The results in Sect.
Hydrological modelling of stream flows in the Rmel watershed using SWAT model
However, the best possible PDM parameters vary for each catchment. In this section, we describe how we developed a universal method for optimising parameter estimation based on topographic data. The marker size represents the NS values larger circles for higher NS values. For those catchment in which mcs does not select the test of the best NS metric Tay, Ure, Derwent, Avon , the best-performing tests are also represented with a degree of transparency.
In Fig. Plots on the left in Fig. On top of each plot the name of the catchment and the performance metrics mean bias and the NS are given. Hence, we propose a new criterion for simulating river flow for catchments in Great Britain Fig. When applying this new mcs criterion daily river flow time series and evaluation metrics shown in Fig. A clear exception is the Avon catchment, where the NS metric is reduced from 0. The other catchments where mcs does not exactly match the best-performing tests are the Tay, Ure and Derwent.
By applying the mcs criterion we are able to introduce catchment variability in the JULES performance related to different topography characteristics, which reaches high NS values in flat catchments, such as the Thames, with baseflow-dominated runoff, and in steeper catchments, such as the Ouse and Ribble, with fast surface runoff production during rainfall events and low BFI. A key driver for this work in the context of developing a UK regional coupled environmental prediction system UKC2; Lewis et al. Effectively, the inclusion of this slope dependency limits the saturation excess runoff production within flatter regions in wet situations, and it enhances runoff production within steeper regions due to a high b parameter with no limitation due to soil water content.
Overview of Runoff
We stress at this point that the results do not have predictive skill at the Avon and Ock catchments. The Avon is the main outlier in this study due to an unsaturated chalk zone which has known fast flow in the subsurface and might require a different soil hydrology modelling altogether Rahman and Rosolem, ; Blyth et al. We define three performance categories in Fig. It has shown very high performance for catchments in Great Britain Crooks et al.
The infiltration excess surface runoff is rarely invoked in JULES as the rate of water reaching the surface at each time step does not reach the maximum infiltration rate, which is defined as the saturation conductivity at the upper soil layer enhanced by a land cover factor Best et al. Although the bias shows very little improvement from the no hyd runs due to a low estimation of surface runoff, the NS metric shows an improvement in all catchments, as the surface runoff production by saturation excess is active and the rainfall peaks do produce river flow peaks. The baseflow production during dry periods is not as delayed as it is in the no hyd runs Fig.
The markers in different shades of blue in Fig.
Tang et al. Simulated river flows using LSMs will result from physical processes represented in the model and the imposed meteorological driving data. Both of these factors affect the simulations on a range of different timescales. In particular, this allows assessment of the average amplitude of discharge on different timescales and, separately, the average phase difference lead or lag of the modelled compared to the observed discharge Weedon et al.
The ideal model performance at a particular frequency leads to an amplitude ratio of exactly 1. For clarity in Figs. Here positive phase differences mean that the model variations lag behind the observations and negative values indicate the model leading the observations. For each catchment the top two panels show the power or variance spectra.
In the form of spectral analysis applied here the power directly indicates the mean squared amplitude at each frequency rather than the area under the plot; Weedon et al. The ideal model performance results in amplitude ratios third row indistinguishable from 1. Theoretical phase-difference trends are shown with black dashed lines bottom row.
Note that the JULES discharge performance was assessed against observations with cross-spectral analysis by Weedon et al. Here RFM routing was applied subdaily, thereby avoiding the artefacts. The timescales on which amplitude ratios and phase differences have been assessed are annual, slow-response scale SR and quick-response scale QR. The upper limits of the SR and QR timescales are determined for each catchment as the time that river flow takes to flow from the upper most point of the catchment to the outlet, using the wave speeds that RFM uses in JULES for subsurface and surface flow, respectively.
The lower limits of the timescales are defined as one-third of the upper limits. Results of the cross-spectral analysis of the daily river flow power spectra, amplitude ratio spectrum and phase difference or phase spectrum are shown in Fig.
Selected other publications
If a time series is compared to itself, but offset by a few time steps, there is a resulting trend in the high-frequency part of the phase-difference spectrum Eq. A10 in Weedon et al. The modelled phase-difference trends that approximate the results are shown using black dashed lines in Fig. Note that phase differences distinguishable from zero degrees can result from both simple offsets in the timing of model output compared to the observations causing phase-difference trends and from incorrect modelling of the response times of hydrological stores Weedon et al.
A further comparison between amplitude and phase differences with observations using different parameterisations and on different timescales helps to clarify the implications of our final parameterisation and model development Fig.
The annual-scale spectral peak is very marked in the wet northern Dee catchment, and all flavours of JULES represented capture it accurately amplitude ratios are close to 1. However, in the Severn2 and Thames catchments we start to see compromised performances on the annual scale. For each catchment, amplitude ratios red and phase differences blue are shown on the annual scale top two rows , SR scale middle two rows and QR scale bottom two rows. On the QR scale we expected results to resemble the NS metric analysis, and we find that over the three catchments the new parameterisation results are the best or as good as the mcs criterion results as seen in Sect.
The cross-spectral results illustrated in Figs. To our knowledge, this is the first study that analyses river flow model outputs from a LSM over a wide enough area the 13 selected catchments driven by the CHESS-met dataset Robinson et al.
CO Meeting Organizer EGU
The availability of this dataset opens new possibilities to study land surface hydrology and interactions with the atmosphere using LSMs that typically require gridded forcing datasets on the kilometre scale driven by gridded rainfall derived from gauge stations. A recent study Blyth et al. These authors find that, when compared to flux tower data, the model overestimates evapotranspiration rates. The sources of this evaporation bias are beyond the scope of this work and other studies in the model community are investigating the issue e.
Blyth et al. However, the new runoff development reduces the negative runoff bias as shown here, mostly from increased surface runoff during the rainy season over mountainous regions. Hence, the evapotranspiration rates in the Blyth et al. Knowing whether this reduction in evapotranspiration in Great Britain by lower soil moisture availability is consistent with soil moisture observations remains a challenge. However, we have found that for regional integration in Great Britain the surface runoff production by PDM allows for a better characterisation of the topographical variability through the S 0 parameter.
This finding within the JULES model and the regional framework of Great Britain can have significant impacts over other regions and be applied to other models that need to account for subgrid variability in the runoff generation process, using a widely available parameter from digital elevation model datasets like the grid-cell mean slope as the only input, whereas other physical characteristics might be more difficult to obtain or are simply unavailable. The poor performance at the Avon catchment by the PDM scheme has not been solved by our new parameterisation, pointing to geological rather than topographical characteristics driving the subsurface water flow Rahman and Rosolem, We argue that a combination of the PDM scheme for surface runoff generation and TOPMODEL, or another scheme that incorporates the representation of groundwater dynamics and persistence at the subsurface e.
Fan et al. We stress that the issue of the infiltration excess surface runoff rarely being invoked in JULES needs to be further investigated e. Mueller et al. The maximum infiltration rate that will produce surface runoff is theoretically reached from sudden and intense rainfall events.
Such a rate is difficult to reach by LSMs, since they are currently driven by precipitation datasets that lack the temporal and spatial resolution necessary to correctly represent intense rainfall events Balsamo et al. The saturation excess surface runoff is overwhelmingly dominant in this study and might be compensating for underestimations of infiltration excess. The performance loss when using grid-slope dependency instead of mean catchment slope dependency is a compromise that we accept since our development and recommended configuration need to be applicable for the whole of Great Britain and even for other regions and space scales where particular catchment information might not be available.
We acknowledge that human activities groundwater abstractions, dams, reservoirs affect the observed river flow in Great Britain and therefore JULES outputs of natural river flow are not expected to exactly reproduce the observed NRFA records. As mentioned in Sect. However, the effects of human activities on river flow are difficult to quantify given the lack of data and heterogeneity of activities in the studied catchments. A recent study over Great Britain, for instance, showed an increase in drought duration in catchments affected by groundwater abstractions and varying effects on drought occurrence depending on the activities Tijdeman et al.
The model development described here on the kilometre scale and over the domain of Great Britain is based on the inclusion of a terrain slope dependency in the soil wetness parameter that switches on the saturation excess runoff scheme. Even though the parameter values need to be re-examined for other regions and resolutions, this physical dependency should also be valid on the global scale and its implications in the performance of the JULES model global simulations at 0.
Motivated by the search for the best representation of hydrological processes over the land in the context of a coupled UK land—ocean—atmosphere model UKC2; Lewis et al. Best et al. The parameter S 0 controls the soil water content necessary to start producing surface runoff. The parameterisation that produces the best results for each catchment uses the mean catchment slope. When applied to a gridded model, a new linear function of slope on the model resolution scale can produce performance metrics comparable to those using the mean catchment slope.
The new parameterisation constrains surface runoff production to wet soil conditions over flatter regions, whereas over steeper regions the model produces surface runoff for every rainfall event, regardless of the soil wetness conditions. We stress that this finding should be tested for other regions and scales on JULES and other LSMs, as topography datasets are available at very fine resolution e.
The capability of an LSM to reproduce the water balance on regional scales with a performance in terms of river flow generation comparable to that of hydrological models can potentially impact weather forecast and climate predictions using regional coupled modelling systems such as UKC2. We have also shown that cross-spectral analysis for evaluating model performance against observations quantifies the mismatches in variability and separately mismatches in phase at different timescales that are not otherwise apparent from metrics such as NS and RMSE.
The cross-spectral analysis comparing the modelled river flow with observations has reinforced the choice of the new parameterisation for surface runoff production. The new version of the model with the new parameterisation recommended here has been used to study evaporation and water budgets in Great Britain during the last 55 years by Blyth et al. The development is also incorporated into the UKC2 system Lewis et al. The new code development is described in ticket no. The CHESS-met driving data and the rest of the ancillary datasets used here are publicly available through references given in Sect.
The SACSMA model is ideally suited for the simulation of large drainage basins — greater than 1, square kilometers.