Abstract
The objective of this PhD study is to investigate possible ways towards a better integration of spatial observations into the modelling process via spatial pattern evaluation. It is widely recognized by the modelling community that the grand potential of readily available spatial observations is not fully exploited by current modelling frameworks due to the lack of suitable spatial performance metrics. Furthermore, the traditional model evaluation using discharge is found unsuitable to lay confidence on the predicted catchment inherent spatial variability of hydrological processes in a fully-distrbuted model. Therefore this thesis elaborates on possible statistical measures and rigorously tests them at three test sites across spatial scales: (1) a small headwater catchment (0.5 km2) located in Germany, (2) a meso-scale catchment (2500 km2) situated in Denmark and (3) a large-scale domain over the contiguous United Sates (10^6 km2).
To this end, the thesis at hand applies a set of spatial performance metrics on various hydrological variables, namely land-surface-temperature (LST), evapotranspiration (ET) and soil moisture. The inspiration for the applied metrics is found in related fields of environmental science, such as meteorology, geostatistics or geography. In total, seven metrics are evaluated with respect to their capability to quantitatively compare spatial patterns. The human visual perception is often considered superior to computer based measures, because it integrates various dimensions of spatial information in a holistic assessment. Opposed, statistical measures typically only address a limited amount of spatial information. A web-based survey and a citizen science project are employed to quantify the collective perceptive skills of humans aiming at benchmarking spatial metrics with respect to their capability to mimic human evaluations.
This PhD thesis aims at expanding the standard toolbox of spatial model evaluation with innovative metrics that adequately compare spatial patterns. Driven by the rise of more complex model structures and the increase of suitable remote sensing products as observations it can be anticipated that the demand for such measures will grow even more in the near future. The community can benefit from the quantified human perception as a benchmark when introducing and testing novel spatial performance metrics.
To this end, the thesis at hand applies a set of spatial performance metrics on various hydrological variables, namely land-surface-temperature (LST), evapotranspiration (ET) and soil moisture. The inspiration for the applied metrics is found in related fields of environmental science, such as meteorology, geostatistics or geography. In total, seven metrics are evaluated with respect to their capability to quantitatively compare spatial patterns. The human visual perception is often considered superior to computer based measures, because it integrates various dimensions of spatial information in a holistic assessment. Opposed, statistical measures typically only address a limited amount of spatial information. A web-based survey and a citizen science project are employed to quantify the collective perceptive skills of humans aiming at benchmarking spatial metrics with respect to their capability to mimic human evaluations.
This PhD thesis aims at expanding the standard toolbox of spatial model evaluation with innovative metrics that adequately compare spatial patterns. Driven by the rise of more complex model structures and the increase of suitable remote sensing products as observations it can be anticipated that the demand for such measures will grow even more in the near future. The community can benefit from the quantified human perception as a benchmark when introducing and testing novel spatial performance metrics.
| Originalsprog | Engelsk |
|---|
| Forlag | Department of Geosciences and Natural Resource Management, Faculty of Science, University of Copenhagen |
|---|---|
| Antal sider | 221 |
| Status | Udgivet - sep. 2016 |