Effects of horizontal resolution and air-sea coupling on simulated moisture sources for regional East Asian precipitation

Precipitation over East Asia (EA) simulated in six experiments using the Met Office Unified Model (MetUM) is compared with observations and ERA-Interim reanalysis. These simulations are performed at resolutions of 200 40 km and with both atmosphere-only and air-sea coupled configurations. EA precipitation in MetUM is systematically overestimated, especially over southeastern EA and the Tibetan Plateau. Moisture sources for simulated and observed regional EA precipitation are traced using the Water Accounting Model-2layers (WAM-2layers) a moisture tracking model that traces moisture sources 5 using a combination of evaporation, atmospheric moisture and circulation. Biases in moisture sources are linked to biases in precipitation. For southeastern EA, positive precipitation biases are linked to errors in mid-latitude sources rather than tropical ocean sources. Increasing resolution reduces precipitation biases over the Tibetan Plateau. From the perspective of moisture source, this reduction comes from reduced remote moisture source that is blocked by the better representations of topography at higher resolution. Including coupling does not necessarily improve EA precipitation, however, coupling alters 10 moisture sources. Because the effect of coupling on moisture source varies on location and sign, the collective impact on EA precipitation may not be noticeable. By using WAM-2layers, these changes in moisture sources can be attributed to changes in SST, circulation and associated evaporation. WAM-2layers can be a useful tool to identify model biases that cause biases in regional precipitation.

via the coupler OASIS3 (Valcke, 2013) of 3 hourly frequency. GC2 showed an improvement over previous configurations, particularly in terms of modes of variability, e.g., mid-latitude and tropical cyclone intensities, the Madden-Julian Oscillation and El Niño Southern Oscillation (Williams et al., 2015).
Six MetUM simulations are used, which can be grouped into three pairs. Each pair includes an atmosphere-only simulation (A) and a coupled simulation (C), which have the same atmospheric resolution. Three resolutions are used, 192 longitude points× 90 145 latitude points (N96), 432×325 (N216) and 1024×769 (N512). Therefore, six simulations used here are denoted as AN96, CN96, AN216, CN216, AN512 and CN512. The equivalent atmospheric grid spacing in longitude at the equator is 200km, 90km and 40km, respectively. The atmosphere model has 85 hybrid height levels in the vertical covering 0-85km. The ocean model uses 75 vertical levels and the ORCA025 tri-polar grid which has 0.25 • resolution at the equator. Periods of simulation are listed in Table 1. Simulations match the period of ERA-Interim, except for the N512 simulations, which have a shorter 95 simulation period.

Water Accounting Model-2layers
WAM-2layers is a moisture tracking model developed by van der Ent et al. (2013van der Ent et al. ( , 2014. WAM-2layers is based on the atmospheric water conservation equation and combines information from precipitation, evaporation, atmospheric circulation and moisture to determine sources or sinks of moisture originating from a specified region. In this study, WAM-2layers is applied 100 to backward trace moisture sources for regional precipitation over EA in both ERA-Interim reanalysis and MetUM simulations. Daily precipitation in the target EA region is fed into WAM-2layers, which is integrated backward using circulation and humidity data on model/pressure levels. The domain and magnitude of the moisture source are calculated using the point of last evaporation for precipitation falling in the target regon. A detailed description about WAM-2layers and its setup over EA is given in Guo et al. (2019). As EA crosses several climatic zones and has inhomogeneous hydrological features, EA is first 105 divided into five subregions according precipitation minus evaporation and topography ( Figure 1). These regions are southeastern EA (region 1), Tibetan Plateau (region 2), central-eastern EA (region 3), northwestern EA (region 4) and northeastern EA (region 5). A similar division has been used in Guo et al. (2018), where detailed discussion about the division is given.
3 Differences to observation/reanalysis 3.1 Precipitation 110 Figure 2 shows annual mean EA precipitation in APHRODITE, MetUM AN96 and ERA-Interim, and biases of AN96 and ERA-Interim against APHRODITE. AN96 (Figure 2b) captures the major features of precipitation over EA, i.e., the southnorth precipitation gradient, the precipitation maxima over the Sichuan Basin and southeastern China. However, compared to APHRODITE (Figure 2c), AN96 overestimates precipitation over the Tibetan Plateau, the Sichuan Basin and southeastern China, but underestimates precipitation over the southern slope of the Himalayas. There are also biases over southern Asia, i.e., 115 4 https://doi.org/10.5194/gmd-2020-104 Preprint. Discussion started: 8 May 2020 c Author(s) 2020. CC BY 4.0 License. the Indian Peninsula, Bangladesh and the Indochina Peninsula. These biases are also common in other MetUM simulations and have been reported in previous studies (e.g., Stephan et al., 2018a). Comparing ERA-Interim to APHRODITE, Figure 2e shows that ERA-Interim overestimates precipitation over southwestern China and the Tibetan Plateau. These biases in ERA-Interim will affect the accuracy of the moisture source calculations over these regions. However, using ERA-Interim precipitation remains as only option because it is physically consistent with other ERA-Interim variables, i.e., wind, humidity and evaporation, 120 therefore offers a reasonably closed water budget for WAM-2layers.
These biases are also reflected in seasonal and regional means over EA subregions (Figure 3). Both ERA-Interim and Me-tUM simulations overestimate precipitation over southeastern EA (region 1), with biases in MetUM being larger. Precipitation over the Tibetan Plateau (region 2) is also overestimated in ERA-Interim and MetUM. With increased resolution, biases in simulations reduce, especially with comparing from low (N96) to medium resolution (N216). A detailed analysis of the resolution-125 related changes in moisture sources is discussed in Section 4.1. Biases over regions 3 and 4 are smaller in coupled simulations compared to atmosphere-only simulations, especially in JJA and in medium and high-resolution simulations. This difference is due to a weak western North Pacific subtropical high (WNPSH) is in the coupled simulations (not shown). The weak WNPSH reduces moisture transport from low latitudes (region 1) to mid latitudes (regions 3 and 4), which therefore reduces the positive biases (Figure 4f, h). This weak WNPSH has also been identified in previous studies (e.g., Rodríguez et al., 2017). To further 130 investigate these biases, moisture sources of simulated EA precipitation are traced and compared to ERA-Interim. Figure 4 shows the annual mean moisture sources and the vertically integrated moisture fluxes for five EA subregions using ERA-Interim. The differences between AN96 and ERA-Interim are also shown in Figure 4. Compared to ERA-Interim, AN96 transports less moisture from low latitudes but more from middle latitudes for all EA subregions. These differences are largely 135 associated with differences in moisture fluxes (Figure 4b, d, f, h and j). In AN96, the cross-equatorial flow along the Somali Jet is too weak, but the mid-latitude westerlies is too strong. The moisture flux over region 1 is too zonal, which is consistent with a weak WNPSH, illustrated as a cyclonic moisture flux anomaly shown on Figure 4b. These differences cause less moisture to be transported from low latitude to middle latitude and reduced moisture contribution from southern China to regions 3, 4 and 5 in AN96 (Figure 4f, h and j). Over the Tibetan Plateau, AN96 transports less moisture than ERA-Interim over the whole 140 source domain, except eastern Tibet. This indicates that the simulated precipitation over the Tibetan Plateau in AN96 relies more on local sources but less on remote sources.

Moisture source
The contribution of local moisture to precipitation is measured using the precipitation recycling ratio, defined as the proportion of precipitation in the target region that is contributed by the evaporation over the same region. Figure 5 shows the annual cycle of the precipitation recycling ratio for five EA subregions. The simulations can reproduce the annual cycles over regions 145 1, 3 and 4 (Table 2), but overestimate the recycling ratio over regions 2 (summer and autumn) and 5 (spring and autumn).
The simulated and ERA-Interim remote moisture sources are compared using its shape. To assist comparison, instead of showing maps of seasonal moisture source, the mass centres of the moisture sources from the simulations and ERA-Interim are compared collectively in Figure 6. Figure 6 shows the mass centres for all five regions in DJF and JJA. The mass centres are measured using the moisture source regions that account for 80% of precipitation in the target regions, similar to those in Figure 4. The mass centres were also measured using threshold at 50% and 65% of precipitation; the results are consistent and not sensitive to the choice of threshold. As shown in Figure 6, the simulated mass centres show consistent seasonal variations as in ERA-Interim. However, there are systematic differences between the simulations and ERA-Interim, as well as among the simulations.
Over region 1 in JJA (coloured triangles in Figure 6a), the mass centres of simulated moisture sources are located approxi-  Figure 6d, e). This is because regions 4 and 5 are situated over the mid-latitudes, therefore, the tropical impacts associated with the EASM are less. In DJF over regions 4 and 5 (coloured circles in Figure 6d, e), simulated mass centres are located east of the ERA-Interim mass centre, especially, for high-resolution simulations, CN216, AN512 and CN512, wherein the eastward shift can be as large as 30 • . To explain this difference, moisture sources for region 5 in DJF from 165 CN512 are compared to ERA-Interim ( Figure 7f). In CN512, too much moisture originates from regions east of region 5, i.e., from the Seas of Japan and Okhotsk, but too little from regions west of region 5, especially, from the Mediterranean Sea, the Red Sea and the Persian Gulf. This eastward shift is smaller in low-resolution simulations (AN96, CN96 and AN216). More details are discussed in Section 4.3.
The remote moisture sources are further divided into four sections (tropical sea, tropical land, extra-tropical sea and extra-170 tropical land). Together with the local source (measured as the precipitation recycling ratio), the contributions from these sections are compared in Table 3. For the annual mean contribution, the simulations can reproduce the primary sources identified from ERA-Interim for all EA subregions, i.e., tropical sea for region 1 and extra-tropical land for regions 2, 3, 4 and 5.
On the other hand, for seasonal mean contributions, discrepancies between the simulations and ERA-Interim are large. Even the primary sources are different between the simulations and ERA-Interim (as boldface values highlighted in Table 3). Since 175 these discrepancies are also sensitive to resolution and coupling, more details are discussed in Section 4.3.

Differences in moisture source due to model resolution and coupling
Comparisons in the previous section show that MetUM simulations have systematic biases against ERA-Interim in both precipitation and moisture sources. Despite increasing resolution and including coupling, these systematic biases between MetUM simulations and ERA-Interim remain. However, there is evidence to show that precipitation and moisture sources are sensitive 180 to resolution and coupling. Therefore, these changes in moisture source are discussed here; links between changes in moisture source and precipitation are also made in the section.

Change with resolution
As shown in Figure 3b, over the Tibetan Plateau, the simulations have smaller precipitation biases against APHRODITE than ERA-Interim, especially in MAM and JJA. From the perspective of moisture source (Figure 4d), this improvement is attributed 185 to the reduced remote contribution. Also shown in Figure 3b, increasing resolution reduces the mean precipitation bias over the Tibetan Plateau. Similar results have also been found in previous studies (e.g., Curio et al., 2015), which showed that higher resolution improves the representation of topography in models; steeper topography blocks remote moisture transport. On the other hand, the local moisture contribution is enhanced with resolution ( Figure 5b) because the steeper topography reduces the outflow of local moisture. In MetUM simulations, the enhanced local moisture source is located over eastern Tibet (Figure 8a,   190 b and c). This is partly because eastern Tibet is wetter than western Tibet, partly because the north-south orientated valleys over southeast Tibet are the major pathway for remote moisture. By reducing the remote contribution, the local contribution becomes larger. This is demonstrated (Figure 8d) by the opposing trends in the tracked local evaporation over eastern Tibet (increase) and the low-level wind along its southern boundary (decrease).

Change with coupling 195
To investigate impacts of coupling on moisture sources, we focus on region 1, where the oceans are the major contributors (according to Table 3). Differences in moisture source over region 1 in JJA between coupled and atmosphere-only simulations in different resolutions are shown in Figure 9. Regardless of resolution, coupled simulations always show consistent differences against atmosphere-only simulations, which include a reduced moisture contribution from the Indian Ocean but an increased moisture contribution from the Pacific Ocean. The reduced moisture contribution over the Indian Ocean is linked to the cold 200 SST bias over the Arabian Sea, as demonstrated in Figure 9d. This cold SST bias reduces local evaporation, which reduces its contribution to precipitation over region 1. This cold SST bias has also been reported in previous studies with MetUM (e.g., Marathayil et al., 2013). On the other hand, there is not a consistent SST bias associated with the increased moisture contribution over the Pacific Ocean ( Figure 9d). Instead, there is a consistent increase in the low-level wind in all coupled simulations ( Figure 9d). As mentioned in Section 3, this wind bias is due to the EASM flow being too zonal in all coupled 205 simulations. The positive low-level wind bias over the South China Sea intensifies local evaporation. Associated with this wind bias is a weak WNPSH, which is demonstrated in Figure 9a, b and c as the cyclonic anomaly over the southeast coast of EA. This cyclonic anomaly converges the enhanced evaporation over the same location and transports more moisture from the South China Sea into region 1. As these two changes in moisture sources due to coupling show opposite signs and similar magnitudes, when attributing them collectively to changes in precipitation, the differences of precipitation over region 1 in JJA 210 between atmosphere-only and coupled simulations are small (Figure 3a). However, it does not necessarily mean that changes in SST, evaporation or circulation are also small. terms of reproducing the mean and interannual variability of EA precipitation (Lin et al., 2014;Su et al., 2015).
The length of simulations for N512 is shorter than other simulations (Table 1). This may cause inconsistency when comparing results from different resolutions. Only one ensemble member per simulation is used in this study. Therefore, the robustness of the results will need further test when more ensemble members are available.

250
In this study, moisture sources of EA precipitation simulated by MetUM are traced using WAM-2layers and compared to ERA-Interim. The purposes of this study are: first, to link systematic biases in simulated EA precipitation to biases in moisture sources; second, to investigate sensitivities of moisture sources to model atmospheric horizontal resolution and air-sea coupling.

255
MetUM simulations can reasonably capture patterns of EA precipitation but show systematic biases against observation regardless of resolutions or coupling. These biases include overestimations over southeastern China and the Tibetan Plateau and underestimations over the southern slope of the Himalayas. Simulated precipitation is sensitive to both resolution and coupling. However, systematic errors in precipitation between simulations and observations cannot be eliminated with either resolution or coupling.

260
Moisture sources traced from simulations agree with those from ERA-Interim in terms of the seasonal cycles of local moisture contribution and mass centres. However, systematic differences in moisture sources between simulations and ERA-Interim are noticeable. Simulated precipitation over southeastern EA (region 1) have smaller contributions from tropical oceans but more from mid-latitude sources, which is due to a weak cross-equatorial moisture flux over the Indian Ocean. For the same reason, all simulated mass centres of region 1 are situated north of the mass centre in ERA-Interim, especially in JJA. Although 265 the mean precipitation biases are small for mid-latitude EA regions (regions 4 and 5), simulated mass centres, however, are situated east of the mass centre in ERA-Interim, especially in DJF. This is because moisture in simulations originates from the Pacific Ocean for precipitation over mid-latitude EA, while it mainly originates from the Mediterranean Sea and the Atlantic Ocean in ERA-Interim. Simulated precipitation over the Tibetan Plateau has smaller biases compared to ERA-Interim, which is manifested as a reduced remote contribution but an enhanced local contribution in moisture sources.

270
Although increasing resolution can not eliminate systematic errors in precipitation between MetUM simulations and ERA-Interim, improvements are noticeable, especially over the Tibetan Plateau. Better representation of topography at higher resolutions reduce remote moisture contribution and enhance local moisture contribution. This change is more noticeable over the eastern Tibetan Plateau, where the sruface is wetter and the remote moisture can enter its southern boundary along the north-south orientated valleys.

275
EA precipitation is not necessarily improved with coupling. However, the moisture sources associated with EA precipitations can still be changed. For example, although differences in precipitation with coupling over region 1 in JJA are small, a negative moisture source change over the Arabian Sea and a positive change over the South China Sea are found in all cou-9 https://doi.org/10.5194/gmd-2020-104 Preprint. Discussion started: 8 May 2020 c Author(s) 2020. CC BY 4.0 License. pled simulations. These changes are associated with changes in SST and circulation that are introduced by including air-sea coupling. These changes in moisture sources have different signs, therefore, the collective impact on precipitation over a target 280 region cannot necessarily be detected through analysis of mean precipitation alone. However, this study shows the usefulness of WAM-2layers in identifying model biases in a range of variables that relate to precipitation, and evaluating the sensitivity of those biases to changes in model resolution or physics.
Simulations at higher resolutions capture a seasonal shift of the major moisture contributor over regions 4 and 5 in DJF, in which the major contributor shifts from mid-latitude land to mid-latitude ocean. In reanalysis, this shift is caused by a 285 reduced land surface evapotranspiration and an enhanced moisture transport via mid-latitude westerlies from oceans west of the target regions. However, none of the simulations can reproduce this mechanism. Instead, shifts in simulations are caused by an increasing moisture source over the Pacific, which is related to the increase in resolution.
In summary, we have shown in this study that, to better understand precipitation biases in simulations, it is necessary to go a step further by connecting precipitation biases to biases in other related variables using a moisture tracking tool that is built 290 upon the atmospheric moisture conservation equation. We have also shown that the WAM-2layers is suitable candidate for this purpose. Especially, for its computational efficiency, it can be readily applied simulations with large ensembles and resolutions, such as, the Coupled Model Intercomparison Project Phase 6.