본문 바로가기
  • Home

Deep Learning-based Estimation and Mapping of Evapotranspiration in Cropland using Local Weather Prediction Model and Satellite Data

  • Journal of the Korean Cartographic Association
  • Abbr : JKCA
  • 2018, 18(3), pp.105-116
  • DOI : 10.16879/jkca.2018.18.3.105
  • Publisher : The Korean Cartographic Association
  • Research Area : Social Science > Geography > Geography in general > Cartography
  • Published : December 31, 2018

Soo-Jin Lee 1 Kwang-Jin Kim 1 Yeong-Ho Kim 1 Ji-Won Kim 1 Seong-Wook Park 1 Ye-Seul Yun 1 Kim, Na Ri 2 Yang-Won LEE ORD ID 3

1부경대학교 지구환경시스템과학부 공간정보공학전공
2부경대학교 지오메틱연구소
3부경대학교

Accredited

ABSTRACT

Evapotranspiration is an important factor for Earth energy balance which transports vapor into atmosphere in the form of latent heat using net radiation energy. Measuring the evapotranspiration is actually limited to a point, so the modeling on a spatially continuous grid has been conducted for a long time using meteorological and satellite data. In addition to the PM (Penman-Monteith) equation-based METRIC (Mapping Evapotranspiration with Internalized Calibration) model and the PT (Priestley-Taylor) equation-based MS-PT (Modified Satellite-based Priestley-Taylor) model, the DNN (deep neural network) as an emerging technique can be a viable option. We conducted a DNN modeling of evapotranspiration through optimization for hidden layer structure, loss function, optimizer, active function, L1/L2 regularization, and dropout ratio. The result showed a quite favorable accuracy with the RMSE of 0.326 mm/day and the correlation coefficient of 0.975. This is because we used optimal input variables associated with the mechanism of evapotranspiration from numerical weather prediction model and satellite data, in addition to the DNN optimization. However, a more delicate modeling by increasing the volume and kind of training dataset will be necessary for future work.

Citation status

* References for papers published after 2022 are currently being built.