得出精确的数据，气象站必须保持定期的维护和校准。尽管国际标准的气象站确实这样做了，但许多其他站点并未如此。因为这意味着要做大量的数据清理和质量管理，工作量十分庞大。
5.2（气象数据）从点到网
在气象站的专业分析中，气象数据被看作是一系列的点，每个气象站代表一个点并且每个点在地球上都有自己的地理位置。为了估计两个气象站之间某个特定位置的准确温度，就必须要使用插值法。当然，可使用的插值法有很多种。要注意的是，两个站点之间的距离越远，内插值的准确性就越低。
与气象站的数据是点数据不同，由于使用的测量仪器的性质不同，卫星数据是“网格”状的。大气层在水平和垂直两个维度上被分成多个网格，每个单元格被赋予一个具体的参数值。
有时候，当地法规也可能会限制气象站的使用。几十年来，气象学家一直都在使用卫星来获取天气数据。由于卫星上携带着各种不同的传感器，它们可以测量到地球上任意一点的、大气层不同维度的天气参数。无论是极地轨道卫星还是地球同步卫星，气象卫星都提供了获取所需气象信息的永久访问权。但是气象卫星同样也有一定的局限性，例如：不能对本地天气事件（如降水）进行准确的度量，或者在某些特殊用途中因分辨率太低而无法对山区进行观测（单个单元格内高度差太大）。
虽然气象指数并非适用于所有保险指数产品的灵丹妙药，它们确是保险行业忽视的好用工具。目前已有很多基于气象指数研发的产品案例，例如非洲风险能力机构开发的保险产品，以及印度的农作物气象保险方案。
大气模型采用相同的管理方法，因此可能会将观测站点和卫星两者的数据进行合并。
气象学家运用数据同化方案和大气模型，已开发出融合各种可用数据资源（无线电探控仪、卫星、地面站、飞行器以及太空船报告）的数据集。实际观测值和经过数据同化产生的模型模拟值都被用来进行数值化天气预报和再分析。与预测未来天气状况的数值化天气预报不同，再分析产品反映的是过去的天气状况，往往涉及到过去几十年。目前，数值天气预报和再分析使用到的数据大部分来自于卫星。但无论是用来校准卫星产品，还是作为气象模型（预测，再分析）的基础数据，一定量的地面观测总是有必要的。
与其他所有模型一样，再分析数据模型也有一定的局限性，但对于农业监测而言——其主要目的是描述作物整个生长季气象的累积效应，再分析数据模型是非常有价值的。
究竟是选择原始卫星技术，还是二手加工数据集，这取决于目标用途。例如，对于降水预测，卫星资料往往是更好的选择，因为降水（特别是对流性降水）是亚格子尺度模型的过程结果，这并不能由气象模型直接模拟得出。再举一个例子，通过再分析得出的空气温度往往比卫星数据提供的空气温度更准确，因为卫星只能测量到表层温度，想要获得某一参考高度的空气温度，就必须由所测量到的表层温度计算得来。
原文：
How are weather satellites used?
THE HISTORY OF WEATHER SATELLITES
When insuring agricultural production, the vast majority of the mainperils are weather-driven. Of course, traditional loss adjustment methods withfield visits directly assess losses in the field, but they require a heavyinfrastructure. Index insurance does not need field visits, it is based on anindex that can proxy production. The quality and the correlation between lossand index will set the level of basis risk. In the case of a weather index, the quality of the initial weather data is crucial to ensure a good product (see“basis risk”).
Weather stations constitute the first obvious source of weather data.Ground-based, and measuring parameters, on a very frequent basis, weather stations should be best source for weather-based insurance. But somelimitations must be mentioned, the first one being the number and location ofthe stations. Stations are often located far from agricultural fields, or inlocations that are not representative of the fields (e.g. airports and otherbuilt-up areas). This is a key issue, as the distance between the insured fieldand the weather station can introduce a severe bias in the index.
The second limitation is the quality of the data. In order to produceaccurate data, a weather station has to be maintained and regularly controlledand calibrated. While this is the case for international standard-level weatherstations, it is often not true of many other stations. This implies aconsiderable amount of data cleaning and quality control.
FROM POINT TO GRID…
In the terminology of weather stations, weather data is treated as anumber of points, each station representing a point and each point having ageographical location on the globe. In order to estimate the specifictemperature at a specific point between two weather stations, interpolationmust be performed. Of course, many different methods of interpolation can beused. But the greater the distance between stations, the higher the uncertaintyaround the interpolated value.
Unlike the “points” system used by weather stations, satellite data ishandled as a “grid”, due to the nature of the measuring instrument itself. Theatmosphere is divided into individual cells at horizontal level and verticallevel, and each cell is then given a value for a specific parameter.
In some cases, local regulations can limit the use of weather stations.
For decades now, weather scientists have been using satellites toretrieve weather data. Thanks to the different sensors onboard, they canmeasure weather parameters at any point on the globe and at different altitudesin the atmosphere. Whether they are polar-orbiting or geostationary, weathersatellites give permanent access to the required information.
But weather satellites also have certain limitations, such as notbeing able to provide accurate measurements for very local events likeprecipitation, or having special resolutions that are sometimes too crude to beused in mountainous areas (too much altitude variation within a single gridcell).
Although not the panacea for all insurance index products, theyconstitute a great tool that is often under-used in the insurance industry.There are various examples of products based on weather indices, such as thosecreated by the African Risk Capacity agency, and by the Weather-Based CropInsurance Scheme in India.
Atmospheric models are managed in the same way, hence the possibilityof merging data between stations and satellites.
Weather scientists have developed datasets, mixing all available datasources - radiosondes, satellites, ground stations, aircrafts, ship reports –and using data assimilation schemes and atmospheric models. The combination ofobservations and model output through data assimilation is used in numericalweather prediction and in reanalysis. Unlike numerical weather prediction,which estimates future conditions, reanalysis products reflect past conditions,often for several decades. Today, most of the input data in numerical weatherprediction and reanalysis comes from satellites. However, some groundobservations are always required, either to calibrate the satellite products,or as input in the meteorological models (prediction, reanalysis) themselves.
Like all models, reanalysis data models have certain limitations, butfor agriculture monitoring, where the main objective is to capture the cumulatedeffect of the weather throughout the crop season, they are a very valuableasset.
The choice between raw satellite technology and reanalysis datasetsdepends on what the resulting data is going to be used for. For example,satellite data is often a better choice for precipitation, becauseprecipitation (especially convective precipitation) is the result of sub-gridscale processes, which are not directly simulated in weather models. To take adifferent example, air temperature from reanalysis tends to be more accuratethan air temperature provided by satellite data, because satellites onlymeasure skin temperature, which then has to be converted to air temperature ata certain reference height.
END
农险论坛
关注农业保险价值传播