Informations et ressources scientifiques
sur le développement des zones arides et semi-arides

Accueil du site → Master → Pays Bas → Evaluation of Two Automated Remote Sensing-Based Surface Energy Balance Models for Estimating Daily Evapotranspiration

Delft University of Technology (TU Delft) 2016

Evaluation of Two Automated Remote Sensing-Based Surface Energy Balance Models for Estimating Daily Evapotranspiration

Duijndam, L.M.

Titre : Evaluation of Two Automated Remote Sensing-Based Surface Energy Balance Models for Estimating Daily Evapotranspiration

Auteur : Duijndam, L.M.

Université de soutenance : Delft University of Technology (TU Delft)

Grade : Master thesis 2016

Résumé
Remote sensing based surface energy balance models are often used to produce actual evapotranspiration (ET) estimates over large spatial scales , which provide vital information of water consumption for many water resources applications . In the past these m odels have required trained user s fo r a model calibration process, which needed to be completed during every model run. This requirement limited the number of people capable of acquiring the data and the speed at which ET estimates could be generated . Two of these remote sensing based surface energy balance models have recently been automated : the automated Surface Energy Balance Algorithm for Land (SEBAL3.0) and Earth Engine Ev apotranspiration Flux (EEFlux ). Automation of the se mod el s increases the number of potential users and increases the speed at which estimations can be produced. In this study SEBAL3.0 and EEFlux have been used to produce dail y ET estimates for locations around the world with different climates and land covers. These results were com pared with ground measurements from flux tower systems to evaluate the performance of the models. In addition to comparing daily ET estimates to flux tower measurements, SEBAL3.0 estimates for other surface energy balance comp onents were compared to flux t ower measurements . Eighteen flux tower sites were split into two sets. Set 1 was used for the initial model runs and for a calibration of constants used to calculate the hot and cold anchor pixel surface temperatures from the group of hot and cold pixels i denti fied by SEBAL3.0. Set 2 was used for additional runs for EEFlux and for validation of the calibrated constants developed using Set 1. Set 1 SEBAL3.0 results showed systematic underestimations of daily ET caused by systematic overestimations of the sensible heat flux. Set 1 EEFlux results showed significantly different performance levels between the sites in the Netherlands and the sites in the United States. The Netherlands ’ sites had many unreasonable over and underestimates for daily ET. Set 1 SEB AL3.0 daily ET estimates had a RMSE of 2.48 mm/d and EEFlux daily ET estimates had a RMSE of 6.24 mm/d . A calibration performed using select sites and days from Set 1 sites showed a variation between optimal constant values on a spatial and temporal scale. New constants were selected based on the constant combinations with the ma jority of optimum combinations. Although no changes were made to EEFlux between Set 1 and Set 2 runs, Set 2 daily ET results had an improve d RMSE of 4.46 mm/d. Set 2 EEFlux results did not have the same unreasonable estimations as Set 1 , however , several sites estimated negative daily ET rates for many days. Set 2 SEBAL3.0 results had a slightly improved RMSE of 4.16 mm/d with the calibrated constants . SEBAL3.0 inaccurately selected cold pixels for t wo sites in very dry semi- arid regions due to a lack of water bodies . Although there was still a large amount of scatter, sensible heat flux results improved from the cali bration and were no longer systematically underestimated. SEBAL3.0 2 4 hr averaged net radiation for both sets showed a reasonable fit with the measurement data. The cumu lative values of all daily ET estimates for each site were calculated for SEBAL3.0, EEFlux and flux tower measurements. SEBAL3.0 cumulative estimates had percent errors ranging from 12% to 132% and 4 out of 6 sites had cumulative errors under 20%. EEFl ux cumulative estimates had percent errors ranging from 4% and 176% and only 1 out of 6 sites had a cumulative error under 20%.

Sujets  : hydrology ; remote sensing ; evapotranspiration ; flux towers

Présentation

Version intégrale (3,04 Mb)

Page publiée le 14 janvier 2018