Drought Monitoring And Forecasting Based On AMSR-E Microwave Remote Sensing And TIGGE Data

Drought is one of the most serious natural disasters that affect China’s agricultural development, and has brought huge losses to the national economy. Soil moisture is an important indicator for drought. It is very important for drought research to monitor and forecast soil moisture. And this paper is about drought monitoring and forecasting.Firstly, in the surface drought monitoring section, pixel-based and plane-based artificial neural network models with a back-propagation learning algorithm (BPNN) have been developed using AMSR-E observations to retrieve soil moisture for the Huaihe River Basin. By comparing the simulation soil moisture with AMSR-E soil moisture products, such as standard algorithm by NASA and Land Parameter Retrieval Model (LPRM), we can conclude that:Pixel-based BPNN models have obvious advantages over AMSR-E soil moisture products in the study area. Soil moisture retrieved by pixel-based BPNN models has a better correlation with in situ soil moisture. The RMSE and MAE of the modelled soil moisture are the smallest of all the soil moisture products. For a pixel scale of 25 km, soil moisture by plane-based BPNN models are very different from in situ soil moisture. While for a watershed-scale simulation, plane-based BPNN models can better capture soil moisture spatial distribution. Generally, soil moisture retrieved by pixel-based and plane-based BPNN models can get a better performance on soil moisture spatial distribution and temporal dynamic change than AMSR-E products, and then can be used to monitor drought effectively.Secondly, in the drought forecasting section, according to H.L. Penman’s formula we have calculated evaporation using TIGGE dataset and validated the evaporation. And multimodel consensus forecasts, including multiple forecasts from a single model and forecasts from different models, are investigated by using outcomes of 24h-240h ensemble forecasts for 24h accumulated total precipitation taken from ECMWF, NCEP, UKMO and CMA global ensemble prediction systems (EPS). The multimodel consensus technique used in this study is super-ensemble. Then the TIGGE forecast datasets are used to drive XXT hydrological model to simulate soil water deficit depth and runoff. Finally, in order to unify the drought indexes of the two sections, the paper explores a unification approach.The results show that it is reasonable and feasible to calculate soil water deficit depth and runoff based on XXT distributed (rainfall-runoff) hydrological model driven by Penman evaporation and 24h accumulative total precipitation forecasting data of the EPSs.The results of the four ensemble forecasting models, as well as the members of the same single-model, vary widely. CMA has a maximum deviation of dispersion and poor performance on time-effectiveness and forecasting value. NCEP and UKMO have a great advantage on precipitation forecasting for a value of 0 or close to 0. ECMWF and NCEP’s RMSEs of 24h-144h 24h accumulative total precipitation forecasts are smaller than other models’.In addition, among the single-model ensembles, CMA has a worst performance on 24h accumulative total precipitation forecasts. ECMWF and UKMO forecasts have higher skill than CMA and NCEP in terms of temporal correlation coefficient. Generally, the super-ensemble of the individual member forecasts often produces a more accurate forecast than any single member forecast because merits of the individual forecasts tend to be integrated.The multimodel consensus forecasts of the 24h accumulated total precipitation are more accurate than the predictions from the individual ensemble members, especially for a value of 0 or close to 0. And the discrete degree of the multimodel consensus forecasts is lower. But not all multimodel consensus forecasts are better than single-model ensemble forecasts. Adding models with poor predictions to a multimodel consensus prediction system may significantly reduce the accuracy of the multimodel consensus forecasts. On average, the multimodel consensus prediction systems of 24h accumulated total precipitation provide superior forecasts than most single-model ensembles.As the driven data of XXT hydrological model, the accuracy of the TIGGE datasets directly affects the output of the model. The simulating soil water deficit depth and runfall driven by ENU forecasts are more accurate than other super-ensembles forecasts in terms of temporal correlation coefficient with values of 0.89 and 0.93. After regression analysis, the ENU forecasts performs best on driving XXT model to simulate soil water deficit depth among the multimodel consensus prediction systems.Finally, unify drought indexes needs further study. Through regression analysis of soil relative moisture content and soil water deficit depth, the deterministic coefficient and correlation coefficient are 0.60 and 0.77 respectively. The common methods of linear regression analysis can only assess drought level in a unified way to some extent.