基于遥感数据的区域近地面气温最高/最低值(8天平均)的估算研究

Air temperature plays an important role in many disciplines, such as meteorology, climatology, ecology, hydrology and public health. Traditionally, air temperature is observed at meteorological stations and the data are point-based. Although interpolation methods can be used to get spatial information on air temperature, the results is uncertain due to the sparse distribution of stations, city heat island influence and other factors. With the development of various meteorolgocial, hydrological, ecological models, the need for spatial information on air temperature becomes increasingly urgent and has driven more and more researchers to look for satellite-based methods. LST (Land Surface Temperature) is a parameter most closely related to air temperature and can be derived using satellite data. A lot of LST products can be easily accessed near real time. Many studies have attempted to estimate air temperature from LST and favorable results have been obtained. The objecitve of this researche is to estimate 8 day averaged minimum and maximum air temperature in China from MODIS/LST product (MOD11A2, MYD11A2, 8-day composite at 1 km spatial resolution) and other ancillary inputs. Three issues will be investigated. 1) Buiding Regression Tree models for air temperature prediction. Given the complex relationship between LST and air temperature and the bias of LST (only clear sky) compared to air temperature (all sky), many features will be considered in building the model, including cloud cover, vegetaion index, land surface water index, land cover type, elevation, solar zenith angle, viewing zenith angle, DOY etc. 2) Residual correction. Calcualte the prediction residual for each meteorological station and produce residual maps using interpolations methods. Add residuals to model predictions to ensure satellite-derived air temperature at meteorological stations are correct. 3) Results comparison. Compare interpolated grid of air temperature with satellite-derived air temperature at different spatial resolutions and analyze spatial pattern of the disparities between them across different scales.

近地面气温在气候、生态、灾害、公众健康等领域都至关重要。各种气候模式、估产模式、水文模式的迅速发展，对气温空间分布信息的需求日益强烈。站点测量气温在空间上不连续，无法满足应用需要。本项目旨在利用1公里分辨率8天合成的MODIS地表温度(LST）产品，估算同等时空分辨率的中国区域近地面气温最高/最低值。研究包括：1）建立基于回归树算法的地面气温估算模型。以LST、云出现频率、植被指数、地表水分指数、太阳高度角、卫星观测角、地表覆盖、高程、日序等因素作为输入特征，以回归树模型为训练算法，建立中国区域气温最高/低值估算模型。2）结合地面台站数据的模型订正。计算模型对站点气温最高/低值的预测残差，通过插值获取残差的面上分布，对模型结果进行修正，提高气温估算精度；3）将遥感估算气温与基于站点的插值结果在不同空间分辨率下进行对比，分析差异的空间分布及其随尺度的变化趋势。

近地面气温在气候、生态、灾害、公众健康等领域都至关重要。各种气候模式、估产模式、水文模式的迅速发展，对气温空间分布信息的需求日益强烈。站点测量气温在空间上不连续，无法满足应用需要。本项目旨在利用1公里分辨率8天合成的MODIS地表温度(LST）产品，估算同等时空分辨率的中国区域近地面气温最高/最低值。. 研究内容包括：中国区域遥感LST与气象站点观测气温数据之间的差异，以及两者的差异如何受到其他因素（季节、高程、植被、气候区、观测角度、晴空天数等）的影响；对比多种气温估算方法，确定最佳的估算模型；气温估算结果的空间均匀性分析；气温估算结果的应用示范。. 研究结果显示：. （1）白天地气温差明显高于夜间地气温差，且两者受不同因素的影响，白天地气温差主要受到季节、海拔、地表覆盖等因素的影响；夜间地气温差主要受到晴空天数、观测天顶角、地形指数的影响。. （2）在气温估算方面，数据挖掘算法的精度低于样条插值算法，尤其是对于最高气温Tmax而言，暗示气温作为一种空间均匀性较强的气象要素，插值方法具有更大的优势；综合考虑精度高和简单性的原则， 最佳的气温估算形式为：. Tair=spline(lat,lon)+dem+MeanLSTday+MeanLSTnight，.其中MeanLSTday和MeanLSTnight为同期多年平均的日间和夜间LST。夜间LST对气温估算模型的贡献超过白天LST，LST对冬春季气温估算的改善超过夏秋季气温，最低气温估算精度的改善超过最高气温。. （3）增加了LST信息的气温估算结果不仅精度高，在小尺度上（<30 km）也具有更强的空间异质性，特别是对于冬春季气温而言，城乡之间、林草地之间、湖泊和周围环境之间的气温差异得以明显体现。. （4）为对气温估算结果进行应用，开展了三个研究示范，西藏冬季气温升温速率计算、三峡地区气温直减率计算以及“火炉”城市评估，结果表明：西藏冬季气温上升速度明显高于LST的上升速度，西藏北部地区升温更明显，色林错升温明显；三峡地区气温直减率在夏秋季最高：约为5-5.5度/km，而在冬春季最低：约为4-5度/km，明显低于常用的气温直减率（6度/km）；基于遥感气温的“火炉”城市评估和其他研究结果相似，并可体现城市内部炎热程度分布。