利用遥感技术快速、准确地监测洪水淹没范围对于洪水灾害管理工作具有重要的意义。永久性淹没区域和季节性淹没区域之外的洪水淹没范围往往对人类生命财产安全的危害性较大。而季节性淹没区域的提取需要分析地表水体长期复杂的变化特征。目前常用于季节性淹没区域识别的水体淹没频率(Water Inundation Frequency, WIF)阈值法，由于缺乏对水体范围动态变化规律的分析，提取精度较低、鲁棒性较差，为洪水淹没范围的准确提取带来了较大的不确定性。

本研究以鄱阳湖为研究区域，首先提出了基于贝叶斯模型平均法(Bayesian Model Averaging, BMA)综合历史、临近和邻域先验信息的水体范围提取方法(Water extraction based on BMA, WBMA)。通过对长时序水体变化的谐波分析，提出了基于谐波模型的水体淹没区域识别方法(Water inundation area Identification based on Harmonic Model, WIHM)。在WIHM获取高精度季节性淹没区域和永久性淹没区域的基础上，精确剔除了季节性水体对洪水淹没范围的不确定性影响，提高了洪水淹没范围的提取精度。最后基于长时序遥感数据提取了近30年鄱阳湖区域的洪水淹没范围，以洪水超越概率(Flood Exceeding Probability, FEP)为指标，分析近30年鄱阳湖区域的洪水灾害风险时空变化特征。主要成果如下：

(1)提出了一套基于贝叶斯模型平均的水体范围提取方法。该方法通过引入临近先验信息和BMA法，具备捕捉水体范围临近变化、综合多源先验信息的优势，能够更加精细地提取地表水体范围，总体精度可以达到97.24%，Kappa系数可以达到95.43%。

相较于历史先验水体概率，临近先验水体概率的可靠性更强，引入临近先验信息能够将提取的水体范围错误率降低1.66%。基于BMA综合历史、临近和邻域先验信息得到的水体概率可靠性优于基于单独采用某一类先验信息得到的水体概率。WBMA提取的地表水体范围准确性更高、噪声引起的水面内部孔洞以及孤立细小水体误差更少，总体精度可以达到97.24%，Kappa系数可以达到95.43%。WBMA具有能够捕捉水体范围临近时期的动态变化和有效综合多源先验信息的优势，适用于水文变化活跃地区和洪水期的地表水体范围精细提取工作。

(2)提出了一套基于谐波模型的水体淹没区域识别方法。该方法通过对长时序水体变化的谐波分析，具备量化水体变化规律的优势，能够更加精确地识别季节性淹没区域，剔除季节性水体对洪水淹没范围的不确定性影响，提高了洪水淹没范围提取精度，总体精度可以达到95%，Kappa系数可以达到86.11%。

WIHM中基于长时序水体范围计算的谐波模型振幅和均值参数能够有效提取季节性淹没区域。利用WIHM法识别季节性淹没区域和永久性淹没区域具有更高的精度和鲁棒性，总体精度可以达到94.75%，Kappa系数可以达到90.57%。基于WIHM提取的高精度季节性淹没区域和永久性淹没区域构成的参考水面，提取洪水淹没范围具有更高的精度，总体精度可以达到95%，Kappa系数可以达到86.11%，适用于水文变化活跃、复杂地区的洪水灾害应急工作，能够为自然灾害管理等相关部门提供重要的技术支撑。

(3)揭示了近30年鄱阳湖区域的洪水灾害风险时空变化趋势。约有9.36%的区域洪水灾害风险变小，主要分布在鄱阳湖体的周围和河流两侧；约有3.80%的区域洪水灾害风险变大，主要分布在人口密集的建成区域。

鄱阳湖区域接近10%的区域处于中等及以上的洪水灾害风险中，较高等级的洪水灾害风险主要分布在鄱阳湖体西侧、西南侧、南侧和东南侧以及河流的边缘和城市建成区。近30年鄱阳湖区域的洪水灾害风险整体上呈现变小的趋势，尤其是在2000年之后。约有9.36%的区域洪水灾害风险变小，主要分布在鄱阳湖体的周围、河流两侧，与防灾减灾能力的提高有关。约有3.80%的区域洪水灾害风险变大，主要分布在人口、建成区较为密集的区域，与建成区扩张和人口增多有关。直接利用对地观测数据获取历史时期的洪水淹没频率，能够更加直观、准确地反映未来洪水发生的可能性。同时，对近30年鄱阳湖区域的洪水灾害风险空间分布、变化趋势和影响因素进行了系统地分析，对于完善洪水灾害理论体系、理解洪水灾害风险的变化机制具有重要的科学意义，也对决策者开展洪水灾害预警工作、资源调度、提高防灾减灾能力具有重要的实践意义。

The utilization of remote sensing technology for rapid and precise flood inundation monitoring holds significant importance in flood disaster management. Flood inundation occurring beyond permanently inundated and seasonally inundated areas often poses greater risks to human life and property. Currently, the Water Inundation Frequency (WIF) threshold method, commonly employed for identifying seasonal inundation areas, suffers from low extraction accuracy and limited robustness due to insufficient analysis of the dynamic change patterns in water extent. This limitation introduces considerable uncertainty in accurately delineating the flood inundation.

Focusing on Poyang Lake as the study area, a set of Water extraction method based on Bayesian Model Averaging (WBMA) is proposed, which integrates historical, proximity, and neighborhood prior information using the BMA method. Following that, the study introduces the Water inundation area Identification based on Harmonic Model (WIHM), through the harmonic analysis of long time series water changes. Subsequently, building upon the precise delineation of seasonal and permanent inundation areas achieved by WIHM, the flood inundation extraction has been improved. Finally, the flood inundation area of the Poyang Lake region over the past 30 years is extracted. Flood Exceeding Probability (FEP) served as an indicator to analyze the spatial and temporal characteristics of flood disaster risk in the area during the same period. The main achievements are outlined below:

(1) A set of water extent extraction methods based on Bayesian model averaging is proposed. By introducing proximate a priori information and utilizing the BMA method, the method captures the proximate changes of water extent and integrates multi-source a priori information. This allows for finer extraction of surface water extents, with an overall accuracy reaching 97.24% and a Kappa coefficient of 95.43%.

Compared to historical a priori water probability, proximate a priori water probability is more reliable. Introducing proximate a priori information reduces the error rate of extracted water ranges by 1.66%. The reliability of water probability, based on combined historical, proximate, and neighborhood a priori information using BMA, surpasses that based on a single type of a priori information alone. WBMA achieves higher accuracy in extracting surface water ranges, reducing errors in noise-induced internal holes on the water surface and isolated fine water bodies. Overall accuracy reaches 97.24%, with a Kappa coefficient of 95.43%. WBMA excels in capturing dynamic changes in water extents and effectively integrates multiple a priori information. It is particularly suitable for finely extracting surface water extents in regions with active hydrological changes and during flooding periods.

(2) A suite of water inundation area identification methods, based on Harmonic Model, is introduced. The method quantifies the change rule of water bodies through harmonic analysis of their long time series. It can accurately identify seasonal inundation areas, eliminate the uncertainty caused by seasonal water variations on flood inundation ranges, and improve the accuracy of flood inundation range extraction. The overall accuracy can reach 95%, and the Kappa coefficient can reach 86.11%.

The amplitude and mean parameters of the harmonic model, calculated based on long time-series water body ranges in WIHM, effectively extract seasonal inundation areas. The WIHM method demonstrates higher accuracy and robustness in identifying seasonally inundated areas and permanently inundated areas, achieving an overall accuracy of 94.75% and a Kappa coefficient of 90.57%. Using a reference water surface composed of high-precision seasonally inundated and permanently inundated areas extracted by WIHM, flood inundation range extraction shows higher accuracy at 95% overall accuracy and an 86.11% Kappa coefficient. This method is applicable for emergency response to flooding in areas with active hydrological changes and complexity, providing crucial technical support for natural disaster management and related departments.

(3) Spatial and temporal trends of flood risk in the Poyang Lake region in the last 30 years is revealed. Roughly 9.36% of the area experiencing decreased flood risk is mainly concentrated around Poyang Lake and riverbanks. Conversely, approximately 3.80% of the area witnessing increased flood risk is primarily located in densely populated built-up areas.

Approximately 10% of the Poyang Lake area faces medium or higher-grade flood risk, predominantly clustered in the western, southwestern, southern, and southeastern regions, including the lake's periphery, riverbanks, and urban centers. Over the past three decades, there has been an observable decrease in flood risk, particularly pronounced post-2000. Approximately 9.36% of the region has experienced a reduction in flood risk, primarily concentrated around Poyang Lake and its adjacent riverbanks. This decrease can be attributed mainly to enhancements in disaster prevention and mitigation capacities. Conversely, about 3.80% of the area has seen an increase in flood risk, primarily within densely populated urban areas. This rise is largely due to urban expansion and population growth. The utilization of earth observation data for obtaining flood inundation frequency in historical periods offers a more intuitive and accurate depiction of future flood occurrences. Additionally, the systematic analysis of spatial distribution, changing trends, and influencing factors of flood risk over the past 30 years in the Poyang Lake region provides valuable insights into understanding and managing frequent flood disasters. Absolutely, the analysis not only contributes to advancing the theoretical understanding of flood disasters but also provides crucial practical implications for decision-makers and stakeholders involved in disaster management.