基于BO-CNN-LSTM的锡林郭勒草原干旱预测模型

Drought prediction model for the Xilingol grasslandbased on BO-CNN-LSTM

摘要:

【目的】 建立基于BO-CNN-LSTM 耦合神经网络的干旱预测模型，探索干旱预测的适用性。【方法】 首先，基于长短期记忆网络（LSTM）的记忆功能，将其嵌入卷积神经网络（CNN）全连接层。其次，为确定最优超参数，将贝叶斯优化算法（BO）的概率代理模型和采集函数引入至LSTM。最后，建立BO-CNN-LSTM 耦合神经网络模型用以预测干旱状况。【结果】 1） BO-CNN-LSTM 预测精度随时间尺度的增大而提高，对SPEI-12 模拟精度最高，且判定系数R2 均在98% 以上；2） 与LSTM模型SPEI-12 的模拟结果进行比较，BO-CNN-LSTM 表现出更高拟合精度。其中R2 相对提高值为[4. 63%，8. 67%]，MSE 的数量级由10-2 降至10-3；3） 通过BO-CNN-LSTM 预测2023 年锡林郭勒草原干旱空间分布，结果显示该区域整体呈干旱态势。其中东乌珠穆沁旗站点区域属于中旱，其它区域均属于重旱。【结论】 BO-CNN-LSTM 具有较高的计算精度，尤其适用于预测SPEI-12，故可将其应用于年时间尺度干旱预测。

Abstract:

【Objective】 Reliable and effective monitoring can mitigate the impact of drought disasters on socio-74economic development and natural ecosystems. This study adopted the BO-CNN-LSTM coupled neural networkas adrought prediction model.【Method】 First， the memory function of long short-term memory（ LSTM） was integratedinto the fully connected layer of the convolutional neural network （CNN）. Second，to determine the optimalhyperparameters for LSTM，the probability surrogate model and acquisition function from the Bayesian optimization（BO） algorithm wereintroduced. Finally，a BO-CNN-LSTM coupled neural network model was constructed topredict the drought conditions in the Xilingol grassland.【Result】 （1） The prediction accuracy of the BO-CNNLSTMmodel increased with the time scale， withthe highest prediction accuracy observed under the 12-month scalefor the Standardized Precipitation-Evapotranspiration Index （SPEI）. The determination coefficient R² of SPEI-12for each site exceeded 98%.（2） Compared to the simulation results of the LSTM model for SPEI-12，the proposedmodel exhibited higher fitting accuracy， showinga relative improvement in R² of [4. 63%，8. 67%]. The order ofmagnitude of mean squared error（ MSE） at each site had decreased from 10-2 to 10-3.（3） Using the model to predictthe spatial distribution of drought in the Xilingol grassland for 2023. indicated that the region as a whole was experiencingdrought. Especially，the Dongwuzhumuqin Banner area was classified as experiencing moderate drought，whileother areas were classified as severe drought.【Conclusion】 The results demonstrate that the BO-CNN-LSTMmodel has high computational accuracy，making it particularly suitable for predicting SPEI-12. Therefore， the methodsin this study can be effectively applied to drought prediction on an annual time scale.