基于LSTM和GRU模型的森林物候预测研究  

Research on Forest Phenology Prediction based on LSTM and GRU Model

作　　者：关鹏 郑一力[1,2,3] GUAN Peng;ZHENG Yili(School of Technology,Beijing Forestry University,Beijing 100083,China;Beijing Laboratory of Urban and Rural Ecological Environment,Beijing Municipal Education Commission,Beijing 100083,China;Key Lab of State Forestry Administration for Forestry Equipment and Automation,Beijing 100083,China)

机构地区：[1]北京林业大学工学院,北京100083 [2]北京城乡生态环境实验室,北京100083 [3]国家林业局林业装备与自动化重点实验室,北京100083

出　　处：《Journal of Resources and Ecology》2023年第1期25-34,共10页资源与生态学报（英文版）

基　　金：The Fundamental Research Funds for the Central Universities (2021ZY74)。

摘　　要：森林物候是研究气候与环境变化之间关联的重要参量。本研究将光学相机作为近地遥感卫星设备获取森林图像,计算出绝对绿度指数(GEI)数据,通过双逻辑斯蒂模型和归一化处理方法,拟合GEI数据物候季节变化曲线;引入LSTM和GRU深度学习模型对物候数据进行训练和测试分析,并验证深度学习模型合理性和性能评估,最后预测GEI数据未来60天变化趋势。结果表明:在森林物候训练和预测方面,GRU与LSTM模型经直方图和自相关图验证,显示预测数据分布与真实数据趋势吻合,LSTM与GRU模型呈现的结果数据具有可行性且模型具有平稳性;LSTM模型与GRU模型的MSE、RMSE、MAE和MAPE差值分别为0.0014、0.013、0.008和5.26%,GRU模型的性能优于LSTM;LSTM与GRU模型对GEI数据未来60天预测,均表现出与上半年GEI数据变化趋势相符合的走势图;通过GRU和LSTM对GEI数据进行深度学习模型预测,实现LSTM和GRU深度学习模型在森林物候预测中的响应,同时验证了GRU性能优于LSTM模型,从而进一步揭示未来森林物候的生长和气候变化,为森林物候预测的应用提供理论依据。Research on forest phenology is an important parameter related to climate and environmental changes.An optical camera was used as a near-earth remote sensing satellite device to obtain forest images, and the data of Green excess index(GEI) in the images were calculated, which was fitted with the seasonal variation curve of GEI data by double Logistic method and normalization method. LSTM and GRU deep learning models were introduced to train and test the GEI data. Moreover, the rationality and performance evaluation of the deep learning model were verified, and finally the model predicted the trend of GEI data in the next 60 days. Results showed: In the aspects of forest phenology training and prediction, GRU and LSTM models were verified by histograms and autocorrelation graphs, indicating that the distribution of predicted data was consistent with the trend of real data,LSTM and GRU model data were feasible and the model was stable. The differences of MSE, RMSE, MAE and MAPE between LSTM model and GRU model were 0.0014, 0.013, 0.008 and 5.26%, respectively. GRU had higher performance than LSTM. The prediction of LSTM and GRU models about GEI data for the next 60 days both showed a trend chart consistent with the change trend of GEI data in the first half of the year. GRU and LSTM were used to predict GEI data by deep learning model, and the response of LSTM and GRU deep learning models in forest phenology prediction was realized, and the performance of GRU was better than that of LSTM model. It could further reveal the growth and climate change of forest phenology in the future, and provide a theoretical basis for the application of forest phenology prediction.

关 键 词：森林物候 绝对绿度指数 LSTM模型 GRU模型 预测 

分 类 号：X703[环境科学与工程—环境工程]