王钧, 李广, 刘强. 基于LSTM神经网络模拟的陇中黄土高原沟壑区保护性耕作下土壤贮水量变化[J]. 中国生态农业学报(中英文), 2019, 27(8): 1226-1237. DOI: 10.13930/j.cnki.cjea.180949
引用本文: 王钧, 李广, 刘强. 基于LSTM神经网络模拟的陇中黄土高原沟壑区保护性耕作下土壤贮水量变化[J]. 中国生态农业学报(中英文), 2019, 27(8): 1226-1237. DOI: 10.13930/j.cnki.cjea.180949
WANG Jun, LI Guang, LIU Qiang. Soil water storage under conservation tillage based on LSTM neural network simulation in the Loess Plateau Gully Region of central Gansu[J]. Chinese Journal of Eco-Agriculture, 2019, 27(8): 1226-1237. DOI: 10.13930/j.cnki.cjea.180949
Citation: WANG Jun, LI Guang, LIU Qiang. Soil water storage under conservation tillage based on LSTM neural network simulation in the Loess Plateau Gully Region of central Gansu[J]. Chinese Journal of Eco-Agriculture, 2019, 27(8): 1226-1237. DOI: 10.13930/j.cnki.cjea.180949

基于LSTM神经网络模拟的陇中黄土高原沟壑区保护性耕作下土壤贮水量变化

Soil water storage under conservation tillage based on LSTM neural network simulation in the Loess Plateau Gully Region of central Gansu

  • 摘要: 为分析陇中黄土高原沟壑区不同保护性耕作措施的贮水效果,本研究利用春小麦/豌豆(W/P)、豌豆/春小麦(P/W)轮作的长期定位试验,分别设置传统耕作(T)、免耕(NT)、传统耕作秸秆覆盖(TS)和免耕覆盖(NTS)4种耕作措施,以当地月平均气温、月降水量、月平均辐射量、月平均蒸发量、月作物耗水量为输入因子,以月土壤贮水量为输出,建立基于长短期记忆(Long Short-Term Memory,LSTM)神经网络的土壤贮水量预测模型,并对该模型的有效性进行评估。研究结果表明:1)基于LSTM神经网络建立的土壤贮水量模型对陇中黄土高原沟壑区保护性耕作下土壤贮水量预测具有较好的适用性,模型模拟结果的平均均方根误差为7.76 mm,平均绝对误差为6.95 mm,相对误差控制在-5%~+5%的范围内。2)P/W轮作序列中各处理的土壤贮水量均比W/P轮作序列增加1.09%~1.43%。3)不同轮作序列,NTS处理的贮水效果均优于其他3种耕作措施,在W/P轮作序列中,NTS处理的年均土壤贮水量比T、NT和TS分别增加2.89%、1.70%和2.46%;在P/W轮作序列中,NTS处理的年均土壤贮水量比T、NT和TS分别增加3.03%、1.91%和2.57%。4)不同降水年型,NTS处理的土壤贮水量最高,且干旱年效果更加显著,其中丰水年NTS处理的土壤贮水量比T、NT和TS平均增加2.71%、1.48%和2.19%,而干旱年平均增加3.97%、2.54%和3.64%。5)保护性耕作措施的贮水效果随季节发生变化,作物生长前期(3-5月)保护性耕作措施的贮水优势较为明显,进入作物生长旺盛期(5-6月)保护性耕作措施与传统耕作的贮水效果差异不显著,而作物生长后期(7月)保护性耕作措施较传统耕作土壤贮水量明显增加。基于LSTM神经网络模拟环境下免耕覆盖的贮水保墒效果最好,为陇中黄土高原沟壑区最适宜的保护性耕作措施。

     

    Abstract: In order to analyze the effects of soil water storage for four different tillage measurestraditional tillage (T), no-tillage with no straw cover (NT), traditional tillage with straw incorporation (TS) and no-tillage with straw cover (NTS), the field research was conducted in the Loess Plateau Gully Region of central Gansu. The objectives of the study were to establish a prediction model for soil water storage based on Long Short-Term Memory (LSTM), and to evaluate the model's effectiveness using the long-term positioning experiment of the rotation sequence for spring wheat/pea (W/P) and pea/spring wheat (P/W) crops. In the experiment, monthly average temperature, monthly precipitation, monthly average radiation, monthly average evaporation and monthly crop water consumption constituted input factors, and value for soil water storage constituted the output factor in the prediction model. The results of the present study showed that:1) The water storage model based on LSTM neural network showed good applicability for predicting soil water storage, particularly in conservation tillage practice, in the Loess Plateau Gully Region of central Gansu. The average root mean square error and mean absolute error of the model simulation were 7.76 mm and 6.95 mm, respectively; moreover, the relative error was controlled between -5% and +5%. 2) In P/W rotation sequence, the soil water storage of various treatments increased by 1.09%-1.43% as compared to W/P. 3) Within different rotation sequences, water storage effect of NTS treatment turned out to be better than those for other three tillage measures. In W/P rotation sequence, the annual average soil water storage of NTS treatment was 2.89%, 1.70% and 2.46% higher than that of T, NT, and TS, respectively. In P/W rotation sequence, the average annual soil water storage of NTS treatment increased by 3.03%, 1.91% and 2.57%, respectively, compared to that of T, NT, and TS. 4) In the years with different precipitation, the soil water storage of NTS treatment was the highest, and this effect was markedly more significant in the dry year. The soil water storage of NTS treatment increased by 2.71%, 1.48% and 2.19%, on average, in the wet year, and it increased by 3.97%, 2.54% and 3.64%, on average, in the dry year as compared to the values recorded for T, NT and TS. 5) The water storage effect for conservation tillage measures varied with the season. There was obvious water storage advantage of conservation tillage measures during the early stages of crop growth (March-May). However, there was not significant between the results of conservation tillage measures and traditional tillage during the full growth stage (May-June). Nevertheless, the soil water storages for conservation tillage increased significantly during the late stage of crop growth (July). Going by the effect on soil water storage based on LSTM neural network, no-tillage mulching remains the best practice as well as the most suitable measure of protective tillage in the Loess Plateau Gully Region of central Gansu.

     

/

返回文章
返回