Estimating reference evapotranspiration in an arid region using a MODIS sensor and machine learning algorithms

Creative Commons License

Talebi H., Samadianfard S., Çıtakoğlu H.

9th Advanced Engineering Days (AED), Tabriz, İran, 9 - 10 Temmuz 2024, ss.1-3

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Basıldığı Şehir: Tabriz
  • Basıldığı Ülke: İran
  • Sayfa Sayıları: ss.1-3
  • Erciyes Üniversitesi Adresli: Evet

Özet

Agricultural water management and irrigation planning depend on accurate estimation of reference evapotranspiration (ET0). By using satellite images, it is possible to compensate for the lack of weather information in areas without meteorological stations. Therefore, we have implemented the random forest (RF) and multi-layer perceptron (MLP) algorithms in this study to estimate the monthly ET0 at Ahvaz station (dry climate) using parameters extracted from MODIS sensor images. Then the estimated values of monthly ET0 and calculated monthly ET0 were compared with the FAO-Penman-Monteith equation. Based on the data of satellite images collected from 2005 to 2023, a database was created. During the development of the above models, 75% was used for the model training phase and 25% for the model testing phase. The input variables consisted of land surface temperature (LST) and evapotranspiration index (ET). Results showed that the MLP-1 model with only one LST input parameter had the best accuracy for monthly ET0 estimation in Ahvaz station (R2=0.973, RMSE=0.387, and NS=0.928).Finally, by examining the results of both models, it was found that the MLP model has better accuracy than the RF model in all scenarios.

Keywords: evapotranspiration, machine learning, remote sensing, MODIS, land surface temperature