Estimating daily maximum air temperature from MODIS in British Columbia, Canada
Estimating daily maximum air temperature from MODIS in British Columbia, Canada
复制标题
DOI：
10.1080/01431161.2014.978957
复制
发表时间：
2014-01-01
影响因子：
3.4
通讯作者：
Ho, Hung Chak
中科院分区：
文献类型：
--
作者：
Xu, Yongming;Knudby, Anders;Ho, Hung Chak
气温（T-a）是森林研究和管理的重要气候变量。由于气象站密度低、分布不均匀，传统的地面观测无法准确捕捉T-a的空间分布，特别是在地形复杂、局部变异性较大的山区。本文利用卫星遥感估算了加拿大不列颠哥伦比亚省每日最大T-a。收集2003年至2012年夏季（6月至8月）的Aqua MODIS（中分辨率成像光谱仪）数据和气象数据来估算T-a。选择九个环境变量（地表温度（LST）、归一化植被指数（NDVI）、修正归一化水分指数（MNDWI）、纬度、经度、距海洋的距离、海拔、反照率和太阳辐射）作为预测变量。对观测 T-a 与空间平均遥感 LST 之间关系的分析表明，7 x 7 像素大小是根据 MODIS 数据估计 T-a 的统计模型的最佳窗口大小。使用两种统计方法（线性回归和随机森林）来估计最大 T-a，并通过逐站交叉验证来验证其性能。结果表明，随机森林模型比线性回归模型（MAE = 2.41 摄氏度，R-2 = 0.64）获得了更好的精度（平均绝对误差，MAE = 2.02 摄氏度，R-2 = 0.74）。基于7×7像素尺寸的随机森林模型，推导了2003-2012年夏季不列颠哥伦比亚省1 km分辨率下日最大T-a，并讨论了该地区夏季T-a的空间分布。令人满意的结果表明该建模方法适用于地形复杂的山区气温估计。
Air temperature (T-a) is an important climatological variable for forest research and management. Due to the low density and uneven distribution of weather stations, traditional ground-based observations cannot accurately capture the spatial distribution of T-a, especially in mountainous areas with complex terrain and high local variability. In this paper, the daily maximum T-a in British Columbia, Canada was estimated by satellite remote sensing. Aqua MODIS (Moderate Resolution Imaging Spectroradiometer) data and meteorological data for the summer period (June to August) from 2003 to 2012 were collected to estimate T-a. Nine environmental variables (land surface temperature (LST), normalized difference vegetation index (NDVI), modified normalized difference water index (MNDWI), latitude, longitude, distance to ocean, altitude, albedo, and solar radiation) were selected as predictors. Analysis of the relationship between observed T-a and spatially averaged remotely sensed LST indicated that 7 x 7 pixel size was the optimal window size for statistical models estimating T-a from MODIS data. Two statistical methods (linear regression and random forest) were used to estimate maximum T-a, and their performances were validated with station-by-station cross-validation. Results indicated that the random forest model achieved better accuracy (mean absolute error, MAE = 2.02 degrees C, R-2 = 0.74) than the linear regression model (MAE = 2.41 degrees C, R-2 = 0.64). Based on the random forest model at 7 x 7 pixel size, daily maximum T-a at a resolution of 1 km in British Columbia in the summer of 2003-2012 was derived, and the spatial distribution of summer T-a in this area was discussed. The satisfactory results suggest that this modelling approach is appropriate for estimating air temperature in mountainous regions with complex terrain.