Drought Characteristics And Prediction Models In Eastern Inner Mongolia

Frequent occurrence of drought disasters impact economic development of agriculture and animal husbandry in China.The frequency of droughts in eastern Inner Mongolia,the largest base of mixed cereal and bean production in China,has had a huge impact on agriculture in the region.It is important to carry out the analysis of drought evolution characteristics and prediction research in the main grain producing areas.In this study,the meteorological data recorded at42 meteorological stations in eastern Inner Mongolia from 1960 to 2021 were used as the basis for drought analysis based on hierarchical clustering using Standardized Precipitation Evapotranspiration Index(SPEI)for interannual and seasonal drought frequency in eastern Inner Mongolia from 1960 to 2021.Then M-K trend analysis was used to analyze the trend and spatial distribution of the eastern part of Inner Mongolia in the last 62 years.In addition,drought prediction models based on deep learning techniques for LSTM and combined model CNN-LSTM were also developed,and LSTM performance advantages over the combined model CNN-LSTM model and different methods to improve its prediction performance are compared and analyzed.Finally,the time series prediction and spatial distribution prediction of SPEI at different time scales were performed using a long short-term memory network(LSTM)drought prediction model.The main findings are as follows:(1)Drought analysis of eastern Inner Mongolia based on drought index.The study area is divided into the southern and northern drought areas by the line of Alshan-Solun-Zalait Banner.The SPEI-based drought frequencies in eastern Inner Mongolia were relatively consistent in terms of spatial and temporal distribution characteristics.Therefore,SPEI is applicable to this study to evaluate the drought conditions in the eastern part of Inner Mongolia(2)Spatial and temporal evolutionary characteristics of drought in eastern Inner Mongolia.The drought changes in the last 62 years can be divided into three stages.Before the 1980 s,the drought was more consistent and less severe in the north and south;From the 1980 s to the 21 st century,drought conditions differed between the north and south,with consecutive droughts occurring in the south;after the 21 st century,drought conditions changed more consistently from north to south,with increased frequency and drought conditions in the region.In the past 62 years,there is a trend of drying in spring,summer and autumn in the south,with summer and autumn being the most significant.In the north,the drying trend is more consistent in spring,summer and autumn,and in winter,it develops towards wetness.The northern and northeastern regions of Hulunbuir are concentrated in areas with more summer droughts;The higher frequency of drought in spring was concentrated in the western part of Chifeng.The higher frequency of drought in summer is concentrated in the northern and northeastern part of Hulunbeier.In autumn,the higher frequency of drought occurs in the northeastern part of Tongliao.The areas with higher frequency of drought occurrence in winter than 37% are concentrated in the southwestern part of Hulunbeier City,the eastern part of Chifeng City and the southeastern part of Tongliao City.(3)Research on drought prediction based on deep learning methods in drought prediction.The accuracy of LSTM and CNN-LSTM models improves with increasing time scale,and the prediction performance of LSTM models is better than that of CNN-LSTM models in all aspects,and CNN models have excellent robustness and generalization ability.Different training parameters and network structure in the LSTM model can improve the prediction accuracy to a small extent,but the prediction performance is not necessarily improved when the network structure is more complex.In terms of spatial distribution prediction,the LSTM model predicted SPEI12 and SPEI24,and the predicted values were very similar to the spatial distribution of the true SPEI values.Therefore,LSTM can be an effective method to predict drought conditions in eastern Inner Mongolia as a single feature regional long-term and able to be more accurate on a large scale,which provides a scientific basis for the formulation of drought prevention,drought resistance and drought management countermeasures proposed to reduce the losses caused by drought.