西南近50年实际蒸散发反演及其时空演变

利用CRU 4.0及GLDAS Noah 2.1数据集,采用随机森林算法对1966年至2016年中国西南陆面月尺度实际蒸散发(ETa)进行逐像元反演,结合袋外误差均方值(MSEOOB)、解释方差百分数(PVE)和均方根误差(RMSE)评价模型以及与其他典型数据集对比的方法对模型和反演结果进行精度评价,在对中国西南ETa的空间格局及时空演变特征进行分析的基础上,利用因子置换重要性评价模型(PIM)对特征因子进行重要性评价。结果表明:(1) MSEOOB均值为4.14,标准偏差仅为3.73,PVE均值为99.36%,标准偏差仅为0.33,模型基于2000年至2016年月尺度拟合结果的RMSE均值仅为1.04 mm/月,标准偏差为0.52,反演结果与GLDAS 2.1、2.0及MOD16数据的R2分别为0.99、0.89、0.95,总体而言模型及拟合结果可信度和精度较高;(2)西南地区ETa整体上表现出随着纬度的降低而增加的特征,从西北高原地区向东南沿海区域逐步增加,不同季节上西南的ETa空间分布差异较为明显,从春季到夏季先呈现出由东南向西北逐步增加的态势,夏季到冬季则呈现出从西北向东南减弱的特征,在每年的7、8月份左右各区域的ETa达到最大值,在1、2月份左右为最低值,并呈现起伏的周期特征;(3)以横断山脉为分界,横断山脉以南的丰水区的ETa主要受云覆盖百分数、月均气温日较差与月均日最高温共同驱动,而横断山脉以北的少水区域主要受云覆盖百分数、月霜日频率与月均水汽压共同驱动,而无论是在丰水区还是少水区,云覆盖百分数都是所有因素中最主要的驱动因子。

Inversion and spatiotemporal evolution of actual evapotranspiration in southwest China for the past 50 years

In this study, we combined the CRU 4.0 and GLDAS Noah 2.1 datasets and used the random forest algorithm (RF) to calculate the monthly actual evapotranspiration (ETa) in southwestern China between 1966 and 2016. The accuracy of our model and the inversion results were evaluated by the MSE_OOB, PVE, and RMSE indices and a comparison with other typical datasets. Based on this, the spatiotemporal distribution and evolution pattern of ETa over multiple time scales were fully discussed. We extended our research by using the PIM model to evaluate the importance of the feature factors in each pixel. We obtained the following results:1) the mean value and the standard deviation (Std Dev) of MSE_OOB were 4.14 and 3.73, and those of PVE were 99.36% and 0.33; in addition, the mean RMSE of the monthly inversion results for 2000 to 2016 was 1.04 mm per month and the corresponding Std Dev was 0.52; moreover, the R² values for the inversion results of ETa from GLDAS 2.1, GLDAS 2, and MOD16 were 0.99, 0.89, and 0.95, respectively. All these evaluation indices illustrated the credibility and precision of the model and the inversion results were sufficiently high. 2) Our inversion results indicated that ETa increased with a decrease in latitude and gradually increased from the northwest plateau to the southeast coastal area; in addition, the spatial distribution patterns in southwestern China in different seasons were quite different:from spring to summer, high ETa expanded from the southeast to the northwest; in contrast, in winter, ETa diminished remarkably from northwest to southeast. The maximum value was reached around July every year and had the lowest value was reached around February, which showed periodic characteristics with fluctuations. 3) We found that the Hengduan Mountains were the boundary of the driving factors for ETa in arid and humid regions. The ETa in the humid regions south of the mountains was jointly driven by the cloud cover percentage (CCP), diurnal temperature range (DTR), and monthly average daily maximum temperature (TMX). In contrast, ETa in arid regions north of the mountains was mainly affected by CCP, frost day frequency (FDF), and vapor pressure (VAP). Notability, CCP is the most important driving factor weather in both humid and arid areas.