High spatiotemporal resolution precipitation grid data serve as crucial input for various hydrological and cli-mate models.This study introduced longitude,latitude,temperature,wind speed,and specific humidity as explanatory variables.The multiple linear regression,long short term memory neural network,and random forest models were used to downscale GPM-IMERG precipitation data.Using measured data from the Yongxing Island meteorological station as ground truth,accuracy verification was conducted on the original GPM-IMERG precipitation data and downscaled data u-sing metrics such as Bias,correlation coefficient,root mean square error,and mean absolute error.The results indicate that at the daily scale,the deviation between GPM-IMERG precipitation data and measured precipitation data at the mete-orological station is-0.004 mm,with a mean absolute error of 3.537 mm.The 10 m spatial resolution precipitation data in the Yongxing Island is obtained through GPM-IMERG downscaling.The RF model yields the highest accuracy in downscaled results,with a 46.9%improvement in root mean square error and a 30.1%increase in the correlation coeffi-cient compared to the original GPM-IMERG precipitation data.This study provides a new approach for obtaining high spatial resolution precipitation data for islands and reefs in situations with sparse meteorological stations.
关键词
GPM-IMERG/降尺度/机器学习/南海岛礁
Key words
GPM-IMERG/downscaling/machine learning/islands and reefs in south china sea
维普
万方数据