土壤水分在水文、大气和生物系统中其起着非常重要的作用。在地球水文系统中，虽然其数量仅为整个系统的0.001%，但它直接影响到地表蒸发、地面热通量、植被生长和陆地系统的生物化学过程等。红壤在我国的分布比较广泛，约占整个国土面积的22.7%，但红壤地区由于其土壤结构性和水热资源年内分布不均匀等问题，使得该地区极易发生季节性干旱。目前，土壤水分原位实测方法有中子仪和TDR等方法。但获取土壤水分的长期变化实测数据要投入大量的人力物力，且土壤水分时空演变的复杂性使得监测难度加大。所以寻求合理有效地土壤水分动态变化的模拟方法，对研究区土壤水分及时准确的预测预报有着重要的实际及理论意义。因此,本研究基于上述原因，建立了土壤水分动态模拟及预测系统。 研究的主要内容及结果如下： （1）应用小波分析对气象因素的年变化规律进行研究发现：该区气象因素普遍存在27-29年的主周期，在主周期尺度上除了最低气温，其它气象因素存在“高值-低值-高值”的变化规律。同时，可以推断未来12-15年该区域除年最低气温有上升趋势外，其它气象因素都呈下降趋势，降水量的相对下降速度大于蒸发量。 （2）基于BP神经网络对土壤水分与气象因素进行敏感性分析探讨在非线性的条件下土壤水分与气象因素的相互关系。综合局部及全局敏感性分析表明：土壤水分含量对降水量的敏感度最大，蒸发量次之，对地温和相对湿度敏感度最小，在建模过程中可以不考虑或少考虑地温和相对湿度对土壤水分的影响。与此同时，推断了红壤区未来12-15年土壤含水量有可能有下降的趋势，所以研究该区土壤水分动态变化，并建立模拟及预测预报体系具有实际意义。 （3）在加入气象因素的土壤水分动态随机模型中，LS-SVM模型与ANFIS模型和BP-ANN模型相比，该模型的运行效率、稳定程度、模拟性能和建模数学意义更具有较强优越性和可靠性。应用LS-SVM对土壤水分动态日变化进行模拟，并采用bior3.3小波函数5层分解提取日变化趋势图的结果可以将研究区土壤水分动态变化分划为四个阶段：土壤水分充盈期、土壤水分亏缺期、土壤水分补充恢复期和土壤水分干旱期。该结果可为研究区水分合理利用和土壤水分的预测预报提供科学依据。 （4）混沌时间序列土壤水分预测模型中，原始信号、降噪信号和小波分解的趋势部分应用C-C算法确定的最优嵌入维数（m）及延滞时间(τ)将不会改变。M2模型应用启发式阈值算法降噪会使信号失真，采用经典时间序列预测方法预测结果不佳。M3模型应用自适应阈值算法降噪效果比较好，能去除原始信号的大部分白噪声。但M3模拟的效果略不及M5模拟的效果，这是因为M5不仅能精确地模拟预测出信号的趋势部分，而且能大体地模拟出部分信号的细节部分。通过M4和M5发现寻求合理的小波层次分解数，将有助于模型预测能力的提高。在模拟预测中当Lyapunov指数（λ1<下标!>）高时，预测能力会很快下降，合理的层次分解数可以提高预测的精度和预测范围。

Soil water plays a significant part in the hydrosphere, atmosphere and biosphere systems. In terms of the global hydrological cycle, the quantity of soil water is smaller than 0.001%. However, it is an important factor affecting evaporation, land surface energy fluxes, plant growth and biogeochemical cycling. The red soils which are the most important soil resource in southern China occupy approximately account for 22.7% of the total Chinese land. Presently, Soil water can be measured in-situ by Neutron Probes or Time Domain Refectometry (TDR). However, long-term direct measurements of soil water are expensive and impractical. Therefore, simulations and predictions based on limited measured data must be used to develop understanding of soil water dynamics in this region. Results were as follows: (1) The wavelet transformation was applied to analysis the variation of meteorological factors within 51 years. It showed that periodic variations (27-29 years) of meteorological factors are localized in the time domain, and exhibited “high-low-high” variation regularity except the minimum temperature. The results suggested that meteorological factors will decrease except the low temperature, and the decreasing rate of precipitation will be more than the evaporation. (2) The sensitivity analysis basing on the BP-ANN was applied to study nonlinear relationship between the soil water and meteorological factors. It was showed that the sensitivity between the soil water and precipitation were most than others. The result suggested that the soil water will decrease with 12-15 years in this study region. (3)In this study, the various nonlinear Stochastic Model of soil water simulation systems and chaotic time series analysis methods of prediction systems had been set up. In the nonlinear Stochastic Model of soil water simulation systems, the daily soil water content simulated by Least squares support vector machine (LS-SVM) with the meteorological factors had more stabilities and advantages in soil water simulation performance over the Back Propagation Artificial Neural Network (BP-ANN) and Adaptive Neuro-Fuzzy Inference System (ANFIS). The soil water dynamic diurnal variation was simulated by LS-SVM model and extracted its trend by bior3.3 making five layers wavelet decomposing. The trend implied that the soil water dynamic change can be divided into four stages: filling period, deficit period, recovery period and drought period. The results can provide scientific data for the water utilization and the soil water prediction in the study region. (4) In chaotic time series analysis method of prediction systems, the various signal preprocessing methods including the appropriate de-noising methods and wavelet decomposition methods were applied to preprocess the original chaotic soil water signal. The results of the prediction systems showed that the appropriate de-noising methods and the tendency of wavelet transformation had less effect on the delay time (τ) and embedding dimension (m). The de-noising methods may ignore the detail information of the signal; however the appropriate wavelet transformation to get smaller Maximum Lyapunov Exponent (λ1<下标!>) of the chaotic soil water signal detail and tendency information can improve the predicting capacity.