Insights into seasonal variation of wet deposition over Southeast Asia 1 via precipitation adjustment from the findings of MICS-Asia III

. Asia has attracted research attention because it has the highest anthropogenic emissions in the world, and the 26 Model Inter-Comparison Study for Asia (MICS-Asia) Phase III was carried out to foster our understanding on the status of 27 air quality over Asia. This study analyzed wet deposition in Southeast Asian countries (Myanmar, Thailand, Lao People's 28 Democratic Republic (PDR), Cambodia, Vietnam, the Philippines, Malaysia, and Indonesia) with the aim of providing 29 insights into the seasonal variation of wet deposition. Southeast Asia was not fully considered in MICS-Asia Phase II due to 30 a lack of observational data; however, the analysis period of MICS-Asia III, namely, the year 2010, is covered by ground 31 observations of the Acid Deposition Monitoring Network in East Asia (EANET), and the coordinated simulation domain 32 was extended to cover these observation sites. The analyzed species are wet depositions of S (sulfate aerosol, sulfur dioxide 33 (SO 2 ), and sulfuric acid (H 2 SO 4 )), N (nitrate aerosol, nitrogen monoxide (NO), nitrogen dioxide (NO 2 ), and nitric acid 34 (HNO 3 )), and A (ammonium aerosol and ammonia (NH 3 )). The wet deposition simulated with seven models driven by 35 unified meteorological model in MICS-Asia III was used with the ensemble approach, which effectively modulates the 36 differences in performance among models. By comparison with EANET observations, although the seven models generally 37 captured the wet depositions of S, N, and A, there were difficulties capturing these in some cases. This failure of models is 38 considered to be related to the difficulty in capturing the precipitation in Southeast Asia, especially during the dry and wet 39 seasons. To overcome this, a precipitation-adjusted approach which scaling the modeled precipitation to the observed value 40 was applied, and it was demonstrated that the model performance was improved. Satellite measurements were also used to 41 adjust for precipitation data, which worked well to account for spatio-and-temporal precipitation patterns, especially in the 42 dry season. As the statistical scores were mostly improved by this adjustment, the estimation of wet deposition with 43 precipitation adjustment was considered to be superior. To utilize satellite measurements, the spatial distribution of wet 44 deposition was revised. Based on this revision, it was found that Vietnam, Malaysia, and Indonesia were upward-corrected 45 and Myanmar, Thailand, Lao PDR, Cambodia, and the Philippines were downward-corrected; these corrections were up to 46 ±40%. The improved accuracy of precipitation amount was key to estimating wet deposition in this study. These results 47 suggest that the precipitation-adjusted approach has the potential to obtain accurate estimates of wet deposition through the 48 fusion of models and observations. Hereafter, precipitation of 50 mm/month is used as the index to divide the dry and wet seasons. Based on this criterion, the dry and wet seasons were clearly characterized from observed precipitation; however, the model simulated light precipitation of around 20 mm even during the dry season, and underestimated precipitation during the wet season. Due to the seasonal variation in the observed precipitation, the observed wet deposition of S, N, and A also exhibited a clear seasonal dependency during the dry and wet seasons. Compared with the observed wet deposition, the model generally overestimated the wet deposition during the dry season and underestimated it during the wet season. These results indicate

that the model performance for precipitation could be a critical factor in determining the model performance for wet 159 deposition. The statistical performance of the simulated wet deposition of S, N, and A is listed in Table 2. The ENS results  160 showed a good correlation with the observed data, with an R of around 0.8; however, there was a large underestimation for 161 wet deposition, with an NMB greater than −70% and an NME greater than 80%. As suggested by the observed monthly wet 162 deposition amount shown in  Table 3. ENS showed underestimation for the wet deposition of S, N, and A, with an NMB of -20 to -50% and 176 an NME larger than 80%. Additionally, the correlation between the observed and simulated data was small, especially for S, 177 which showed no linear correlation. The observed wet deposition amount was higher in the wet season, but the amount were relatively well captured at the four sites in Vietnam. Accordingly, the wet depositions of S, N, and A obtained by the 195 ENS can generally reproduce the observed data to an acceptable level. The results of the statistical analysis are shown in 196 Table 5. As can be seen from the table, as well as from Fig. 5, the statistical scores for Vietnam were better than those for the 197 other countries in continental Southeast Asia. The R value was around 0.5-0.6, while the NMB was around -35% for the wet 198 deposition of S and N and around +15% for the wet deposition of A. The NME was around +50%, which was smaller than 199  Table 6. For the wet deposition of S, R was 0.79 and NMB and NME were +11.4% and +58.0%, respectively. The ENS 208 captured the wet deposition of S adequately. However, the NME values were worse for the wet deposition of N and A. For 209 example, the NME for the wet deposition of A was greater than +100%. In particular, as shown in Fig. 6 Table 7. There was a moderate correlation between the observations and simulations for 225 the wet deposition of S and N, and the NMB and NME were highest for the wet deposition of S. It should be noted that the 226 wet deposition of A showed much higher NMB and NME values and a lower value of R; this is due to the fact that the wet 227 deposition of A was overestimated at all four sites in Malaysia (Fig. 7) Island; in August in Kototabang (No. 50), which is located on Sumatra Island; and in January and February in Maros (No. 235 54), which is located on Sulawesi Island. The observed wet deposition of S, N, and A in these limited dry seasons was 236 generally lower than during the wet season; however, no difference in the simulated wet deposition of S, N, and A was 237 observed between the wet and dry season. The reason for this failure was that the model did not show the reduced 238 precipitation amounts seen in observations during these dry seasons. The results of the statistical analysis are listed in Table  239 8. A moderate correlation between observations and simulations was found for the wet deposition of S, N, and A, but the 240 ENS overestimated the wet deposition of S, N, and A, especially for S, with an NMB of +65.6% and an NME larger than 241 scales. The precipitation-adjusted approach using EANET observational data is hereafter called AO (adjusted by observation 260 at EANET site). 261 The precipitation-adjusted approach was shown to be effective for improving the modeling reproducibility in MICS-Asia III 262 (Itahashi et al., 2020). However, this approach has a limitation in that the adjusted wet deposition was obtained only at 263 locations corresponding to EANET observation sites, and hence the adjusted wet deposition was spatially limited. To 264 overcome this limitation, in this study, we additionally used a satellite dataset; this precipitation-adjusted approach is

Improvements of wet deposition modeling through a precipitation-adjusted approach for each country in 287
Southeast Asia 288

Myanmar 289
At the Yangon (No. 30) site in Myanmar, the wet deposition of S, N, and A was underestimated, with an NMB exceeding -290 70%, as listed in Table 2. Table 2 also provides the results of the statistical analysis for the AO and AS approaches, 291 demonstrating that the underestimation in the ENS was improved by both approaches; most of the statistical scores were 292 improved compared with the ENS, though there was still underestimation compared with the observed wet deposition of S, 293 N, and A. Fig. 10 shows the annual accumulated wet deposition of S, N, and A from the observational data, ENS, AO, and 294 AS. As shown in the figure, the wet deposition was higher with the AO and AS approaches compared with ENS; that is, the 295 underestimation was partly improved. Fig. 10 also shows the fractions of wet deposition occurring during the dry and wet 296 seasons as bar graphs for the observational data, ENS, AO, and AS. It can be clearly seen that, for the wet deposition of S, N, 297 and A, the fraction during the dry season was overestimated with ENS but was well matched with the AO and AS 298 approaches. 299 300

Thailand 301
The wet depositions of S, N, and A was generally underestimated at the six sites in Thailand, as shown in Table 3. The 302 statistical scores for AO and AS are also provided in this table. For the R value and the NME, AO and AS obtained superior 303 values for Thailand compared with the ENS, showing a stronger correlation with the observational data. For AO and AS, the 304 although the fractions of wet deposition occurring during the dry and wet seasons were improved with the AO and AS 315 approaches, underestimation was worsened. 316 317

Cambodia 318
At Phnom Penh (No. 38) in Cambodia, there were some difficulties capturing the wet deposition of S, N, and A using the 319 ENS. As shown in Table 4, there was a low correlation between the observed values and the ENS for the wet deposition of S, 320 and an even lower correlation for the wet deposition of N and A. The NMB was around -70% and the NME was 70-80% for 321 the wet deposition of S, N, and A using the ENS. These deficiencies in the ENS were adequately improved using AO and 322 AS. For AO and AS, all statistical scores showed an improvement compared with the ENS.

Vietnam 329
For the four EANET sites in Vietnam, the statistical scores for the ENS were superior to those for other countries in 330 continental Southeast Asia. In most cases, for AO and AS, the scores were improved compared with the ENS for the wet 331 deposition of S, N, and A, as shown in Table 5.

Indonesia 359
In Indonesia, during the short dry season, wet deposition showed a steep decline; however, models did not show such a 360 dramatic decrease. As shown in Table 8, the statistical scores for AO and AS were mostly superior to those of the ENS; the 361 moderate correlation found for the wet depositions of S, N, and A in the ENS were improved by AO and AS. For the wet 362 deposition of S, the NMB of +65.6% and NME of +100.2% in the ENS were improved by AO and AS. Based on the analysis and statistical results of the precipitation-adjusted approaches using surface observations and satellite 371 measurements, it was found that these approaches improved the simulation of the wet deposition amount, as well as the 372 fraction of wet deposition occurring during the dry and wet seasons. Although there were still difficulties in some cases, the 373 precipitation adjustment was shown to be an effective way to improve the simulated wet deposition. One of the advantages 374 of the adjustment using satellite measurements is that it provides the spatial distribution of adjustment factors; hence, it is of S, N, and A are mapped. Both the ENS and AS simulated hot spots with high depositions of S, N, and A in regions such 377 as northern Vietnam, the southern Malay Peninsula, and Sumatra Island and Java Island in Indonesia. However, there were 378 clear differences between AS and ENS. These differences were similar for the wet depositions of S, N, and A. As shown in 379 by EANET. Generally, the ensemble model can capture the observed wet deposition; however, the models failed to capture 401 the wet deposition, even using the ensemble mean, obtaining low correlations and/or large biases and errors. Based on a 402 detailed analysis of the observed precipitation at each EANET observation site, it was found that this failure to capture the 403 wet deposition was related to the poor representation of the precipitation amount. In some cases, the model did not 404 adequately simulate the precipitation pattern during the dry and wet seasons. 405 To overcome this modeling difficulty for precipitation, in this study, two precipitation-adjusted approaches were applied 406 using EANET surface observations and TRMM satellite measurements, respectively. Both approaches have been shown to             Units are g S ha -1 month -1 for the wet deposition of S, and g N ha -1 month -1 for the wet depositions of N and A. Improvements in the 694 statistical score with AO and AS compared with ENS are highlighted in gray. Units are g S ha -1 month -1 for the wet deposition of S, and g N ha -1 month -1 for the wet depositions of N and A. Improvements in the 698 statistical score with AO and AS compared with ENS are highlighted in gray. Units are g S ha -1 month -1 for the wet deposition of S, and g N ha -1 month -1 for the wet depositions of N and A. Improvements in the 702 statistical score with AO and AS compared with ENS are highlighted in gray. Note: Units are g S ha -1 month -1 for the wet deposition of S, and g N ha -1 month -1 for the wet depositions of N and A. Improvements in 706 the statistical score with AO and AS compared with ENS are highlighted in gray.