Abstract River damming for electric power production generally triggers a set of anthropic activities that strongly impact on aquatic ecosystem, especially in small reservoirs located in urbanized and industrialized areas. Among the possible adverse effects is the over-abundance of aquatic macrophytes resulting from the input of high concentration of nutrients in the ecosystem that can affect the health of the ecosystem. In these situations, macrophytes are treated as weeds that need to be continuously monitored and analysed over time. Historically, remote sensing has played a prominent role in change detection studies and, nowadays, considering the open data sources of multi-temporal images and the high computational performance that allows for larger volumes of historical images to be mined, water monitoring is a recurrent object of analysis. The Salto Grande reservoir is a small water body located in the metropolitan region of Campinas, Sao Paulo, Brazil, characterized by high rates of urbanization and industrialization. The intense anthropic occupation around the reservoir triggered the degradation of the landscape and the decrease of water quality. This study explored the potential of image-attributes’ time series to monitor the spatio-temporal behavior of aquatic macrophytes in the Salto Grande Reservoir. Our assumption was that the combination of techniques for analyzing large multi-temporal datasets enables us to understand the trends and changes in the macrophytes occurrence in this small reservoir. To achieve this, quarterly Normalized Difference Vegetation Index (NDVI) time series based on Landsat data imagery from 1984 to 2017 were built to analyze the occurrence and persistence of these aquatic plants in the reservoir. A principal component analysis (PCA) was applied to the NDVI time series, which allowed us to identify typical years in the abundance of macrophytes and twelve regions of greater and lesser temporal variability in its abundance, by a K-means aggregation of the first principal component scores. For these regions, the Breaks for Additive and Seasonality Trend (BFAST) algorithm was used to analyze the trend, cyclic behaviour, and changes in the time series of the average NDVI. BFAST was able to detect gradual and abrupt changes for each of the twelve areas by searching for breakpoints in the temporal series. It was observed that the regions near the dam and where the conditions of the river are still maintained are most affected by the occurrence of macrophytes, characterized by an average NDVI greater than 0.4. Although subject to more subtle seasonal variations, all these regions defined at least one breakpoint, suggesting abrupt changes such as sharp interventions to control the overabundance of macrophytes at specific time. The regions located in the middle of the reservoir, with a more lacustrine influence, had lower average NDVI and small variations over time. Thus, it was possible to identify the critical regions of the studied reservoir with excess of growing macrophytes through the applied method, which also can be applied to similar areas.

中文翻译：

---

以 Salto Grande（巴西）为例的富营养化热带水库中大型植物丰度变化：利用 Landsat 遥感数据的趋势和时间分析

摘要 用于发电的河流筑坝通常会引发一系列对水生生态系统产生强烈影响的人类活动，特别是在位于城市化和工业化地区的小型水库。可能的不利影响之一是由于在生态系统中输入高浓度营养物质而导致水生大型植物过多，这会影响生态系统的健康。在这些情况下，大型植物被视为杂草，需要随着时间的推移进行持续监测和分析。从历史上看，遥感在变化检测研究中发挥了突出的作用，如今，考虑到多时相图像的开放数据源和允许挖掘大量历史图像的高计算性能，水监测是一个经常出现的对象的分析。萨尔托格兰德水库是位于巴西圣保罗坎皮纳斯大都市区的一个小型水体，其特点是城市化和工业化率高。水库周围强烈的人为占领引发了景观退化和水质下降。本研究探索了图像属性时间序列在监测 Salto Grande 水库中大型水生植物时空行为方面的潜力。我们的假设是，分析大型多时相数据集的技术组合使我们能够了解这个小型水库中大型植物发生的趋势和变化。为了达成这个，基于1984年至2017年的Landsat数据影像，建立季度归一化植被指数（NDVI）时间序列，分析水库中这些水生植物的发生和持续时间。将主成分分析 (PCA) 应用于 NDVI 时间序列，这使我们能够通过第一主成分的 K 均值聚合来确定大型植物丰度的典型年份和其丰度具有较大和较小时间变异性的十二个区域组件分数。对于这些区域，使用加性和季节性趋势中断 (BFAST) 算法来分析平均 NDVI 的时间序列的趋势、循环行为和变化。BFAST 能够通过搜索时间序列中的断点来检测十二个区域中每个区域的逐渐和突然变化。据观察，大坝附近和河流条件仍然保持的地区受大型植物发生的影响最大，其特征是平均 NDVI 大于 0.4。尽管受更微妙的季节性变化影响，但所有这些区域都至少定义了一个断点，这表明发生了突然的变化，例如在特定时间进行剧烈干预以控制大型植物的过多。位于水库中部的区域受湖泊影响较大，平均NDVI较低，随时间变化较小。因此，通过应用的方法可以识别出研究水库中大型植物生长过剩的关键区域，该方法也可以应用于类似区域。