DescriptionReliable quantitative, high resolution, spatially disaggregated and easily obtainable information about water resources is essential for assessments and research relating to water. Relatively reliable information can be obtained via different hydrological and land surface models, however their application requires (1) skills that potential users may not have, (2) resources to set up a modelling exercise, and (3) data, which may not be available in sufficient quality or resolution. On the other hand, outputs from global/regional models are readily available online, but they may not provide information at desired resolution.
Downscaled climate input data is regularly used in hydrological modelling applications, but the outputs of hydrological models are rarely, if ever, downscaled. In this study we (1) downscale off-the-shelf runoff products using spatial relationship between areal units of runoff and explicit river network (with R package ‘hydrostreamer’), (2) perform data assimilation via model averaging, (3) compare the performance of downscaled and model averaged discharge prediction, (4) investigate the spatiotemporal variation in model averaging weights, and their uncertainty, and (5) regionalize the model weights to arbitrary points in the river network. Our approach has minimal data requirements; only a runoff timeseries, a river network (e.g. HydroSHEDS), and some river discharge observations. Uncertainty in the downscaling process is assessed in another presentation (Virkki et al, 2018).
Our results in the 3S (Sekong, Sesan and Srepok) tributaries of the Mekong River suggest that downscaling improves the performance compared to discharge outputs from global models. Performance is significantly improved further by model averaging. However, the weights assigned to each model vary significantly over time and at different observation stations, highlighting the role of different strengths of each model over time and space. It therefore makes sense to use an ensemble of weights when providing regionalised model averages across sub-catchments.
It is possible to downscale runoff using relatively little data and technical expertise. It can produce useful discharge estimates at arbitrary river segments, and significantly improve performance of streamflow estimates from global models.
|Aikajakso||14 joulukuuta 2018|
|Tapahtuman otsikko||AGU Fall Meeting: null|
|Sijainti||Washington, Yhdysvallat, District of Columbia|
Asiakirjat ja linkit
Tähän liittyvä sisältö