Zhio Go , Jingshn Zhu , Yn Guo , Xioong Yn , Xiujun Wng c , Huoqing Li ,Shuwen Li

a State Key Laboratory of Earth Surface Processes and Resource Ecology, Beijing Normal University, Beijing, China

b Key Laboratory of Cloud-Precipitation Physics and Severe Storms, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing, China

c Jilin Province Technology Center for Meteorological Disaster Prevention, Changchun, China

d Institute of Desert Meteorology, China Meteorological Administration, Urumqi, China

Keywords:Short-range ensemble forecast Land-surface parameter South china region

ABSTRACT In order to compare the sensitivity of short-range ensemble forecasts to different land-surface parameters in the South China region, three perturbation experiments related to the land surface model (LSM), initial soil moisture(ISM), and land—atmosphere coupling coefficient (LCC) were designed, and another control experiment driven by the Global Ensemble Forecast System (GEFS) was also performed. All ensemble members were initiated at 0000 UTC each day, and integrated for 24 h for a total of 40 days from the period 1 April to 10 May 2019 based on the Weather Research and Forecasting model. The results showed that the perturbation experiment of the LSM (LSMPE) had the largest ensemble spread, as well as the lowest ensemble-mean root-mean-square error among the three sets of land-surface perturbed experiments, which indicated that it could represent more uncertainty and less error. The ensemble spread of the perturbation experiment of the ISM (ISMPE) was generally less than that of LSMPE but greater than that of LCCPE (the perturbation experiment of the LCC). In particular,although the perturbation of the LCC could not produce greater spread, it had an effective influence on the intensity of precipitation. However, the ensemble spread of all the land-surface perturbed experiments was smaller than that of GEFSPE (the control experiment). Therefore, in future, land-surface perturbations and atmospheric perturbations should be combined in the design of ensemble forecasting systems to make the model represent more uncertainties.

1. Introduction

The land surface is an important component of the earth system and one of the key factors affecting short-range weather forecasts( Pielke, 2001 ; Trier et al., 2004 ). The accuracy of the representation of land-surface states in numerical weather prediction is critical to the skill of short-range forecasting ( Orth et al., 2016 ). Due to the complexity of land-surface parameters, ensemble forecasting is widely used to estimate the uncertainty and produce probabilistic information ( Deng et al.,2016 ; Duda et al., 2017 ). However, under-dispersion is currently a common problem for ensemble forecasting, which shows that ensembles cannot completely represent the uncertainties attached to the land surface( Hou et al., 2001 ). Therefore, it is necessary to conduct ensemble experiments on different land-surface parameters and compare their effects on short-range forecasting, which can help us identify the most influential parameters and expand the ensemble spread.

There are many land-surface parameters that can influence shortrange forecasts. For instance, numerous studies have revealed the importance of the initial soil moisture (ISM) in the skill of forecasts. Relatively wet and unevenly distributed soil moisture in the morning is conducive to the generation of afternoon convection ( Berg et al., 2013 ; Hsu et al.,2017 ). Zhu et al. (2018) used five soil moisture analyses as the ISM for their ensemble and found that perturbing the ISM produced notable ensemble spread at the lower level and a skillful ensemble mean for the precipitation forecast. In addition to soil moisture, uncertainties in representing the land—atmosphere coupling coefficient (LCC) can also impact the forecasting of convection systems ( Koster et al., 2004 ; Chen and Zhang, 2009 ). Zheng et al. (2014) simulated some heavy precipitation cases using the Weather Research and Forecasting (WRF) model and found that the LCC C(for further details see Section 2 ) could change the intensity of the simulated total precipitation. Chen et al. (2019) used 4-km WRF simulations to confirm that the Cparameter also has an important impact on regional climate. Besides, different land surface models (LSMs) have different representations of the land-surface condition,which also makes it necessary to change the LSM to generate the ensemble ( Zhu and Xue, 2016 ; Duda et al., 2017 ). The above three aspects of land-surface uncertainty are the focus in this paper.

Many previous studies only evaluated the uncertainty of a certain aspect of the land-surface process, such as the LSM, ISM, or LCC, while the objective of this research is to compare the sensitivity of short-range ensemble forecasts to these different aspects of uncertainties under the same framework. Two scientific issues are the main concern: (1) Among these land-surface factors, which factor’s perturbation can produce the greatest spread for short-range ensemble forecasts? (2) How does each land-surface perturbed member affect the forecast result? Analysis of these issues can help to improve the problem of under-dispersion of ensemble forecasting, and it is also helpful in improving the skill of forecasts. Previous studies have shown that pre-rainy-season rainfall in South China is sensitive to the lower atmospheric state, especially from April when vegetation is photosynthetically active ( Luo et al., 2017 ;Sun et al., 2019 ). Therefore, this paper focuses on the impact of landsurface perturbations on short-range ensemble forecasts in South China during the pre-rainy season.

2. Data and methods

Three perturbation experiments related to land-surface conditions were designed, including an ISM-perturbed experiment (ISMPE),LCC-perturbed experiment (LCCPE), and LSM-perturbed experiment(LSMPE). In addition, based on the Global Ensemble Forecast System(GEFS), an atmospheric perturbed experiment (GEFSPE) was designed for comparison with the three land-surface perturbation experiments.Each experiment had four members, including a common control member. All the above perturbations were carried out in the Advanced Research version of the WRF model, version 4.0 ( Skamarock, 2008 ). All ensemble forecasts were initiated at 0000 UTC each day for the period 1 April to 10 May 2019 and integrated for 24 h. The model ran in two nested domains with resolutions of 15 km and 3 km, respectively, with the inner domain covering the South China region ( Fig. 1 (a)).

Fig. 1. (a) The two nested model domains and terrain height (units: m). (b) Averaged volumetric soil moisture of the nested domain at 0—10 cm below ground of the four analyses over 40 days. The lines represent the medians, the boxes the 25th and 75th percentiles, and the whiskers the 5th and 95th percentiles.

The four ensemble experiments were performed as follows:

(1) In ISMPE, four soil moisture analysis datasets from four different research centers were used as the ISM to initiate the ensemble forecast. They were: the National Centers for Environmental Prediction Global Forecast System (GFS) reanalysis dataset( https://www.ncdc.noaa.gov ); the fifth major global reanalysis produced by the European centre for Medium-Range Weather Forecasts (ERA5; https://www.ecmwf.int ); the China Meteorological Administration’s Land Data Assimilation System (CLDAS;http://data.cma.cn ); and NASA’s Global Land Data Assimilation System (GLDAS) Noah LSM dataset ( https://ldas.gsfc.nasa.gov/gldas ).In these soil moisture datasets, the domain-averaged ISM from ERA5 is the wettest at 0—10 cm below ground level, while the ISM from GFS is the driest, indicating that the forecasting of soil moisture is quite different ( Fig. 1 (b)).

(2) In LCCPE, a key parameter ( C) was perturbed. In the Noah LSM,the relationship between C zil and roughness is expressed by

wherezisthemomentumroughnesslengthandzrepresentsthermalroughnesslength, κis theKarman constant, and Reis the Reynolds number. C zil is an empirical parameter and is set to 0.1 in the Noah LSM. The value of Ccan vary from 0.01 (strong coupling)to 1.0 (weak coupling), or depending on the vegetation (abbreviated as dep-veg). In this experiment, the three values used were 0.1, 0.05,and 0.5, and the value of the other member changed dynamically with the vegetation.

(3) In LSMPE, four LSMs (Noah, Rapid Update Cycle (RUC), Community Land Model (CLM), and the multi-parameterization version of Noah(Noah-MP)), were used to form the ensemble forecasts.

(4) In GEFSPE, the four ensemble members were driven by GFS and the first three GEFS members (gep01-gep03), respectively. The GEFS dataset uses the GFS model for integration, and its initial perturbations are selected from the Global Data Assimilation System 80-member Ensemble Kalman Filter 6 h forecast since December 2015( Wei et al., 2008 ; Zhou et al., 2017 ). It runs four times per day(0000, 0600, 1200, and 1800 UTC) and has 20 perturbed members ( https://www.ncdc.noaa.gov/data-access/model-data/modeldatasets/global-ensemble-forecast-system-gefs ). The specific information of the ensemble members of each experiment is given in Table 1 . In addition, for the other physical schemes, all members had the same settings. The Morrison microphysics scheme ( Morrison et al., 2009 ) was used for the microphysical processes, the RRTMG scheme ( Iacono et al., 2008 ) for the longwave and shortwave radiation, and the YSU scheme ( Hong et al., 2006 ) for the planetary boundary layer.

Table 1 Description of each ensemble member in the four experiments.

Fig. 2. Profiles of the ensemble spread of (a) U -wind, (b) V -wind, (c) specific humidity, and (d) potential temperature in the four experiments at 6 h, averaged from 1 April 2019 to 10 May 2019.

The CPC morphing technique grid precipitation analysis dataset was used as the verification data of precipitation ( http://data.cma.cn ). This dataset is based on hourly precipitation observed by automatic weather stations in China and satellite retrieval data ( Shen et al., 2013 ). It is archived at a resolution of 0.05°×0.05° and in 1 h intervals. The near-surface variables were validated by the CLDAS dataset, which is archived at a resolution of 0.0625°×0.0625° in 1 h intervals. The ensemble spread was calculated by

Fig. 3. RMSE of the ensemble mean (dashed line) and ensemble spread (lines) of the four experiments for (a) 2 m temperature and (b) 10 m wind speed. The 40-day averaged (c) sensible heat flux and (d) latent heat flux of all the ensemble members related to land-surface conditions.

3. Results

3.1. Profiles of ensemble spreads

The ensemble spreads of zonal wind ( U -wind), meridional wind ( V -wind), specific humidity, and potential temperature, spatially averaged over the nested domain and temporally averaged over the 40 days, were verified for the four experiments form 1000 hPa to 200 hPa ( Fig. 2 ). Although the three land-surface perturbed experiments had less ensemble spread, about a third to a half of that of GEFSPE, their effects were also non-negligible. Among the three experiments, LSMPE had the largest ensemble spread and LCCPE the smallest, from 1000 to 200 hPa. Except for the U -wind variable, the ensemble spread of other variables at the lower layer was greater than that at the upper layer, which shows that the impact of land-surface perturbations is more obvious at the lower layer.

3.2. Near-surface variables

The ensemble spread and the ensemble-mean root-mean-square error(RMSE) of the four experiments for 2-m temperature (T2) and 10-m wind speed (Wind10) were calculated ( Fig. 3 (a, b)). Previous studies have proven that the ensemble spread and ensemble-mean RMSE of a perfect ensemble forecast should mostly be equal ( Hou et al., 2001 ). For the ensemble spread, GEFSPE was still the largest, followed by LSMPE.The values of LCCPE and ISMPE were relatively small. Most of the time,the spread of ISMPE for Wind10 was slightly greater than that of LCCPE,while it was the opposite for the T2 field. The larger spread of GEFSPE and LSMPE indicates that perturbations in the LSM and atmospheric conditions can represent more uncertainties for near-surface variables.

Fig. 4. (a) RMSE of the ensemble mean (dashed line) and ensemble spread(lines) of the four experiments for 24 h precipitation. (b) Ratio of domainaveraged 24 h precipitation under different Czil values on all 40 days.

As for the ensemble-mean RMSE, LCCPE had the largest value and the other three experiments showed small differences in their forecasting of T2. However, for Wind10, LSMPE was the smallest, followed by GEFSPE, ISMPE, and LCCPE. The smaller ensemble-mean RMSE of LSMPE and GEFSPE implies that smaller errors were included in their experiments. In addition, relatively speaking, LCCPE could neither represent more uncertainty nor reduce the error, which indicates that its effect on improving the forecasting of near-surface variables is very limited.

The sensible heat flux and latent heat flux were averaged over the inner domain and over 40 days for nine members associated with landsurface perturbations ( Fig. 3 (c, d)). It can be seen that increasing (decreasing) the coupling strength increased (decreased) both the sensible heat and the latent heat. Among the LSMs, Noah has the highest sensible heat release and the lowest latent heat release, while RUC and CLM will release more latent heat. In addition, the soil of ERA5 will release more latent heat and less sensible heat, which may be related to its relatively wet state.

3.3. Precipitation

Fig. 4 (a) shows the ensemble spread and ensemble-mean RMSE of the four experiments for 24 h total precipitation on all 40 days. GEFSPE also had the largest spread and the smallest RMSE, which indicates it can represent more uncertainty and less error. Among the land-surface perturbation experiments, their RMSEs were relatively close but LSMPE had the largest ensemble spread, which is consistent with the distribution of near-surface variables. The spread of LCCPE was the smallest,but it should be mentioned that C zil showed a capability to influence the overall intensity of precipitation in a large number of cases in this study. Fig. 4 (b) shows the ratio of domain-averaged 24 h precipitation under different C zil values on all 40 days. It can be seen that reducing the coupling strength reduced the total precipitation by approximately 8%, but the effect of increasing the coupling strength was not so obvious; it could only increase the precipitation by 2%, which shows that the impact of coupling strength on precipitation is nonlinear.

The Fractional Skill Scores (FSSs) of 24 h precipitation exceeding 1,10, 25, and 50 mm over the nested domain for all members on 40 days are sorted in Fig. 5 . The FSS ranges from 0 to 1, with a higher score indicating a better forecasting skill ( Ebert, 2008 ). The scores of all the members decreased as the threshold increased, which indicated that the forecasting skill of the high threshold precipitation was still low. Among the LSMs, Noah and Noah-MP (member m0 and m3, respectively) performed better overall. The moderate coupling strength ( C zil = 0 . 1 ) and GFS ISM were generally better for the 1, 10, and 25 mm rainfall thresholds. For the 50-mm threshold, changes in the LCC and ISM improved the forecasting results to a certain extent. For GEFSPE, the results of the control member were also optimal overall, except for the 25-mm threshold. The above results indicate that perturbation of the land surface or atmospheric conditions alone offers limited improvement in terms of the skill of precipitation forecasting, and thus more perturbation methods need to be studied in the future.

Fig. 5. FSS scores of all members for thresholds of (a) 1 mm, (b) 10 mm, (c) 25 mm and (d) 50 mm sorted on the 40 days. The lines represent the medians; the boxes the 25th and 75th percentiles. The abscissa represents the index of each member in Table 1 .

4. Summary

In this study, based on the WRF-ARW model, three sets of perturbation experiments related to land-surface conditions, along with a control experiment related to atmospheric perturbation, were designed for the 40-day daily short-range forecasts in South China from 1 April to 10 May 2019, so as to explore the sensitivity of short-range ensemble forecasts to different land-surface parameters and provide guidance for the design of future ensemble forecasts. Among the three land-surface perturbation experiments, LSMPE had the largest ensemble spread from 1000 to 200 hPa, as well as in the near-surface variables and precipitation. It had the smaller ensemble-mean RMSE for the near-surface variables, which indicates that perturbations of the LSM is an effective method for improving the uncertainty and reducing the error. The ensemble spread of ISMPE was mostly higher than that of LCCPE, and the error of LCCPE in the near-surface variables was relatively large, which shows that perturbing the coupling strength alone has a limited effect on expanding the spread and reducing the error. The Noah and Noah-MP LSMs, combined with moderate coupling strength ( C zil = 0 . 1 ) and the ISM of GFS, produced higher forecasting scores for precipitation, but changing the ISM data and coupling strength slightly improved the prediction skill for heavy rainstorms. In addition, weakening the coupling strength ( C= 0 . 5 ) was also effective at reducing the regional-average precipitation.

However, the ensemble spread of GEFSPE was the largest in all the experiments, including the near-surface variables. This indicates that perturbation of land-surface parameters alone is not enough to improve the ensemble spread. Therefore, in future research, atmospheric perturbations and land-surface perturbations need to be combined to make the model represent more uncertainties, and it is also necessary to evaluate its forecasting skills.

Declaration of Competing Interest

No potential conflict of interest was reported by the authors.

Funding

This work was supported by the National Key R&D Program on the Monitoring, Early Warning and Prevention of Major Natural Disasters[grant number 2017YFC1502103], the Key Special Project for the Introducing Talents Team of the Southern Marine Science and Engineering Guangdong Laboratory (Guangzhou) [grant number GML2019ZD0601 ],and the National Natural Science Foundation of China [grant numbers 41875136 , 41305099 , and 41801019 ].