| Precipitation is a key element of the global water cycle and an important indicator for global climate change research.The acquisition of high precision and high resolution precipitation data can provide an important role in water resources management,drought monitoring,agricultural and livestock production and other research.With the development of remote sensing technology,comprehensive and effective precipitation information can be obtained on a large scale,and GPM precipitation data with high spatial and temporal resolution and a large observation range have been widely used.Its spatial resolution is slightly coarse against the background of the demand for high-precision hydrological research in local regions and watersheds and the increasing spatial resolution of other remote sensing products.Therefore,it is important to improve its spatial resolution through downscaling models and to further refine the analysis of drought characteristics through downscaled precipitation data.This paper uses the Mongolian Plateau as the study area.Firstly,the accuracy of the GPM precipitation data in the study area was evaluated using the accuracy evaluation index with the meteorological station data as the true value.Secondly,using land cover type data,the study area was divided into vegetated and non-vegetated areas,and the lagged response of vegetation to precipitation in vegetated areas was analysed,as well as the correlation between precipitation and land surface temperature(LST)dataset values in vegetated and non-vegetated areas.Then elevation,slope,slope orientation data,normalised vegetation index(NDVI)and LST data were used as variables in the vegetated areas.In the non-vegetated areas,elevation,slope,slope orientation data and LST data were used as variables.Multiple linear regression models(MLR),geographically weighted regression models(GWR),random forest(RF)regression models and support vector machine(SVM)regression models were constructed for the vegetated and non-vegetated areas with GPM precipitation data respectively.Monthly 1 km resolution downscaled precipitation data were obtained from 2006 to 2010 for different downscaled models in the study area.The accuracy of the downscaling results of the models was again verified using meteorological station data,and the optimal precipitation downscaling model for the study area was compared.Finally,based on the optimal precipitation downscaling model for the study area,the monthly GPM precipitation data for the study area from 2001 to 2020 were spatially downscaled,and the downscaled precipitation data obtained were used to construct a standardized precipitation index(SPI)to finely analyze the drought characteristics of the study area with high spatial resolution.The following conclusions were drawn:(1)The accuracy assessment of GPM precipitation data shows that.On the monthly scale,GPM precipitation data have a high correlation with meteorological station data,with a correlation coefficient of 0.93.The correlation coefficient is also high on the monthly scale from March to November,ranging from 0.87 to 0.93.On the seasonal scale,GPM precipitation data have a high correlation with meteorological station data in spring,summer and autumn,with correlation coefficients of 0.91,0.90 and0.94 respectively,On the annual scale,the correlation between GPM precipitation data and meteorological station data was high,with a correlation coefficient of 0.92.All time scales showed good correlation between GPM precipitation data and meteorological station data,but all overestimated precipitation and had some errors.At the spatial scale,GPM data and meteorological station data have good spatial correlation at monthly,seasonal and annual scales.(2)Correlation analysis of the GPM precipitation data with the explanatory variables showed that there were differences in the correlation coefficients between precipitation in the vegetated areas of the study area and the NDVI for the current month,the next month,the next two months and the next three months.The highest correlation coefficients were found between precipitation in January,February,May and June and the NDVI of the next three months;the highest correlation coefficients were found between precipitation in March,July and September and the NDVI of the next two months;and the highest correlation coefficients were found between precipitation in April and August and the NDVI of the next month.In the vegetated areas,the highest correlations between precipitation and LST data sets were mainly for land surface temperature at night(LST＿N)and land surface temperature difference between day and night(LST＿D＿N)data.In non-vegetated areas,the data with the highest correlation between precipitation and LST data sets are mainly daytime land surface temperature(LST＿D)and LST＿D＿N data.(3)The downscaling results of GPM precipitation data show that.The results of the multiple linear regression model,the geographically weighted regression model,the random forest regression model and the support vector machine regression model all improve the spatial resolution of the GPM precipitation data in the study area from 0.1° to 1 km,with the highest accuracy of the random forest model at monthly,seasonal and annual scales.The details of precipitation are more refined.(4)The application of precipitation data based on the optimal downscaling model to drought characteristics shows that the study area is dominated by light and moderate droughts from 2001 to 2020.The droughts occur most frequently in the southwestern,central and northeastern parts of the plateau in the months of January to April and November to December.The central part of the plateau is the region with the highest frequency of annual cumulative droughts.January to April and November to December are the months in which the total drought area,light drought area,medium drought area,severe drought area,and special drought area occur more frequently in the study area.2001,2005,and 2007 are the years in which various types of drought occur more frequently. |
PDF Full Text Request