Enhancement and validation of a state-of-the-art global hydrological model H08 (v.bio1) to simulate second-generation herbaceous bioenergy crop yield

The bioenergy crop yield is a critical determinant of the bioenergy potential for various stringent climate change mitigation scenarios. Currently, the bioenergy crop yield is usually determined from a limited number of simulations. However, reliable yield simulation remains a challenge at the global scale. Here, through parameter calibration and algorithm improvement, we enhanced a state-of-the-art global hydrological model (H08) to simulate the bioenergy yield from dedicated the herbaceous bioenergy crops Miscanthus and switchgrass. Site-specific evaluations showed that the 5 enhanced H08 had the ability to simulate yield for both Miscanthus and switchgrass, with the calibrated yields being well within the ranges of the observed yield. Independent country-specific evaluations further confirmed the performance of the enhanced H08. Using this improved model, we found that unconstrained irrigation more than doubled the yield of the rainfed condition, but reduced the water use efficiency (WUE) by 29% globally. With irrigation, the yield in dry climate zones can exceed the rainfed yields in tropical climate zones. Nevertheless, due to the low water consumption in tropical areas, the 10 highest WUE was found in tropical climate zones, regardless of whether the crop was irrigated. https://doi.org/10.5194/gmd-2019-277 Preprint. Discussion started: 5 November 2019 c © Author(s) 2019. CC BY 4.0 License.

temperature is sensitive to the crop growing days. Ranges from 7 to 10°C for Miscanthus and from 8 to 12°C for upland switchgrass were suggested by Trybula et al. (2015). The calibrated values are within the above ranges. The maximum leaf area indices were calibrated at 11 and 8 for Miscanthus and switchgrass, respectively; these values were identical to those suggested by Trybula et al. (2015).

Site-specific performance of enhanced H08
An overview of the performance of the enhanced H08 is provided in Fig. 3. It can be seen that the performance of the enhanced H08 was improved over that of the original H08, with the tendency of overestimation for switchgrass and underestimation for Miscanthus having been successfully fixed. Points in a scatter plot comparing the simulated yield from 150 the enhanced H08 with the observed yield were well distributed along the 1:1 line. More detailed site-specific results are shown in Figs. 4a (Miscanthus) and Fig. 4b (switchgrass). To depict the uncertainties in the observed yield, the minimum and maximum observed yields were added as error bars in Fig. 4. It was found that the simulated yields were within or close to the range of the observed yield. The simulated relative error was randomly distributed, substantially smaller than the range of the observed yield, and showed no climatic bias. This implies that the combination of the Hun identified by Tryubla et al 155 (2015) and the calibrated parameters of this study are valid for climate zones other than that of the midwestern US, where the Hun was observed. Investigating the performance under the irrigated condition (shown in Fig. S1), we found that H08 performed well at sites 1, 2, and 10, but was out of range at the other sites. This could be attributed to the assumptions of irrigation. H08 assumes that irrigation is fully applied to crops. Therefore, if the reported yield is within the range of that between rainfed and irrigated crops, it is considered reasonable. This was found to be the case, as shown in Fig. S1. To 160 investigate the uncertainty in the meteorological data, a simulation using other meteorological data from the S14FD dataset (Iizumi et al. 2017) was conducted; the results are compared in Fig. S2. The comparison showed that the WFDEI driven result was very similar to that obtained with the S14FD data.

Country-specific performance of enhanced H08
165 Figure 5 compares the yield simulated by the enhanced H08 with the collected independent country-specific yields simulated by MISCANMOD (Clifton-Brown et al., 2004), HPC-EPIC (Kang et al., 2014), andLPJmL (Heck et al., 2016). Here, the yield was simulated under rainfed conditions. For Miscanthus, the correlation coefficient of the yield simulated by H08 and MISCANMOD in the scatter plot ( Fig. 5d) was 0.40. A t-test showed that the correlation was not significant at the 0.01 level.
For consistency with the yield collected by MISCANMOD, any area within a country where the yield is less than 10 Mg ha -1 170 yr -1 was excluded from the analyses. Also, the land available for calculation was set as 10% of the pastureland and cropland.
For switchgrass, the correlation coefficient of the yield simulated by H08 and HPC-EPIC in the scatter plot ( Fig. 5e) was 0.80. A t-test showed that the correlation was significant at the 0.01 level. This indicates that the spatial pattern of the yield simulated by H08 was similar to that of HPC-EPIC. For example, high yields were found in Brazil, Colombia, Mozambique, and Madagascar, while low yields were found in Australia and Mongolia by both models.

175
Miscanthus and switchgrass are not distinguished in LPJmL, and we therefore compared the mixed (mean, Miscanthus and switchgrass) yield of Miscanthus and switchgrass simulated by H08 and the C4 grass yield simulated by LPJmL. The correlation coefficient of the yield simulated by H08 and LPJmL in the scatter plot ( Fig. 5f) was 0.78. A t-test showed that the correlation was significant at the 0.01 level. An additional comparison under the irrigated condition is presented in Fig.   180 S3. The correlation coefficient of the yield simulated by H08 and LPJmL, as shown in the scatter plot (Fig. S3), was 0.95. A t-test showed that the correlation was significant at the 0.01 level. The difference was mainly due to Colombia, Sudan, Mozambique, and Mexico, which are located in tropical zones. The difference in these countries was generally equal to the https://doi.org/10.5194/gmd-2020-179 Preprint. Discussion started: 7 July 2020 c Author(s) 2020. CC BY 4.0 License.
consider the availability of renewable water sources, and planetary boundaries of land, food, and water (Heck et al., 2018).
Finally, as with other models, like MISCANMOD (Clifton-Brown et al., 2004), SWAT (Neitsch et al., 2011), andLPJml (Bondeau et al., 2007), we adopted a crop-uniform water stress formulation. However, an earlier study indicated that the water stress could be crop-specific (Hastings et al., 2009). Additional investigations of the water stress formulation for different bioenergy crops are needed.

4 Conclusion
In this study, we enhanced the ability of the H08 global hydrological model to simulate the yield of a dedicated secondgeneration herbaceous bioenergy crop. The enhanced H08 model generally performed well in simulating the yield of both Miscanthus and switchgrass, with the estimations being well within the range of observations and other model simulations.

295
To the best of our knowledge, this study is the first attempt to successfully enable a global hydrological model with consideration of water management, such as irrigation, to separately simulate the yield of Miscanthus and switchgrass. The enhanced model could be a good tool for the future assessment of the bioenergy-water tradeoffs. With this tool, we quantified the effects of irrigation on yield, water consumption, and WUE for both Miscanthus and switchgrass in different climate zones. We found that irrigation more than doubled the yield in all areas under rainfed conditions and reduced the 300 WUE by 32%. However, due to the low water consumption in tropical areas, the highest WUE was generally found in tropical climate zones, regardless of whether the crop was irrigated.
Code and data availability. The code of the model used in this study is archived on Zenodo

305
Technical information about the H08 model and the input dataset are available from the following website: http: //h08.nies.go.jp.
Competing interests. The authors declare that they have no conflict of interest.