| dc.description.abstracteng | Evapotranspiration (ET) is a central flux in the hydrological cycle on a regional and on a global scale. Transpiration from plants is the largest water flux from terrestrial surfaces, accounting for the major part of terrestrial ET. This thesis comprises method comparisons of methods for ET and plant-water-use analyses. Established and well-tested methods are compared with recently emerging methods based on drone remote sensing thermography and modelling approaches. The presented studies were developed and realized in the frame work of the Collaborative Research Centre 990 and the subproject A02. Based on the methods comparisons, a standardized workflow for ET prediction by land surface temperatures was implemented in a user friendly open-source software for ET calculation. The primary goal of this thesis was to test and evaluate the new possibilities that result from the application of drones, radiometric thermal cameras as well as the utilization of causal and machine learning models. Therefore, method comparisons between well-known and tested reference methods and new drone-based methods are implemented. Another objective of this thesis was to streamline measurement efforts in the field and therefore evaluate the most important variables to measure to provide precise predictions. Based on the results of these objectives a standardized workflow for the calculation of ET based on thermal images from a variety of sources is developed and implemented into an open-source software available to a wide range of potential users. For the first study, thermal images of land surface temperatures were recorded in 61 drone recording flights on five days over a commercially managed oil palm plantation. To predict ET from the thermal images three energy-balance-models (EBMs) were applied: The relatively simple one-source EBM DATTUTDUT (Timmermans et al., 2015) and the more complex two-source EBMs TSEB-PT (Norman et al., 1995) and DTD (Norman et al., 2000). Latent heat flux estimates of the DATTUTDUT model with measured net radiation agreed well with eddy covariance measurements (r$^{2}$=0.85; MAE=47; RMSE=60) across variable weather conditions and day times. Confidence intervals for slope and intercept of a model II Deming regression suggest no difference between drone-based and eddy covariance method, thus indicating interchangeability. TSEB-PT and DTD yielded agreeable results, but all three models are sensitive to the configuration of the net radiation assessment. Overall, we conclude that drone-based thermography with energy-balance modeling is a reliable method complementing available methods for evapotranspiration studies. It offers promising, additional opportunities for fine grain and spatially explicit studies. The second study focused on plant transpiration as a key element in the hydrological cycle. Widely used methods for its assessment comprise sap flux techniques for whole-plant transpiration and porometry for leaf stomatal conductance. Recently emerging approaches based on surface temperatures and a wide range of machine learning techniques offer new possibilities to quantify transpiration. The focus of this study was to predict sap flux and leaf stomatal conductance based on drone-recorded and meteorological data and compare these predictions with in-situ measured transpiration. Therefore, a comparatively large data set, consisting of 103 drone recording flights, two weeks of sap flux measurements in 10 min intervals and thousands of stomatal conductance measurements was recorded. The data collection was focused on oil palm, as well as on four local tree species that are native to the Sumatra region. Since no causal, but strictly data-driven machine-learning (ML) models were applied, such a large data set would be necessary for accurate prediction results. The models applied were a multiple linear regression that would serve as a simple base line method to compare with the more complex ML algorithms. The ML algorithms used in the study were a support vector machine (SVM), two types of random forest algorithms (RF) and an artificial neural network (ANN). Random forest predictions yielded the highest congruence with measured sap flux (r$^{2}$=0.87 for trees and r$^{2}$=0.58 for palms) and indicated differences among tree species. Confidence intervals for intercept and slope of a Passing-Bablok regression suggest interchangeability of methods for sap flux prediction using random forest. However, the other algorithms also showed promising results, especially for the prediction of sap flux from oil palm. Predictions for stomatal conductance were less congruent, likely due to spatial and temporal offsets of the measurements. Overall, the applied drone and modelling scheme predicts whole-plant transpiration with high accuracy, especially using random forest algorithms. Various approaches to compute ET via energy balance models exist, but their handling is often complex and challenging. The prevalent aim of the third study was the development of a user friendly open-source software, that would implement central findings of the first two studies and that would be publicly available as an extension for the geospatial QGIS3 platform. Special emphasis was put on the option to use thermal maps from a variety of sources including drones, satellites and handheld thermal cameras. As the previously mentioned studies showed, measured radiation and meteorological variables contribute significantly to the prediction accuracy. An option to input measured variables has therefore been added to the software. To test the performance of the software, land surface temperatures from an oil palm plantation were recorded using a drone and a handheld thermal cameras and radiation measurements were further used as optional input. Typical daily ET patterns were found with all model configurations for both recording types. However, the precision of the ET estimates by the software was significantly improved using solar radiation measurements. QWaterModel is compatible with all versions of QGIS3 and is available from the official QGIS Plugin Repository. In summary, this thesis shows that evapotranspiration and plant-water-use prediction approaches based on drone thermography and subsequent modelling with causal as well as machine learning models are a useful extension or even a potential alternative to well-proven methods such as eddy covariance, sap flux, stomatal conductance. However, using only drone recorded data was often not enough since all predictions benefited from additional information on solar radiation as well as from measurements of relative humidity and air temperature. The open-source software developed in the scope of this thesis is at this time available in four versions and more than 1100 downloads were registered up to this moment. | de |