Abstract The cyanobacterial harmful algal blooms have been observed, over the last decade, in several regions of the southern Caspian Sea, becoming a major threat to human health and aquatic life. The present study aims to develop two models to quantify cyanobacterial concentration in the Caspian Sea using artificial neural networks and multiple band linear regression. The models are based on Moderate Resolution Imaging Spectroradiometer (MODIS) satellite imagery. Data were collected from the west, center, and east of the southern Caspian Sea between September 2015 and August 2016. The field dataset includes 123 samples in seven different transects and is used to define and evaluate the proposed methods. The Root Mean Square Error (RMSE), unbiased RMSE (URMSE), and correlation coefficient (R) values between Multiple Band Linear Regression Algorithm outputs and field dataset are 1.8 × 10−3 mg.m−3, 22.43%, and 0.73, respectively. For Artificial Neural Network (ANN), the outputs are 1.6 × 10−3 mg.m−3, 18.89%, and 0.81, respectively. The performance of the proposed methods is proven suitable under nearly all conditions of the southern Caspian Sea. However, numerical comparison and visual evaluation of the results show that the ANN method is less sensitive to small changes in the environmental conditions, leading to more stable results. Moreover, the ANN model provides accurate results in most cases, and the accuracy of this results are improved by increasing the training data. This study focused on the development and validation of an optimal algorithm for quantifying temporal and spatial variability phycocyanin concentrations in the Caspian Sea using daily satellite data.

中文翻译：

---

使用 MODIS 图像对南里海蓝藻浓度进行定量建模

摘要 在过去十年中，在里海南部的几个地区观察到蓝藻有害藻华，成为对人类健康和水生生物的主要威胁。本研究旨在开发两种模型，使用人工神经网络和多波段线性回归来量化里海中的蓝藻浓度。这些模型基于中分辨率成像光谱仪 (MODIS) 卫星图像。数据是在 2015 年 9 月至 2016 年 8 月期间从里海南部的西部、中部和东部收集的。现场数据集包括七个不同断面的 123 个样本，用于定义和评估所提出的方法。均方根误差 (RMSE)、无偏均方根误差 (URMSE)、多波段线性回归算法输出与现场数据集之间的相关系数 (R) 值分别为 1.8 × 10−3 mg.m−3、22.43% 和 0.73。对于人工神经网络 (ANN)，输出分别为 1.6 × 10−3 mg.m−3、18.89% 和 0.81。所提出的方法的性能被证明适用于里海南部几乎所有条件。然而，结果的数值比较和视觉评估表明，人工神经网络方法对环境条件的微小变化不太敏感，导致结果更稳定。此外，ANN 模型在大多数情况下都能提供准确的结果，并且通过增加训练数据来提高该结果的准确性。