Universidade Federal de Santa Maria

Ci. e Nat., Santa Maria v.42, Special Edition: Micrometeorologia, e17, 2020

DOI:10.5902/2179460X45709

ISSN 2179-460X

Received: 02/06/20  Accepted: 02/06/20  Published: 28/08/20

Special Edition

Evaluation of surface fluxes using the WRF model – a case study to the Bananal wetlands’ region

Avaliação dos Fluxos de Superfície a partir do Modelo WRF - Estudo de Caso para região da Ilha do Bananal

Rayonil Gomes Carneiro ^I

Diogo Nunes da Silva Ramos ^II

Letícia d’Agosto Miguel Fonseca ^III

Camilla Kassar Borges ^IV

Cleber Assis dos Santos ^V

Gilberto Fisch ^VI

Laura De Simone Borma ^VII

^I Instituto Nacional de Pesquisas Espaciais, São José dos Campos, Brazil. E-mail: rayonilcarneiro@gmail.com.

^II Instituto Nacional de Pesquisas Espaciais, São José dos Campos, Brazil. E-mail: diogonsramos@gmail.com.

^III Instituto Nacional de Pesquisas Espaciais, São José dos Campos, Brazil. E-mail: leticiafonseca.geo@gmail.com.

^IV Universidade Federal de Campina Grande, Campina Grande, Brazil. E-mail: camillakassar@gmail.com.

^V Instituto Nacional de Pesquisas Espaciais, São José dos Campos, Brazil. E-mail: cleberassis.ufpa@gmail.com.

^VI Universidade de Taubaté, Taubaté, Brazil. E-mail: gilberto.fisch@unitau.br.

^VII Instituto Nacional de Pesquisas Espaciais, São José dos Campos, Brazil. E-mail: laura.borma@inpe.br.

ABSTRACT

The present work aimed to analyze the simulations of surface fluxes of sensible and latent heat, and global radiation using the mesoscale atmospheric model (WRF) for the Bananal Island (Tocantins state, Brazil) region during three distinct seasonal periods (flooded, dry, and wet) in 2004. The final analysis of the NCEP global model was used as initial and boundary conditions of the WRF, which horizontal resolution (5 km) and physical parameterizations follow the operational settings used at CPTEC/INPE. The global radiation, the simulated sensible and latent heat fluxes were consistent with the observed data for the daily cycle, where the R² was higher than 0.8, showing a good correlation between the data. However, the WRF outputs overestimates/underestimates follow a distinct seasonal pattern between global radiation and heat fluxes. There are some hypotheses for this result, such as potential limitations of the model in describing the surface conditions, whether static or dynamic. Future studies may investigate how sensitive the WRF would be when updating surface conditions for scenarios closer to reality, especially the flooded surface situation.

Keywords: Amazonian Forest; Turbulent surface heat flux; Wetlands.

RESUMO

O presente trabalho teve como objetivo analisar as simulações dos fluxos à superfície de calor sensível e latente, e radiação global usando o modelo atmosférico de mesoescala (WRF) para a região da Ilha do Bananal (Tocantins, Brasil) durante três períodos sazonais distintos (alagado, seco e chuvoso) para o ano de 2004. As análises finais do modelo global do NCEP foram usadas como condições iniciais e de fronteira do WRF, cuja resolução horizontal (5 km) e parametrizações físicas seguem as configurações usadas operacionalmente no CPTEC/INPE. A radiação global, os fluxos de calor sensível e latente simulados se mostraram consistentes com os dados observados para o ciclo diário, em que o R² foi superior a 0,8, denotando boa correlação entre os dados. No entanto, as superestimativas/subestimativas do WRF seguem um padrão sazonal distinto entre a radiação global e os fluxos de calor. Há algumas hipóteses para este resultado, como potenciais limitações na descrição das condições de superfície do modelo, sejam elas estáticas ou dinâmicas. Estudos complementares poderão investigar qual seria a sensibilidade do WRF ao atualizar as condições superficiais para cenários mais próximos da realidade, principalmente a situação de superfície alagada.

Palavras-chave: Floresta Amazônica; Fluxo de calor sensível; Áreas úmidas.

1 Introduction

Wetlands are ecosystems at the interface between aquatic and terrestrial environments, considered global hotspots of biological diversity, ecosystem productivity, and economic activity (JUNK et al., 2018). The humid areas covered by forests located on the banks of large rivers, such as Amazonas and Araguaia, are known as seasonally flooded forests (JUNK et al., 2018). The composition of the plant communities in these forests is mainly determined by the need for adaptation during the flooded phase, also called the aquatic phase (PAROLIN et al., 2010).

Understanding the physical processes of the energy balance that occur between the surface and the atmosphere is of fundamental importance for a correct description of atmospheric phenomena. Since, turbulent surface fluxes serve as sinks or sources of energy, humidity, momentum, and air pollutants that significantly impact the formation and evolution of clouds, consequently the precipitation and dispersion of air pollutants. Surface fluxes are crucial parameters to simulate turbulent mixing, growth of the convective boundary layer, and atmospheric transport (SUN et al., 2017). Thus, numerical atmospheric models, especially those mesoscale scales, which represent the interactions of the Earth's surface to the atmosphere, are increasingly important, particularly in tropical forest regions.

However, these models have limitations regarding the description of surface processes, particularly in flooded areas. The surface parameters are generally defined as static, varying according to the geographic database and with the physical parameterization of the terrestrial surface adopted. Lin et al. (2018) presented a methodology of dynamic meteorological downscaling for regional hydrological simulations, seeking to understand how floods can be predicted during extreme precipitation events. According to the authors, one of the main challenges is the calibration of land surface models, as well as, in the correct description of the vegetation cover of the flooded site.

Therefore, the present study objective is to verify the simulations of surface fluxes of sensible heat, latent heat, and the global radiation balance using the WRF model for the Bananal Island region during three distinct seasonal periods (flooded, dry and wet).

2 Material and Methods

2.1 Study area

The study was conducted in seasonally flooded forest Cantão State Park (PEC), a protection unit located 260 km west of Palmas, Tocantins, Brazil (Figure 1). The region is in the transition between the biomes of the Amazon and the Cerrado (savanna), delimited in the southwest by the region of Bananal Island, which is the largest river island in the world (BORMA et al., 2009). According to the Koppen classification, the region's climate is tropical humid-sub-humid. The annual precipitation ranges between 1,300 and 1,900 mm and temperature ranges from 22 °C in January to 31 °C in September (BORMA et al., 2009; FONSECA et al., 2019). The dry season occurs between May and September and the wet season between October and April, concentrating approximately 90% of the annual precipitation. From February until May, the region is affected by annual floods, which according to the discretion of Junk et al. (2018) are of the monomodal type, long duration (5 months) and low amplitude. During flooding, the water level in the region rises from 1 to 5 m above the land surface (BORMA et al., 2009).

Figure 1 – Location of the study area of the Cantão State Park (PEC) and position of the micrometeorological tower

A micrometeorological tower was installed in the region (9º 49’ 27.9” S, 50º 08’ 98.8” W, 120 m) that belongs to the LBA program (Large Scale Biosphere-Atmosphere Experiment in Amazonia) located about 2 km east of the Javaezinho River, a tributary of the Javaes River (Figure 1). This study used data from the eddy covariance (EC) system composed of a three-dimensional sonic anemometer (CSAT3A Campbell Scientific, United States), together with a gas analyzer (Li-7500, LI-COR Environmental, United States) providing the averages to every 30 minutes of turbulent surface fluxes (sensible and latent heat), as well as measurements of global solar radiation through a Pyranometer (CMP3, Kipp & Zonen, Netherlands).

For the elimination of noise (spikes) and interferences in the measurements of high frequency sensors, caused by the influence of humidity, among other factors, it is necessary to use a filter (BI et al., 2007; BEZIAT et al., 2009; DIAZ and ROBERTI, 2014), described below:

	(1)
	(2)

where  represents the measurements,  is the mean over the interval and  is the standard deviation.

2.2 WRF model setup

The simulations of the energy balance components in the studied area were obtained using the atmospheric mesoscale model WRF, in its version 3.9.1.1 (WANG et al., 2016). Main WRF configurations used are summarized in Table 1, while the version and the physical parameters adopted in this work follow the operational definitions of the numerical weather forecasts of CPTEC/INPE (http://previsaonumerica.cptec.inpe.br/wrf).

Table 1 – Summary of the main configurations selected in the simulations with the WRF model

Configuration	Description
Initial and boundary condition	FNL/GFS/NCEP: 1º x 1º e 6 h
Static conditions	Topography: GMTED 30s; Use of the soil: MODIS+Lakes 30s
NX, NY, NZ	551, 501, 35
Physical parameterizations	Microphysics: Ferrier; Shortwave and Longwave Radiation: RRTMG; Surface layer: MM5-Jimenez; Turbulence: YSU; Land surface: NOAH; Convection: New Tiedke

The WRF integration was defined with 6 + 24 hours, starting at 00 UTC from the 4th to the 9th of March (flooded period), August (dry period) and December (wet period) 2004. The domain's horizontal resolution was 5 km, centered on the position of the micrometeorological tower.

Figure 1 shows that the WRF domain area has different topographic and vegetation cover characteristics. However, the region around the tower indicates a homogeneity of the surface conditions described by the model, with flat topography and vegetation dominated by forests and swampy areas. The static information of the WRF is consistent with the data provided by the programs: Brasil em Relevo (https://www.cnpm.embrapa.br/projetos/relevobr) and SisCob (http://www.cnpdia.embrapa.br/downloads/siscob/) of EMBRAPA.

2.3 Model Assessment Methods

According to Oliveira and Souza (2017) among others, the best way to evaluate numerical models is by using a set of statistical indexes (LIMA et al., 2012). To evaluate the performance of the model´s output in comparison to the observations, some statistical indices were calculated: the BIAS (Equation 3), which is more simple but very clearly defines the systematic error (overestimation or underestimation) of the model; and the Mean Square Root Error (RMSE) (Equation 4), which measures the sensitivity to large differences between the compared series.

	(3)
	(4)

where  is the number of data,  is the simulated value and  is the observed value.

Consequently, the determination coefficients (R²) will also be calculated, which shows the measure of adjustment of the model in relation to the observed values.

3 Results and Discussion

Hourly averages for the six-day simulation for each of the three cases studied (flooded, dry and wet) of the global solar radiation (Rg) and the turbulent surface fluxes (sensible – H and latent – LE), simulated from WRF were compared with the averages of the observations made at the micrometeorological tower.

Analyzing the average of Rg (Figure 2), it can be noticed that the maximum values for the flooded period of the region were overestimated (Figure 5A) by the model presenting a BIAS (28.5 W m⁻²) and RMSE (68.4 W m⁻²) (Table 2). Mainly during the afternoon, since the maximum observed Rg (Figure 2A) was 650.5 W m⁻², while the WRF model (Figure 2B) had a maximum of 730.0 W m⁻². The simulation of the afternoon period did not show the reduction of Rg observed in the observational data. In the dry period the maximum values were 829.2 W m⁻² for the simulated and 809.3 W m⁻² for the observed, thus attesting to the high sensitivity of the WRF (BIAS equal to 4.7 W m⁻² and RMSE equal to 16.3 W m⁻²) throughout the daily cycle. Lastly in the wet season the maximum results were similar, with an average of 782.9 W m⁻² (observed) and 794.5 W m⁻² (WRF). However, the Rg values in the wet period, as well as in the flooded period, did not show a reduction after the maximum in the afternoon, since the BIAS was 21.2 W m⁻² and RMSE was 43.4 W m⁻². This is due to the underestimation of cloud cover by the model (WEHBE et al., 2019).

Figure 2 – Average time evolution of global solar radiation (Rg) (W m⁻²), observed values (A), and simulated (B) during the three analyzed cases

The simulated H flux overestimated in comparison with to the observational data (Figure 3A) in all 3 periods considered. This can be seen in the flooded and wet periods: BIAS was around 9 W m⁻² and the RMSE around 30 W m⁻². For the flooded period, the maximums values were equal to 135.3 W m⁻² (observed) and 171.5 W m⁻² (simulated), and in the wet season, these values were equal to 183.2 W m⁻² (observed) and 225.1 W m⁻² (simulated). While the dry period occurred greater overestimation over the daily cycle (BIAS approximately 17.2 W m⁻² and RMSE 35.0 W m⁻²), and maximum equal to 278.8 W m⁻² (observed) and 338.1 W m⁻² (simulated). The H flux at the simulated surface for all periods presented a greater magnitude in comparison to the observations, results similar to those found by Sun et al. (2017), Hariprasad et al. (2016) for other locations.

Figure 3 – Average temporal evolution of the sensible heat flux (H) (W m⁻²), observed values (A) and simulated (B) during the three analyzed cases

The LE flux estimated via the WRF model (Figure 4B) in the flooded period showed satisfactory results in relation to the observed data (Figure 4A), with a small overestimated (BIAS approximately 2.2 W m⁻² and RMSE of 15.0 W m⁻²), the maximum values being 450.0 W m⁻² (observed) and 503.2 W m⁻² (simulated). For the dry period, there was a remarkable underestimation of the simulations (BIAS approximately -23.7 W m⁻² and RMSE 52.3 W m⁻²), with the maximums equal to 381.1 W m⁻² (observed) and 241.3 W m⁻² (simulated). However, in the wet season the simulated LE was overestimated by about 17.2 W m⁻² (BIAS) and RMSE equal to 49.4 W m⁻², with the maximums equal to 472.0 W m⁻² (observed) and 467.0 W m⁻² (simulated).

Figure 4 – Average temporal evolution of the latent heat flux (LE) (W m−2), observed values (A) and simulated (B) during the three analyzed cases

According to Smallman, Moncrieff and Williams (2013), the incompatibility between the simulated and observed data for H and LE are related to underestimations of the observational data resulting from the non-closing of the energy balance at the surface.

Table 2 – Performance statistics: BIAS (W m⁻²), RMSE (W m⁻²) and R² for the three study periods

Period\Variable	Rg			H			LE
Flooded	0.80	28.5	68.4	0.82	9.3	32.2	0.85	2.2	15.0
Dry	0.95	4.7	16.3	0.95	17.2	35.0	0.87	-23.7	52.3
Wet	0.85	21.1	43.4	0.83	8.9	31.1	0.86	17.2	49.4

In general, the numerical simulations of the studied variables proved to be satisfactory for the different periods analyzed, since they exhibited a strong correlation denoted by R² values greater than 0.80 (Table 2). However, the scatter plots (Figure 5) showed an overestimation of Rg and H (Figure 5A and 5B) in all periods, especially in the flooded and wet periods. Meanwhile, the LE (Figure 5C) in the dry period was underestimated and in the other periods there was an overestimation. These results happened since there are some limitations of the numerical weather forecasting models associated with the description of the earth's surface that must be considered, since the land cover information is commonly static, varying only according to the database used (i.e. MODIS, USGS, CORINE, etc.). In addition, models such as WRF make use of a data table with surface characteristics described (i.e. albedo, roughness length, soil moisture, etc.) considering only the summer and winter seasons (WANG et al., 2016). Although this information has little seasonal variation, any inconsistencies in the type of vegetation and in the bio-geophysical and aerodynamic parameters of the surface can increase the uncertainties in the simulations.

Figure 5 – Scatterplots of WRF in relation to observational data from global solar radiation (Rg) (A), sensible heat flux (H) (B), latent heat flux (LE) (C)

4 Conclusions (of Final Remarks)

The simulations of turbulent heat fluxes were satisfactory for these periods, describing the seasonal pattern similar to observational data. However, the model overestimated the values of Rg, H fluxes in all periods and LE in the wet and flooded periods, whereas LE was underestimating in the dry periods.

As this study addresses a region with seasonal flooding and a horizontally homogeneous surface, those limitations of the WRF model need to be investigated in future studies. There are a few possibilities, such as applying a newer land surface model physics, as NOAH Multi-Parameterization (NOAH-MP), can improve those results. Investigate different land use or the use of newer vegetation cover data are other suggestions that deserve to be analyzed.

Acknowledgments

We thank Project GoAmazon/Ecophysiological controls on Amazonian precipitation seasonality and variability funded by the São Paulo Research Foundation–FAPESP, Brazil, grant 2013/50531-2, and the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior-Brasil (CAPES)-Finance Code 001 (process number 88881.188563/2018-01).

References

BEZIAT, P.; CESCHIA, E.; DEDIEU, G. Carbon balance of a three crop succession over two cropland sites in South West France. Agricultural and Forest Meteorology, v. 149, p. 1628–1645, https://doi.org/10.1016/j.agrformet.2009.05.004, 2009.

BI, X.; GAO, Z.; DENG, X.; WU, D.; LIANG, J.; ZHANG, H.; SPARROW, M.; DU, J.; LI, F.; TAN, H. Seasonal and diurnal variations in moisture, heat, and CO2 fluxes over grassland in the tropical monsoon region of southern China. Journal of Geophysical Research Atmospheres, v. 112, p. 1–14, https://doi.org/10.1029/2006JD007889, 2007.

BORMA, L. S.; ROCHA, H. R. D.; CABRAL, O. M.; RANDOW, C. V.; COLLICCHIO, E.; KURZATKOWSKI, D.; BRUGGER, P.; FREITAS, H.; TANNUS, R.; OLIVEIRA, L.; RENNÓ, C. D.; ARTAXO, P. Atmosphere and hydrological controls of the evapotranspiration over a floodplain forest in the Bananal Island region, Amazonia. Remote Sensing, v. 114, p. 1–12, https://doi.org/10.1029/2007JG000641, 2009.

DIAZ, M.; ROBERTI, D. R. Cálculo das trocas de carbono num agroecossistema de arroz com aplicação de técnicas de preenchimento de falhas. Ciência e Natura, v. 36, p. 27–31, https://doi.org/10.5902/2179460X16211, 2014.

FONSECA, L. D. M.; DALAGNOL, R.; MALHI, Y.; RIFAI, S. W.; COSTA, G. B.; SILVA, T. S. F.; ROCHA, H. R. D.; TAVARES, I. B.; BORMA, L. S. Phenology and Seasonal Ecosystem Productivity in na Amazonian Floodplain Forest. Remote Sensing, v. 11, p. 1–17, https://doi.org/10.3390/rs11131530, 2019.

HARIPRASAD, K.; SRINIVAS, C. V.; NAIDU, C. V.; BASKARAN, R.; VENKATRAMAN, B. Assessment of surface layer parameterizations in ARW using micro-meteorological observations from a tropical station. Meteorological Applications, v. 23, p. 191–208, https://doi.org/10.1002/met.1545, 2016.

JUNK, W. J.; PIEDADE, M. T. F.; CUNHA, C. N. da; WITTMANN, F.; SCHÖNGART, J. Macrohabitat studies in large Brazilian floodplains to support sustainable development in the face of climate change. Ecohydrology and Hydrobiology, v. 18, p. 334–344, https://doi.org/10.1016/j.ecohyd.2018.11.007, 2018.

JUNK, W. J.; PIEDADE, M. T. F.; LOURIVAL, R.; WITTMANN, F.; KANDUS, P.; LACERDA, L. D.; BOZELLI, R. L.; ESTEVES, F. A.; CUNHA, C. Nunes da; MALTCHIK, L.; SCHöNGART, J.; SCHAEFFERNOVELLI, Y.; AGOSTINHO, A. A. Brazilian wetlands: their definition, delineation, and classification for research, sustainable management, and protection. Aquatic Conservation: Marine and Freshwater Ecosystems, v. 24, p. 5–22, https://doi.org/10.1002/aqc.2386, 2013.

LIN, P.; YANG, Z.-L.; GOCHIS, D. J.; YU, W.; MAIDMENT, D. R.; SOMOS-VALENZUELA, M. A.; DAVID, C. H. Implementation of a vector-based river network routing scheme in the community WRF-HYDRO modeling framework for flood discharge simulation. Environmental Modelling and Software, v. 107, p. 1–11, https://doi.org/10.1016/j.envsoft.2018.05.018, 2018.

PAROLIN, P.; LUCAS, C.; PIEDADE, M. T. F.; WITTMANN, F. Flood-tolerant trees of Amazonian floodplains. Revista Pesquisas Botânica, v. 61, p. 7–38, 2010.

SMALLMAN, T. L.; MONCRIEFF, J. B.; WILLIAMS, M. WRFv3.2-SPAv2: development and validation of a coupled ecosystem–atmosphere model, scaling from surface fluxes of CO2 and energy to atmospheric profiles. Geoscientific Model Development, v. 6, p. 1079–1093, https://doi.org/10.5194/gmd-6-1079-2013, 2013.

SUN, X.; HOLMES, H. A.; OSIBANJO, O. O.; SUN, Y.; IVEY, C. E. Evaluation of Surface Fluxes in the WRF Model: Case Study for Farmland in Rolling Terrain. Atmosphere, v. 8, p. 1–23, https://doi.org/10.3390/atmos8100197, 2017.

WANG, W.; BRUYÈRE, C.; DUDA, M.; DUDHIA, J.; GILL, D.; KAVULICH, M.; KEENE, K.; LIN, H.-C.; MICHALAKES, J.; RIZVI, S.; ZHANG, X.; BERNER, J.; FOSSELL, K. WRF ARW Version 3 Modeling System User’s Guide. Boulder, Colorado (EUA), 2016. 408 p.

WEHBE, Y.; TEMIMI, M.; WESTON, M.; CHAOUCH, N.; BRANCH, O.; SCHWITALLA, T.; WULFMEYER, V.; ZHAN, X.; LIU, J.; MANDOUS, A. A. Analysis of an extreme weather event in a hyper-arid region using WRF-Hydro coupling, station, and satellite data. Natural Hazards and Earth System Sciences, v. 19, p. 1129–1149, https://doi.org/10.5194/nhess-19-1129-2019, 2019.