基于MODIS和随机森林的兰州市日最高气温和最低气温估算

近地表气温是衡量城市热环境的重要因素,城市气温被认为是各种城市问题的重要变量,然而小区域气象站严重不足限制了异质城市内气温的空间连续分布表示。为获得空间连续分布的近地表气温,利用遥感数据结合随机森林机器学习法来估算城市近地表空间连续分布的气温。本文以兰州市为研究区,利用MODIS传感器的前一天和当天8个不同时间点的地表温度数据,结合一系列影响因子,利用随机森林来估算城市的每日最高气温和最低气温(Tmax/Tmin)。由于8个时序的地表温度数据与Tmax/Tmin存在不同的相关关系,根据这种关系设计了输入不同地表温度数据的8个模型方案,利用实测气温对不同模型方案的结果进行验证,获得最佳方案估算的日最高气温和最低气温。结果表明,用随机森林模型结合遥感数据来估算城市日近地表气温是可行的,并且前一天的地表温度对气温影响较大,是估算气温的关键参数。

Estimation of daily maximum and minimum temperature of Lanzhou City based on MODIS and random forest

Near-surface temperature is an important factor for measuring the urban thermal environment. Urban temperature is considered to affect various urban problems. However,the meteorological stations in small regions are insufficient to enable sufficient recording of temperature across heterogeneous cities. Remote sensing data combined with the random forest machine learning method was used to estimate the continuous near-surface temperature of Lanzhou. The random forest approach uses the previous day’s surface temperature data from a MODIS sensor for eight different time points in the day,combined with a series of influencing factors,to estimate the daily maximum and minimum temperature (Tmax/Tmin). Because the eight time points of surface temperature data have different correlations with Tmax/Tmin,these different relationships are used to create eight different model schemes with different input surface temperature data,and the results of the different models are verified using the measured temperature to obtain the best estimates of daily maximum and minimum temperatures. We found that it is feasible to use the random forest model combined with remote sensing data to estimate the daily and near surface temperature of Lanzhou city,and that the previous day’s surface temperature has a large effect on the present temperature,making it a key parameter for estimating temperature.