基于多层感知机重建GRACE与GRACE-FO之间数据缺失的黄河流域陆地水储量变化

Reconstructing gap data between GRACE and GRACE-FO based on multi-layer perceptron and analyzing terrestrial water storage changes in the Yellow River basin

GRACE重力卫星于2017年6月结束,GRACE-Follow On(GRACE-FO)任务于2018年5月启动,导致了约一年的数据缺失.为了研究长时间序列陆地水储量的变化情况,需要重建这两代重力卫星之间的数据.本文基于多层感知机的神经网络模型,通过将GLDAS、CPC土壤湿度、土壤温度、蒸散量以及降水量等数据作为输入参数,GRACE等效水高作为输出参数来训练神经网络,以黄河流域为研究区域,得到缺失等效水高.通过分析黄河流域等效水高时间序列,得出在2002—2020年间研究区域水储量整体上呈现以-0.51±0.03 cm·a^(-1)速率减少的趋势.GRACE反演的等效水高相对于降水、蒸散量以及土壤温度滞后时间达2~3个月.通过小波分析验证了GRACE与GLDAS、CPC计算的等效水高具有较强的相关性,且均具有明显的周年共振周期.

The GRACE gravity satellite ended in June 2017,and the GRACE-Follow On mission was launched in May 2018,resulting in about a year of missing data.In order to study the changes in terrestrial water storage over a long time series,it is necessary to fill in the data between the two generations of gravity satellites.This paper is based on the multi-layer perceptron(MLP),which is trained by soil moisture,soil temperature,evapotranspiration and precipitation of GLDAS and CPC as input parameters,and GRACE-derived equivalent water height(EWH)as output parameters to train the neural network.Taking the Yellow River basin as the study area,the missing EWH was obtained.By analyzing the time series of the EWH in the Yellow River basin,it is concluded that the terrestrial water storage in the study area from 2002 to 2020 shows a trend of decreasing at a rate of-0.51±0.03 cm·a^(-1).GRACE-derived EWH has a lag time of 2 to 3 months relative to precipitation,evapotranspiration and soil temperature.Wavelet analysis verifies that the EWHs calculated by GRACE,GLDAS and CPC have a strong correlation,and have obvious annual resonance periods.