Research On Global Terrestrial Meteorological Drought Monitoring And Prediction Based On Artificial Intelligence

Meteorological drought is one of the most common types of drought globally.Compared to other types of drought,meteorological droughts occur suddenly and have a more significant impact on agriculture,water resources,ecosystems,and human society.Therefore,strengthening research on meteorological drought monitoring,prediction,and assessment is crucial to mitigate the negative impacts of droughts on agriculture,water resources,and human society.Traditional methods of meteorological drought monitoring and prediction mainly rely on statistical models and meteorological knowledge.With the rapid development of artificial intelligence technology,its application in the field of meteorological drought monitoring and prediction has also received widespread attention.Artificial intelligence methods,such as deep learning,can learn patterns from a large amount of meteorological data to obtain automatically extracted features,resulting in more accurate monitoring and prediction results.In this study,four widely used drought indices,SPI,SPEI,PDSI,and SCPDSI,were calculated using three param eters,cumulative precipitation,potential evapotranspiration,and surface temperature,from the ERA5 dataset.Artificial intelligence technology was used to analyze and explore global land drought monitoring and prediction from 1959 to 2021.providing a reference for global drought monitoring and prediction research.The main research contents and conclusions of this paper are as follows:(1)Global land latitude and longitude grid drought indices were calculated using ERA5 data,including 1-month-scale SPI-1,SPEI-1.PDSI.and SCPDSI and 3.6,9,and 12-month-scale SPEI-3,SPEI-6,SPEI9.and SPEI-12.The results showed that the four drought indices at the monthly scale can monitor droughts on six continents(excluding Antarctica).Due to the intensification of global climate change,the impact of temperature on droughts is becoming increasingly important.The SPI monitoring ability that only considers precipitation is weak,and the monitoring range of drought is relatively small.SPEI,PDSI,and SCPDSI both consider precipitation and evapotranspiration,so they can effectively monitor drought conditions in various continents.(2)In terms of the time characteristics of drought indices,there are significant differences in the shortterm drought monitoring performance of SPI-1 and SPEI-1,with frequent fluctuations over time,while PDSI and SCPDSI show good consistency in the same region.PDSI,SCPDSI,and SPEI-12 at the 12month scale are suitable for long-term drought monitoring because they exhibit a clear upward or downward trend over time.Moreover,PDSI and SCPDSI have a wider monitoring range than SPI and SPEI and can monitor extreme drought conditions.The spatial distribution of the four drought indices shows that they can all monitor extreme and severe droughts in ten selected periods,with SPI monitoring heavier droughts than other indices.The drought frequency was calculated using the drought indices for 756 months,and the total drought frequency on six continents was in the order of Africa,Asia,Oceania,South America,Europe,and North America,with Africa and Asia experiencing the most frequent droughts.Africa had a total drought frequency of over 50%in 756 months.(3)In order to study the prediction of drought index time series,five models,including ARIMA,SVR,LSTM,GRU,and TCN,were established to predict four drought indices.The results showed that the prediction accuracy of PDSI and SCPDSI by all five models was much higher than that of SPI-1 and SPEI-1,with SCPDSI having the highest prediction accuracy.The short-term SPI-1 and SPEI-1 sequences were not suitable for the five prediction models because they oscillate frequently over time.The deep learning models(LSTM,GRU,and TCN)exhibited higher prediction accuracy than traditional machine learning models(ARIMA and SVR).As the time scale of SPEI increased,its prediction ability was improved in three deep learning models,and SPEI-12 had the highest prediction accuracy in the TCN model,with an RMSE of 0.362.Among the five prediction models,TCN had the most accurate overall prediction results,and among the four drought indices,SCPDSI and SPEI-12 had the highest prediction accuracy,with SPEI-12 having an MAE,MSE,and RMSE of 0.271,0.131,and 0.362,respectively,in the TCN model.In summary,SPEI,PDSI,and SCPDSI have higher global land monitoring capabilities than SPI.PDSI,SCPDSI,and SPEI-12 are suitable for long-term meteorological drought monitoring and prediction,with SCPDSI and SPEI-12 having the highest prediction accuracy among the four drought indices,the TCN model and GRU model being the most accurate in predicting drought index time series.These research results provide valuable insights for drought monitoring and prediction,which can help to develop effective drought management strategies.