Scholarlink

<p>Lake Water Surface Area (WSA) is vital in environmental preservation and future water resource planning and management. Accurately mapping, monitoring, and forecasting Lake WSA changes are critical to regulatory agencies. This study used MODIS satellite images to extract a monthly time series of WSA of two lakes located in Iran from 2001 to 2019. Following a series of image and time series preprocessing to obtain the preprocessed lake surface area time series (A<sub>L</sub> TS), the outcomes were modeled by the Long-Short-Term Memory (LSTM) deep learning (DL) method, the stochastic Seasonal Auto-Regressive Integrated Moving Average (SARIMA) method and hybridization of these two techniques to develop WSA forecasts. After separate standardization and normalization of A<sub>L</sub> TS and reevaluation of the preprocessed data, the SARIMA (1, 0, 0) (0, 1, 1)<sub>12</sub> model outperformed sole LSTM models with correlation index of (R) 0.819, mean absolute error (MAE) of 49.425 and mean absolute percentage error (MAPE) of 0.106. On the other hand, the hybridization (stochastic-DL) enhanced the other characteristics of the estimations, made the reproduction of the primal statistical properties of WSA data more precise and caused better mediation.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Lake surface area forecasting using integrated satellite-SARIMA-long-short-term memory model

  • 期刊论文

  • 发表日期：

    2026年3月17日

查看全文
  • Keyvan Soltani,
  • Arash Azari,
  • Mohammad Zeynoddin,
  • Afshin Amiri,
  • Isa Ebtehaj,
  • Taha B. M. J. Ouarda,
  • Bahram Gharabaghi,
  • Hossein Bonakdari

摘要

Lake Water Surface Area (WSA) is vital in environmental preservation and future water resource planning and management. Accurately mapping, monitoring, and forecasting Lake WSA changes are critical to regulatory agencies. This study used MODIS satellite images to extract a monthly time series of WSA of two lakes located in Iran from 2001 to 2019. Following a series of image and time series preprocessing to obtain the preprocessed lake surface area time series (AL TS), the outcomes were modeled by the Long-Short-Term Memory (LSTM) deep learning (DL) method, the stochastic Seasonal Auto-Regressive Integrated Moving Average (SARIMA) method and hybridization of these two techniques to develop WSA forecasts. After separate standardization and normalization of AL TS and reevaluation of the preprocessed data, the SARIMA (1, 0, 0) (0, 1, 1)12 model outperformed sole LSTM models with correlation index of (R) 0.819, mean absolute error (MAE) of 49.425 and mean absolute percentage error (MAPE) of 0.106. On the other hand, the hybridization (stochastic-DL) enhanced the other characteristics of the estimations, made the reproduction of the primal statistical properties of WSA data more precise and caused better mediation.