Produtividade Primaria Bruta do Cerrado Brasileiro Estimada por Modelos de Sensoriamento Remoto

Agricultural and Forest Meteorology 307 2021 108456 Available online 27 May 2021 01681923 2021 The Authors Published by Elsevier BV This is an open access article under the CC BY license httpcreativecommonsorglicensesby40 Gross primary productivity of Brazilian Savanna Cerrado estimated by different remote sensingbased models Marcelo Sacardi Biudes a George Louis Vourlitis b Maísa Caldas Souza Velasque a Nadja Gomes Machado c Victor Hugo de Morais Danelichen d Vagner Marques Pavao a Paulo Henrique Zanella Arruda a Jose de Souza Nogueira a a Programa de PosGraduaçao em Física Ambiental Instituto de Física Universidade Federal de Mato Grosso Cuiaba Mato Grosso Brazil b Biological Sciences Department California State University San Marcos California United States c Instituto Federal de Mato Grosso Cuiaba Mato Grosso Brazil d Programa de PosGraduaçao em Ciˆencias Ambientais Universidade de Cuiaba Cuiaba Mato Grosso Brazil A R T I C L E I N F O Keywords Satellite products Carbon exchange Light use efficiency Vegetation index A B S T R A C T Gross primary production GPP is the total amount of fixed carbon and depends on vegetation health and water and energy availability GPP has been monitored worldwide by flux towers and models that are coupled with remotely sensed data such as by Moderate Resolution Imaging Spectroradiometer MODIS However models have not been evaluated for tropical savanna which presumably represent a challenge because of large spatial and seasonal variation in GPP Thus our goal was to evaluate the Vegetation Photosynthesis Model VPM Temperature and Greenness Model TG Vegetation Index Model VI and the MOD17A2 product of MODIS in a tropical mixed woodlandgrassland The Normalized Difference Vegetation Index NDVI Enhanced Vegetation Index EVI Land Surface Water Index LSWI Land Surface Temperature LST and Photosynthetically Active Radiation Fraction fPAR derived from the MODIS sensor were used as model inputs and integrated with groundbased micrometeorological variables GPP varied significantly between the wet and dry seasons and was positively correlated with seasonal variations in soil volumetric water content VSWC and precipitation and negatively correlated with the vapor pressure deficit VPD Satellite vegetation indices NDVI LSWI and to a lesser extent the EVI and derived quantities fPAR and LUE also exhibited similar correlations with VSWC and precipitation Thus there were strong positive correlations between the SVIs and GPP All of the models were able to simulate the seasonal variations in GPP however VPM had the best performance with the highest correlation and smallest errors TG VI and MOD17A2 models performed similarly except for the VI model based only on the EVI Given their ability to capture seasonal dynamics remotesensing based models such as those tested here will likely be an important tool for assessing how climate variability alters C cycling dynamics in these spatially and temporally heterogeneous landscapes 1 Introduction The Cerrado known as Brazilian Savannah is the secondlargest biome in Brazil with 24 coverage of the Brazilian territory Fearn side 2000 Vourlitis and da Rocha 2010 The climate has two welldefined seasons throughout the year a wet season from October to April and a dry season from May to September Vourlitis and da Rocha 2010 The dry season is characterized by greater evaporative demand and low soil water availability causing a stressful environment for woody species Rodrigues et al 2014 However most tree species in the Cerrado have resistance to water loss due to their thick barks branches and twisted trunks with rigid and leathery leaves Ribeiro and Walter 1998 Despite the risk for water stress the Cerrado of Mato Grosso has been replaced by agricultural and cattle pastures over the last 45 decades The resulting vegetation is open with shallowrooted grasses which have less biomass Giambelluca et al 2009 The conversion of native vegetation into areas of agriculture and livestock modify the dynamics of energy flux evapotranspiration and net carbon exchange Fearnside 2000 Biudes et al 2015 Besides the occurrence of water stress and Corresponding author at Biological Sciences Department California State University San Marcos California USA Email address georgevcsusmedu GL Vourlitis Contents lists available at ScienceDirect Agricultural and Forest Meteorology journal homepage wwwelseviercomlocateagrformet httpsdoiorg101016jagrformet2021108456 Received 18 May 2020 Received in revised form 7 April 2021 Accepted 2 May 2021 Agricultural and Forest Meteorology 307 2021 108456 2 high temperature during the dry season in Cerrado causes the stomata closure and consequently reduces transpiration and CO2 diffusion Rodrigues et al 2016 Gross primary production GPP is total C uptake from canopy photosynthesis and is a useful variable for assessing controls on C dy namics because it varies depending on soil water availability incident solar radiation temperature vegetation composition and nutrient availability Biudes et al 2014 Baldocchi et al 2018 GPP can be obtained indirectly using the eddy covariance technique in flux towers considering the difference between the net ecosystem exchange NEE and the ecosystem respiration Reco Falge et al 2002 Baldocchi et al 2018 However while extensive flux networks are operational Bal docchi et al 2018 they are spatially limited because of high cost and technically demanding equipment and infrastructure needs Burba 2013 Furthermore eddy covariance methods require relatively ho mogenous areas to fulfill measurement assumptions and resolve fluxes Griebel et al 2016 but often vegetation and terrain are spatially heterogeneous as in the case of the Brazilian Cerrado Souza et al 2014 The application of remote sensing coupled with eddy covariance measurements and models makes it possible to estimate GPP on a regional scale because in addition to being representative for a given location it is very viable for regions where there are no meteorological data Silva et al 2013 One of the main variables considered in GPP models is the Light Use Efficiency LUE which represents the efficiency of the plant in using absorbed solar energy for C assimilation Monteith 1972 Drolet et al 2008 Ma et al 2014 Some studies had related a positive relationship of LUE with spectral vegetation indices derived from satellites such as the Normalized Difference Vegetation Index NDVI and the Enhanced Vegetation Index EVI Drolet et al 2005 2008 Nakaji et al 2007 Cheng et al 2009 In turn several models have been proposed to couple satellitebased vegetation and water indexes with meteorological variables such as the Vegetation Photosynthesis Model VPM Xiao et al 2005a the Temperature and Greenness model TG Sims et al 2008 the Vegetation Index model VI Wu et al 2010 and MOD17A2 product of the Moderate Resolution Imaging Spectroradi ometer MODIS sensor Wu et al 2011 These models have been successfully used to estimate GPP over global scales in agricultural systems and in various ecosystems Xiao et al 2004 2005b Li et al 2007 Sims et al 2008 Wang et al 2010 Wu et al 2010 2011 Biudes et al 2014 Souza et al 2014 Ma et al 2014 Danelichen et al 2015 but their efficacy in spatially diverse ecosystems such as the Brazilian Cerrado has not been evaluated Brazilian cerrado represents a challenge to remote sensingbased models of GPP because of the high spatial variability in vegetation cover and leaf area Ratana et al 2005 and the high seasonal variation in ecosystem CO2 exchange Arruda et al 2016 Thus our objectives were to test the welldescribe GPP models including the Vegetation Photosynthesis Model VPM Temperature and Greenness Model TG Vegetation Index Model VI and the MOD172 product of MODIS in a mixed woodlandgrassland in the Cer rado of Mato Grosso Brazil These mixed woodlandgrasslands are likely particularly difficult to model because of large spatial variations in vegetation structure Vourlitis et al 20132014 and seasonal varia tions in phenology Dalmolin et al 2015 and water availability Rodrigues Rodrigues et al 2016 that affect tree Dalmagro et al 2016 Dalmolin et al 2018 and ecosystem Arruda et al 2016 gas exchange Thus we were interested to assess whether these models were able to adequately capture the temporal variations in GPP for this complex tropical ecosystem and to evaluate potential biases associated with model performance 2 Material and methods 21 Study area The study area is located at Fazenda Miranda Miranda Farm approximately 15 km from the city of Cuiaba 15435365 S 56041881 W in the southern part of the state of Mato Grosso Brazil Fig 1 Longterm 30 years average annual rainfall and temperature for both sites are 1420 mm and 265 C respectively with a dry season extending from MaySeptember Vourlitis and da Rocha 2010 The research area is on flat terrain at an elevation of 181 m above sea level Soils are rocky dystrophic redyellow latosols Radambrasil 1982 The vegetation in the study area is a mixture of trees and grasses referred to as campo sujo dominated by native and nonnative grasses and semideciduous tree species such as Curatella americana L and Diospyros hispida A DC Vourlitis et al 2013 The site is upland with low leaf area index 13 m2m2 scattered trees 533 treesha and over 60 of the ground surface covered by grasses and forbs Vourlitis et al 2013 2014 Vegetation height varies however the scattered trees reach and maximum height of about 5 m while the grasses and forms reach a maximum height of about 115 m Fire suppression on the farm has been active thus the research site had not burned in over 35 years 22 Micrometeorological measurements Micrometeorological measurements were conducted between March 2011 and December 2012 Photosynthetically active radiation PAR was measured by a quantum sensor LI190SB LICOR Lincoln NE EUA the temperature and relative humidity of the air were measured at the top of the tower by a thermohygrometer HMP45C Vaisala Inc Helsínquia Finlandia soil moisture was measured by a timedomain reflectometer CS616L50 Campbell Scientific Inc Logan UT EUA The net exchange of CO2 from the ecosystem NEE was measured using eddy covariance Arruda et al 2016 Eddy covariance sensors were installed at a height of 12 m above ground level and consisted of a threedimensional sonic anemometer CSAT3 Campbell Scientific Inc Logan UT EUA to measure the mean and fluctuations in wind speed and an open path infrared gas analyzer LI7500 LICOR Inc Lincoln NE EUA to measure the mean and fluctuations in CO2 concentration The infrared gas analyzer was installed 5 cm from the sonic anemom eter downwind to minimize the separation of the sensors Raw data were obtained and stored every 01 s in a datalogger CR1000 Campbell Scientific Inc Logan UT USA Details of the system are described by Arruda et al 2016 23 CO2 Flux calculation and data treatment Carbon dioxide and energy fluxes were obtained by calculating the covariance between the fluctuations in vertical wind speed and fluctu ations in virtual temperature H2O vapor or CO2 molar density following a coordinate rotation of the wind vectors McMillen 1988 and averaged over 30minute periods Eddy CO2 flux derived from the openpath gas analyzer was corrected for simultaneous fluctuations in heat and H2O vapor while eddy H2O vapor flux was corrected for fluctuations in heat flux Webb et al 1980 NEE data were screened for quality following guidelines established by Ameriflux and Anthoni et al 1999 Data were rejected when 1 eddy covariance sensors failed or were down because of calibration and system maintenance 2 warming flags were generated by the system software indicating measurement andor processing errors 3 spikes in sonic andor infrared gas analyzer data were excessive such as during heavy rainfall events 4 abrupt changes in wind speed caused nonstationary conditions and 5 eddy flux data were outside physically andor biologically meaningful ranges Arruda et al 2016 Gross primary production GPP was estimated by Eq 1 following methods described by Wohlfahrt et al 2005 GPP NEE Reco 1 where NEE is the daytime PAR 5 µmol photons m2 s1 net ecosystem CO2 exchange measured from eddy covariance and Reco is an MS Biudes et al Agricultural and Forest Meteorology 307 2021 108456 3 average rate of daytime ecosystem respiration Estimates of Reco and GPP were derived using a MichaelisMenton type function Ruimy et al 1995 Wohlfahrt et al 2005 by Eq 2 over 8day intervals to be consistent with MODIS data acquisition described below NEE LUE PAR FGPPsat LUE PAR FGPPsat Reco 2 where LUE is the apparent quantum yield µmol CO2 µmol photons1 PAR is the measured average 30minute average PPFD µmol photons m2 s1 FGPPsat is the lightsaturated rate of GPP µmol CO2 m2 s1 and Reco was estimated as the intercept of Eq 2 where PAR 0 µmol photons m2 s1 Estimates of Reco derived using these methods compare well to those estimated from nighttime data Falge et al 2001 and minimize problems associated with nighttime flux loss from low tur bulence and errors in objectively selecting a turbulence ie frictional velocity threshold that excludes data measured under inadequate tur bulence Wohlfahrt et al 2005 24 Satellite imagery We downloaded 8day composite land surface reflectance data MOD09A1 land surface temperature LST MOD11A2 and the fraction of PAR absorbed fAPAR by a canopyvegetation MOD15A2 from the EROS Data Active Archive Center EDC DAAC httpdaac ornlgovcgibinMODISGLBVIZ1Glbmodissubsetorderglobalco l5pl based on the geolocation information latitude and longitude of eddy covariance flux tower from March 2011 to December 2012 The MOD09A1 datasets include seven spectral bands at a spatial resolution of 500 m and are corrected for the effects of atmospheric gases aerosols and thin cirrus clouds Vermote and Kotchenova 2008 The MOD11A2 provides the daytime and nighttime land surface tem perature LST under clear sky conditions at a 1 km spatial resolution The MOD15A2 contains the fraction of PAR absorbed by the can opyvegetation fPAR at 1 km spatial resolution fPAR is expressed as a fraction of photosynthetically active radiation absorbed by the photo synthesis process Land surface reflectance LST and fPAR values were averaged for the nine pixels covering and surrounding the eddy flux tower and only pixels with highest quality assurance QA metrics were used 25 Vegetation indices Land surface reflectance values from blue ρblue red ρred near infrared ρnir and short wave infrared ρswir were used to calculate the Enhanced Vegetation Index EVI Eq 3 Huete et al 1997 Normalized Difference Vegetation Index NDVI Eq 4 and the Land Surface Water Index LSWI Eq 5 Xiao et al 2005a NDVI utilizes red ρred and nearinfrared ρnir while EVI utilizes ρred and ρnir bands and includes the blue band ρblue for atmospheric correction to account for residual atmospheric contamination eg aerosols variable soil and canopy background reflectance Huete et al 1997 The atmo spheric correction is important in the Cerrado particularly during the dry season when smoke from biomass burning injects large amounts of particulates into the atmosphere Santiago et al 2015 EVI 25 ρnir ρred ρnir 6ρblue 75ρred 1 3 NDVI ρnir ρred ρnir ρred 4 The short infrared spectral band ρswir is sensitive to vegetation water content and soil moisture and a combination of NIR and SWIR bands have been used to derive the LSWI Eq 5 LSWI ρnir ρswir ρnir ρswir 5 The SWIR absorption increases and SWIR reflectance decreases as leaf liquid or volumetric soil water content increases thus increasing LSWI Xaio et al 2005b 26 MODIS Gross primary production product MOD17A2 The MODIS Gross Primary Production product MOD17A2 is designed to provide regular measures of the growth of terrestrial vege tation based on the light use efficiency LUE using the MODIS Earth fPARLAI daily coverage The product is calculated according to the following Equation GPP LUEmax mTmin mVPD FPAR SWrad 045 6 where LUEmax is the maximum light use efficiency obtained from a look up table based on the type of vegetation mTmin and mVPD are scalers Fig 1 Location of Mato Grosso Brazil with the micrometeorological tower in a mixed grasslandwoodland campo sujo cerrado at the Fazenda Miranda MS Biudes et al Agricultural and Forest Meteorology 307 2021 108456 4 that vary between 0 and 1 to reduce εmax under low temperature and high vapor pressure deficit fPAR is the fraction of photosynthetically active radiation absorbed by vegetation and SWrad is shortwave radi ation of which 45 045 is photosynthetically active Restrepo Coupe et al 2015 27 Precipitation measurement Precipitation data millimeters per month for the study area was obtained from the conventional weather station located at Padre Ricardo Remetter World Meteorological Organization Code 83364 which is 10 km from the flux tower at Miranda Farm Data from March 2011 to December 2012 were used for the study area 28 Vegetation photosynthesis model VPM The VPM proposed by Xiao et al 2004 assumes that the leaves and canopy are composed of photosynthetically active vegetation and nonphotosynthetically active vegetation The gross primary production predicted by the model is described as GPP LUE fPAR PAR 7 where PAR is photosynthetically active radiation mol m2 day1 fPAR is the fraction of photosynthetically active radiation absorbed by photosynthetically active vegetation which is assumed as a linear function of EVI in this study the fPAR MOD15 product provided by the MODIS sensor was used and LUE is the efficiency of light use gCmol PAR To execute the VPM model the input parameters must also be determined mainly LUE which is an important variable and difficult to obtain According to the VPM LUE LUEmax Tscalar Pscalar Wscalar 8 where LUEmax is the maximum efficiency of the light use and Tscalar Pscalar and Wscalar are scalars that vary between 0 and 1 for temperature Tscalar phenology Pscalar and water availability Wscalar so that there is maximum efficiency in the light use for vegetation Tscalar T TminT Tmax T TminT Tmax T Topt 2 9 Pscalar 1 LSWI 2 10 Wscalar 1 LSWI 1 LSWImax 11 where T is the air temperature at 12 m in each time interval C and Tmin Tmax and Topt are the minimum the maximum and the optimum air temperature for photosynthetic activities respectively If the air tem perature is lower then Tmin or higher than Tmax the Tscalar will be set to zero and if the air temperature is close to Topt the Tscalar will be approximately 1 Considering the vegetation type and the predominant climate the Tmin Topt and Tmax were set to 10 C 28 C and 48 C respectively Jin et al 2013 LSWImax is the maximum land surface water index value within the plant growth period and Pscalar is set to 1 when it comes to forests that have green cover all year round like semideciduous trees because green foliage is retained over several growing seasons Xiao et al 2005b 29 Temperature and greenness model TG The TG model was developed by Sims et al 2008 and based on EVI Enhanced Vegetation Index and in LST Land Surface Temperature both derived from MODIS products Eq 12 The combination of EVI and LST improves the correlation between the predicted GPP and the GPP obtained in flux towers Wu et al 2011 GPP m LSTesc EVIesc 12 where m is a scalar molC m2 day1 parameterized in this study as suggested by Sims et al 2008 and LSTesc and EVIesc are the scalar functions of LST Eq 13 and EVI Eq 14 LSTesc min LST 30 25 005 x LST 13 EVIesc EVI 01 14 where LSTesc is defined as the minimum of two linear equations The results described by Sims et al 2008 indicate that the relationship between GPP and LST for wet regions does not present an ideal estimate and for drought regions the results are accurate at about 30 C with GPP declining to zero when LST is close to 0 C or increasing when the temperature is close to 50 C This results in a maximum value of LSTesc LSTesc 1 when LST 30 C and minimum values LSTesc 1 when LST 0 C or LST 50 C Sims et al 2008 Wu et al 2010 Previous studies by Sims et al 2008 also show that GPP drops to zero when the EVI is around 01 210 Vegetation index model VI The VI model GPPVI was proposed by Wu et al 2010 and calcu lates GPP by multiplying PAR and two vegetation indices VI Eq 15 This model has been proposed because vegetation indices represent both Light Use Efficiency LUE and Fraction of Photosynthetically Active Radiation fPAR which have the same biophysical characteristics Giltelson et al 2015 Inoue et al 2008 Sims et al 2008 Wu et al 2010 GPP PAR VI VI 15 In this study to evaluate the individual and combined response of the NDVI and EVI in the estimation of the GPPVI the following equations were used GPP PAR NDVI NDVI 16 GPP PAR EVI EVI 17 GPP PAR NDVI EVI 18 211 Data analysis Monthly seasonal and annual average with 95 confidence in terval of satellitederived variable and micrometeorological variables were calculated by bootstrapping 1000 iterations of random resampling with replacement Efron and Tibshirani 1993 using the Package Boot to R software Canty and Ripley 2015 Bootstrapping was used to generate confidence intervals in the time series to determine if averages were significantly different between seasonal andor annual time pe riods Spearman correlation matrix was calculated for all variables Linear regression Willmotts index d Eq 19 the root mean square error RMSE Eq 20 the mean absolute error MAE Eq 21 and the Pearson correlation were used to evaluate the performance of the GPP estimated by TG and VI models d 1 Pi Oi2 Pi O Oi O2 19 RMSE Pi Oi2 n 20 MAE Pi Oi n 21 MS Biudes et al Agricultural and Forest Meteorology 307 2021 108456 5 where Pi is the estimated value Oi the value observed O the average of observed values and n is the number of observations Willmotts statistic relates the performance of an estimation procedure based on the dis tance between estimated and observed values with values ranging from zero no agreement to 1 perfect agreement The RMSE indicates how the model fails to estimate the variability in the measurements around the mean and measures the change in the estimated values around the measured values Willmott and Matssura 2005 The lowest threshold of RMSE is 0 which means there is a complete agreement between the model estimates and measurements As the RMSE MAE has the same size and dimension of the observed values and the estimated value represents the mean difference between estimated and observed The RMSE gives greater weight to large deviations as they are squared while the MAE gives equal weight to all deviations Ideally the values of the MAE and the RMSE should be close to zero Willmott and Matssura 2005 3 Results and discussion 31 Seasonal patterns in micrometeorology Precipitation followed a strong seasonal trend with approximately 80 of the rain in the wet season and sporadic rain in the dry season Table 1 which was a typical seasonal pattern for the Cerrado in this region Vourlitis and da Rocha 2010 Santos et al 2020 The maximum values of precipitation occurred during the wet season in March of 2011 and November of 2012 and months with no record of precipitation during the dry season Fig 2a The volumetric soil water content VSWC was highly correlated with the precipitation with higher values 36 during the wet season Table 1 and Fig 2a The air temperature Tair was on average higher during the wet season than the dry season Table 1 even though the seasonal peak occurred consistently in September each year Fig 2b Low Tair during the dry season is related to a decrease in the solar radiation and the incursion of cold fronts that originate from southern Brazil Rodrigues et al 2014 The vapor pressure deficit VPD exhibited similar trends as Tair Fig 2b and was approximately 30 higher than during the wet season Table 1 VPD was negatively correlated with precipitation and VSWC and positively correlated with Tair Table 2 and high VPD has been implicated in the decline in stomatal conductance transpiration and leaf photosynthesis that occurs in Cerrado trees during the dry season Rodrigues et al 2014 Arruda et al 2016 The photosynthetically active radiation PAR was on average 14 higher during the wet season Table 1 even with an increase in the frequency of cloud cover PAR was positively correlated with precipi tation VSWC and Tair but not VPD Table 2 The minimum monthly PAR was observed in October for all years and maximum in November and March 2011 and 2012 respectively Fig 2c High PAR values during the wet season are caused by greater availability of solar radia tion in the southern hemisphere while low PAR values during the dry period are due to the reduction of solar radiation and an increase in the concentration of atmospheric aerosols caused by biomass burned Rodrigues et al 2014 Biudes et al 2015 32 Seasonal patterns in GPP and spectral vegetation indices The GPP measured in the flux tower GPPEC was highly correlated with precipitation VSWC VPD and PAR and did not correlate with Tair Table 2 The GPPEC in the wet season was 68 higher than in the dry season with high values from December to February and low values from June to August Table 1 and Fig 2c GPP increased considerably at the beginning of the wet season and decreased near the beginning of the dry season Fig 2c The vegetation in the Cerrado expands leaves during the onset of the wet season and consequently has a high photosynthetic capacity as soon as the first rains begin in October Rosa and Sano 2013 Dalmolin et al 2015 Studies have shown that there is a strong correlation between the water potential of the plant and photosynthesis in seasonal tropical forests Sendall et al 2009 and Cerrado Vourlitis and da Rocha 2010 Dalmagro et al 2016 Dalmolin et al 2018 where GPP is largely controlled by the amount of water Seaquist et al 2003 In turn GPP was positively correlated with the NDVI EVI LSWI fPAR and the LUE but was negatively correlated with the LST Table 2 Thus increases in water availability during the wet season led to an increase in the LSWI and LAI which caused a concomitant increase in the NDVI and EVI and thus fPAR and the rate of photosynthesis Ratana et al 2005 Biudes et al 2014 Rodrigues et al 2014 Arruda et al 2016 The NDVI was positively correlated with precipitation VSWC PAR LUE and GPPEC and negatively with Tair and VPD Table 2 NDVI during the wet season was on average 15 higher than in the dry season Table 1 The higher values of NDVI occurred from February to April and the lowest in September Fig 3a The increase in NDVI is related to the period of growth and development of vegetation the increase in the leaf area index and the increase in the concentration of nutrients in the leaves that typically occur in the wet season Xiao et al 2005b This is a pattern closely linked to the Cerrado Ratana et al 2005 even more so than in the Amazon forest Samanta et al 2012 The positive correla tion of NDVI and LEU corroborates with previous research and indicates an increase in fPAR and C gain as leaf area increases Drolet et al 2005 Nakaji et al 2007 Cheng et al 2009 Wu et al 2010 In contrast the Enhanced Vegetation Index EVI had weak corre lation precipitation Tair PAR and GPPEC but EVI not with VSWC DPV and LUE Table 2 The EVI during the wet season was 16 higher in the wet season Table 1 with high values in October and low values in June Table 1 Annual and seasonal total precipitation and average 95 confidence interval volumetric soil water content VSWC m3 m3 air temperature Tair C vapor pressure deficit VPD kPa photosynthetically active radiation PAR mol m2 day1 eddy covariance measured gross primary production GPPEC Normalized Difference Vegetation Index NDVI Enhanced Vegetation Index EVI Land Surface Water Index LSWI Land Surface Temperature LST Photosynthetically Active Radiation Fraction fPAR and Light Use Efficiency LUE gC μmol PAR1 from March 2011 to December 2012 at the Fazenda Miranda Variables 2011 2012 Annual Wet Dry Annual Wet Dry Ppt 761 730 31 1197 1016 181 VSWC 00350002 00450003 00240001 00360001 00410002 00280002 Tair 271 04 276 03 265 07 263 03 273 02 250 05 VPD 131009 103008 160015 105006 092006 123012 PAR 2009 54 2135 83 1878 74 1812 56 1946 74 1636 83 GPPEC 231025 377033 109016 410029 487027 158034 NDVI 052004 059005 046003 053003 056003 049004 EVI 048004 054006 042003 052002 054003 049002 LSWI 005004 003005 011004 001003 003003 008003 LST 321 11 307 06 334 17 314 10 305 11 325 19 fPAR 044002 046002 042002 051002 053003 047003 LUE 003000 004000 003000 003000 004000 003000 MS Biudes et al Agricultural and Forest Meteorology 307 2021 108456 6 Fig 3b The low correlation of EVI and GPPEC and no significant correlation of EVI and LUE was not expected given that EVI has been found to be positively correlated with NEE GPP and evapotranspiration in tropical semideciduous forests Vourlitis et al 2011 and Cerrado Rodrigues et al 2014 Arruda et al 2016 However the NDVI had a higher dynamic range than the EVI Fig 3a suggesting that the NDVI was able to capture the seasonal variation in leaf area development better than the EVI Fig 2 Monthly total of rainfall mm day1 and average 95 confidence interval of volumetric soil water content VSWC A air temperature e vapor pressure deficit B gross primary production GPP and photosynthetically active radiation PAR C from March 2011 to December 2012 at the Fazenda Miranda The shaded portion in each panel depicts the dry season Table 2 Spearman Correlation Matrix of precipitation Ppt volumetric soil water content VSWC air temperature Tair vapor pressure deficit VPD photosynthetically active radiation PAR Gross Primary Production estimated by eddy covariance GPPEC Normalized Difference Vegetation Index NDVI Enhanced Vegetation Index EVI Land Surface Water Index LSWI Land Surface Temperature LST Photosynthetically Active Radiation Fraction fPAR and Light Use Efficiency LUE from March 2011 to December 2012 at the Fazenda Miranda Ppt VSWC Tair VPD PAR GPPEC NDVI EVI LSWI LST fPAR VSWC 090 Tair 018 013 VPD 056 068 067 PAR 041 032 029 005 GPPEC 080 074 014 061 040 NDVI 057 077 037 076 035 069 EVI 043 017 032 004 023 035 001 LSWI 068 082 029 076 033 074 090 001 LST 036 057 049 072 004 048 056 001 059 fPAR 036 047 026 053 039 064 062 035 058 046 LUE 073 081 003 056 044 077 080 009 093 045 050 pvalue 005 pvalue 001 pvalue 0001 MS Biudes et al Agricultural and Forest Meteorology 307 2021 108456 7 The LSWI was positively correlated with precipitation VSWC VPD PAR GPPEC and LUE and negatively correlated with Tair Table 2 The LSWI had a welldefined seasonality with higher values 128 in the wet season Table 1 The highest LSWI values occurred from December to March and the lowest values from September to November Fig 3c The high correlations observed indicate that in the absence of soil moisture data LSWI satisfactorily represents the dynamics of water in a Cerrado environment The Land Surface Temperature LST had more accentuated sea sonality than that observed in the air temperature Tair The LST during the dry season was 7 higher than in the wet season which is the opposite of Tair Table 1 During the analyzed period LST varied from 28 to 38 C with higher values at the end of the dry season Fig 3b LST was negatively correlated with VSWC NDVI LSWI LUE and GPPEC and positively correlated with Tair and DPV Table 2 The fraction of photosynthetically active radiation fPAR was 10 higher in the wet season Table 1 The lowest fPAR values occurred from August to September and the highest from March to April Fig 3c Unlike PAR fPAR correlated positively with VSWC LUE and GPPEC and negatively with Tair VPD and LST Table 2 The Light Use Efficiency LUE was well coupled with all variables except Tair and EVI Table 2 The LUE during the wet season was 27 higher in the wet season Table 1 The highest values of LUE occurred from January to March and e lowest one occurred from July to September Fig 3c 33 GPP Models and product accuracy The annual average of GPP estimated by VPM GPPVPM TG GPPTG and MODIS GPPMODIS did not differ of the annual average of measured GPP GPPEC Table 3 Fig 4 but the annual average of GPP estimated by VI models overestimated the GPP by 43 GPPNDVI 63 GPPEVI and 47 GPPNDVIxEVI GPPVPM GPPTG GPPEVI and GPPMODIS during the wet season were on average similar to GPPEC but the GPPNDVI and GPPNDVIxEVI were 42 and 25 higher than GPPEC On the other hand no model could estimate the correct magnitude GPP during the dry season GPPVPM was overestimated by 60 GPPTG by 99 GPPEVI by 141 GPPNDVI by 148 GPPNDVIxEVI by 137 and GPPMODIS by 81 during the dry season Table 3 Fig 4 GPPVPM was strongly correlated with GPPEC and had a higher Willmott index and lower errors than the other models Table 3 Although GPPNDVI had a strong correlation with GPPEC it overestimated GPP which resulted in greater errors The fact that all of the models provided similar estimates of GPP Fig 4 is not surprising given that they are driven by similar variables namely NDVI EVI LUC LSWI and fPAR which themselves are highly correlated Table 2 Given that the GPPVPM had the lowest MAE and Fig 3 Monthly average 95 confidence interval for Normalized Difference Vegetation Index NDVI and Enhanced Vegetation Index EVI A Land Surface Water Index LSWI and Land Surface Temperature LST B and Photosynthetically Active Radiation Fraction fPAR C from March 2011 to December 2012 at the Fazenda Miranda The shaded portion in each panel depicts the dry season MS Biudes et al Agricultural and Forest Meteorology 307 2021 108456 8 RMSE and the highest Willmott index it is tempting to suggest that this model was the best at estimating GPP for this Cerrado ecosystem This model has been used successfully for assessing GPP for tropical forest Xaio et al 2005b semideciduous forest Biudes et al 2014 Dane lichen et al 2015 so it is not surprising that it performed well in grassdominated Cerrado In addition the flexibility of this model to capture seasonal dynamics in tropical forest and Cerrado which differ significantly in the magnitude and direction of their seasonal GPP dy namics RestrepoCoupe et al 2013 makes the VPM model attractive for climate diverse regions such as the Amazon Basin However if temperature data are not available the TG and VI models also captured the seasonal variations in GPP and these models rely solely on satellite data The NDVI version performed better than the EVI or NDVI x EVI versions presumably because the NDVI had a higher correlation with VSW precipitation and other variables that are important in controlling seasonal variations in leaf area production Table 2 and thus GPP The Modis GPP estimate also performed well which is consistent with other studies in seasonal tropical forests Danelichen et al 2015 4 Conclusions We compared several satellitebased models of GPP to measured values in a grassdominated Cerrado stand in southern Mato Grosso Brazil While our research was potentially limited in spatial 1 site and temporal ca 2 years extent we found that meteorological variables exhibited large seasonal variations and in general the wet season was wetter and warmer than dry seasons which was on average cooler but much drier and with higher VPD These seasonal variations caused large variations in GPP measured by eddy covariance and satellitebased vegetation indices or derived quantities and there were strong posi tive correlations between GPP NDVI EVI LSWI fPAR and LUE Because all of the GPP models relied on some of these remotesensing components all of the GPP models tested here were able to capture the seasonal variations in GPP measured from eddy covariance How ever while the VPM model exhibited the best performance all of the models exhibited biases such that no single model was vastly superior to the others Thus model selection for this and other seasonal grass dominated tropical savannas will likely be a function of data availabil ity andor quality Climate change is expected to cause an increase in seasonal climate variation in Cerrado Boisier et al 2015 and there is evidence that dry season intensification is already increasing in the southern part of the Amazon Basin Fu et al 2013 Gloor et al 2015 Vourlitis et al 2015 Remotesensing based models such as those tested here will likely be an important tool for assessing how climate variability alters C cycling dynamics in these spatially and temporally heterogeneous landscapes Declaration of Competing Interest The authors declare that they have no known competing financial Table 3 Mean 95 confidence interval gross primary production GPP measured from eddy covariance GPPEC and estimated by the VPM GPPVPM TG GPPTG and VI models GPPNDVI GPPEVI and GPPNDVI x EVI and MODIS product GPPMODIS from March 2011 to December 2012 at the Fazenda Miranda Also shown are the per formance statistics for the modeled GPP relative to the measured GPP including the slope intercept and correlation coefficient for the linear regression between modeled GPP dependent variable vs measured GPP Willmonts d and the mean absolute error MAE and the Relative Mean Square Error RMSE GPPEC GPPVPM GPPTG GPPNDVI GPPEVI GPPNDVI x EVI GPPMODIS Annual 293 245 320 282 319 292 419 367 477 419 430 383 379 330 Wet 432 048 419 047 376 031 515 067 614 078 540 056 505 045 Dry 126 029 201 025 251 028 303 044 312 048 298 034 228 047 Slope 066 037 094 062 076 078 Intercept 126 210 200 237 208 151 r 079 071 074 058 083 077 d 089 075 072 071 080 085 MAE 085 120 200 168 143 111 RMSE 109 141 239 213 171 145 pvalue 0001 Fig 4 Monthly average 95 confidence interval measured GPP GPPEC and the GPP estimated by VPM model GPPVPM TG model GPPTG VI models GPPNDVI GPPEVI and GPPNDVI x EVI and MODIS product GPPMODIS from March 2011 to December 2012 at the Fazenda Miranda The shaded portion in each panel depicts the dry season MS Biudes et al Agricultural and Forest Meteorology 307 2021 108456 9 interests or personal relationships that could have appeared to influence the work reported in this paper Acknowledgements The research was supported in part by Universidade Federal de Mato Grosso UFMT Programa de PosGraduaçao em Física Ambiental PPGFAIFUFMT Instituto Federal de Mato Grosso IFMT Coor denaçao de Aperfeiçoamento de Pessoal de Nível Superior CAPES Conselho Nacional de Desenvolvimento Científico e Tecnologico CNPq code numbers 31087920175 and 30576120188 Edital Universal 012016 code number 40746320160 e Fundaçao de Amparo a Pes quisa do Estado de Mato Grosso FAPEMAT PRONEM 2014 code number 5613972014 References Anthoni PM Law BE Unsworth MH 1999 Carbon and water vapor exchange of an opencanopied ponderosa pine ecosystem Agric For Meteorol 95 3 151168 httpsdoiorg101016S0168192399000295 Arruda PHZ Vourlitis GL Santanna GB Pinto Júnior OB Lobo FA Nogueira JS 2016 Large net CO2 loss from a grassdominated tropical savanna in southcentral Brazil in response J Geophys Res 121 21102124 httpsdoiorg 1010022016JG003404 Baldocchi D Chu H Reichstein M 2018 Interannual variability of net and gross ecosystem carbon fluxes a review Agric For Meteorol 249 520533 httpsdoi org101016jagrformet201705015 Biudes MS Souza MC Machado NG Danellichen VHM Vourlitis GL Nogueira JS 2014 Modelling gross primary production of a tropical semi deciduous forest in the southern Amazon Basin Int J Remote Sens 35 4 15401562 httpsdoiorg101080014311612013878059 Biudes MS Vourlitis GL Machado NG Arruda PHZ Neves GARN Lobo FA Neale CMU Nogueira JS 2015 Patterns of energy exchange for tropical ecosystems across a climate gradient in Mato Grosso Brazil Agric Forest Meteorol 202 112124 httpsdoiorg101016jagrformet201412008 Boisier JP Ciais P Ducharne A Guimberteau M 2015 Projected strengthening of Amazonian dry season by constrained climate model simulations Nat Clim Chang 5 7 656660 httpsdoiorg101038nclimate2658 Burba G 2013 Eddy Covariance Method For Scientific Industrial Agricultural and Regulatory Applications A Field Book on Measuring Ecosystem Gas Exchange and Areal Emission Rates LICOR Biosciences Lincoln Nebraska p 331 Canty A Ripley B 2015 Boot ootstrap R SPlus functions R package version 1317 Cheng YB Middleton EM Hilker T Coops NC Black TA Krishnan P 2009 Dynamics of spectral bioindicators and their correlations with light use efficiency using directional observations at a Douglasfir forest Meas Sci Technol 20 15 095107 httpsdoiorg10108809570233209095107 Dalmagro HJ Lobo FA Vourlitis GL Dalmolin AC Antunes Junior MZA Ortíz CER Nogueira JS 2016 Photosynthetic response of a wetland and an uplandadapted tree species to seasonal variations in hydrology in the Brazilian Cerrado and Pantanal Acta Physiol Plant 38 107 httpsdoiorg101007 s1173801621257 Dalmolin AC Lobo FA Vourlitis GL Dalmagro HJ Antunes Junior MZ Ortiz C ER 2018 Physiological adjustments of an invasive tree species to extreme hydrological events in a tropical seasonal wetland Tree 32 13651375 httpsdoi org101007s0046801817188 Dalmolin AC Lobo FA Vourlitis GL Silva PR Dalmagro HJ Antunes Junior M Z Ortíz CER 2015 Is the dry season an important driver of phenology and growth for two Brazilian savana tree species with contrasting leaf habits Plant Ecol 216 407417 httpsdoiorg101007s1125801404455 Danelichen VHM Biudes MS Velasque MCS Machado NG Gomes RSR Vourlitis GL Nogueira JS 2015 Estimating of gross primary production in an AmazonCerrado transitional forest using MODIS and Landsat imagery Anais da Academia Brasileira de Ciˆencias 87 15451564 httpsdoiorg1015900001 3765201520140457 Drolet GG Huemmrich KF Hall FG Middleton EM Black TA Barr AG Margolis HA 2005 A MODISderived photochemical reflectance index to detect interannual variations in the photosynthetic lightuse efficiency of a boreal deciduous forest Remote Sens Environ 98 212224 httpsdoiorg101016j rse200507006 Drolet GG Middleton EM Huemmrich KF Hall FG Amiro BD Barr AG Black TA McCaughey JH Margolis HA 2008 Regional mapping of gross light use efficiency using MODIS spectral indices Remote Sens Environ 112 30643078 httpsdoiorg101016jrse200803002 Efron B Tibshirani R 1993 An Introduction to the Bootstrap Chapman and Hall New York Falge E Baldocchi D Olson RJ Anthoni P Aubinet M Bernhofer C Burba G Ceulemans G Clement R Dolman H Granier A Gross P Grunwald T Hollinger D Jensen NO Katul G Keronen P Kowalski A Lai CT Law BE Meyers T Moncrieff J Moors E Munger JW Pilegaard K Rannik U Rebmann C Suyker A Tenhunen J Tu K Verma S Vesala T Wilson K Wofsy S 2001 Gap Filling Strategies for Long Term Energy Flux Data Sets Agric For Meteorol 107 7177 httpsdoiorg101016S0168192300 002355 Falge E Tenhunen J Baldocchi DD Aubinet M Bakwin P Berbigier P Bernhofer C Bonnefond JM Burba G Clement R Davis KJ Elbers JA Falk M Goldstein AH Grelle A Granier A Grünwald T Guomundsson J Hollinger D Janssens IA Keronen P Kowalski AS Katul G Law BE Malhi Y Meyers T Monson RK Moors E Munger JW Oechel W Paw U Pilegaard K Rannik Ü Rebmann C Suyker A Thorgeirsson H Tirone G Turnipseed A Wilson K Wofsy S 2002 Phase and amplitude of ecosystem carbon release and uptake potentials as derived from FLUXNET measurements Agric For Meteorol 113 14 7595 httpsdoiorg101016S0168192302 00103X Fearnside PM 2000 Global warming and tropical landuse change greenhouse gas emissions from biomass burning decomposition and soils in forest conversion shifting cultivation and secondary vegetation Clim Change 46 115158 https doiorg101023A1005569915357 Fu R Yina L Wenhong L Ariasc PA Dickinson RE Huanga L Chakrabortya S Fernandes K Liebmann B Fisher R Ranga BM 2013 Increased dryseason length over southern Amazonia in recent decades and its implication for future climate projection Proc Natl Acad Sci 110 1811018115 httpsdoiorg 101073pnas1302584110 Giambelluca TW Scholz FG Bucci SJ Meinzer FC Goldstein G Hoffmann W A Franco AC Buchert MP 2009 Evapotranspiration and energy balance of Brazilians savannas with contrasting tree density Agric For Meteorol 149 13651376 httpsdoiorg101016jagrformet200903006 2009 Giltelson AA Peng Y Arkebauer TJ Suyker AE 2015 Productivity absorbed photosynthetically active radiation and light use efficiency in crops implications for remote sensing of crop primary production J Plant Physiol 177 100109 https doiorg101016jjplph201412015 Gloor M Barichivich J Ziv G Brienen R Schongart J Peylin P Cintra BBL Feldpausch T Phillips O Baker J 2015 Recent Amazon climate as background for possible ongoing and future changes of Amazon humid forests Glob Biogeochem Cycles 29 13841399 httpsdoiorg1010022014GB005080 Griebel A Bennett LT Metzen D Cleverly J Burba G Arndt SK 2016 Effects of inhomogeneities within the flux footprint on the interpretation of seasonal annual and interannual ecosystem carbon exchange Agric For Meteorol 221 5060 httpsdoiorg101016jagrformet201602002 Huete AR Liu HQ Batchily K Van Leeuwen W 1997 A Comparison of Vegetation Indices Global Set of TM Images for EOSMODIS Remote Sens Environ 59 440451 httpsdoiorg101016S0034425796001125 Inoue Y Penuelas J Miyata A Mano M 2008 Normalized difference spectral indices for estimating photosynthetic efficiency and capacity at a canopy scale derived from hyperspectral and CO2 flux measurements in rice Remote Sens Environ 112 1 156172 httpsdoiorg101016jrse200704011 Jin C Xiao X Merbold L Arneth A Veenendaal E Kutsch WL 2013 Phenology and gross primary production of two dominant savanna woodland ecosystems in Southern Africa Remote Sens Environ 135 189201 httpsdoiorg101016j rse201303033 Li Z Yu G Xiao X Li Y Zhao X Ren C Zhang L Fu Y Modeling gross primary production of alpine ecosystems in the Tibetan Plateau using MODIS images and climate data Remote Sens Environ 1073510519 doi101016j rse200610003 Ma X Huete A Yu Q Coupe NR Beringer J Hutley LB Kanniah KD Cleverly J Eamus D 2014 Parameterization of an ecosystem lightuseefficiency model for predicting savanna GPP using MODIS EVI Remote Sens Environ 154 253271 httpsdoiorg101016jrse201408025 McMillen RT 1988 An eddy correlation technique with extended applicability to non simple terrain Boundary Layer Meteorol 43 231425 httpsdoiorg101007 BF00128405 Monteith JL 1972 Solar radiation and production in tropical ecosystems J Appl Ecol 9 3 747766 httpsdoiorg1023072401901 Nakaji T Ide R Oguma H Saigusa N Fujinuma Y 2007 Utility of spectral vegetation index for estimation of gross CO2 flux under varied sky conditions Remote Sens Environ 109 3 274284 httpsdoiorg101016j rse200701006 Radambrasil 1982 Levantamentos Dos Recursos Naturais Ministerio das Minas de Energia Secretaria Geral Projeto Radambrasil Rio de Janeiro Brazil Ratana P Huete AR Ferreira L 2005 Analysis of cerrado physiognomies and conversion in the MODIS seasonal temporal domain Earth Interactions 9 3 122 httpsdoiorg1011751087356220050090001AOCPAC20CO2 RestrepoCoupe N Huete A Davies K Cleverly J Beringer J Eamus D Van Gorsel E Hutley LB Meyer WS 2015 MODIS vegetation products as proxies of photosynthetic potential a look across meteorological and biologic driven ecosystem productivity Biogeosci Discussions 12 1921319267 httpsdoiorg105194 bgd12192132015 RestrepoCoupe N da Rocha HR Hutyra LR Araujo AC Borma LS Christoffersen B Cabral OMR de Camargo PB Cardoso FL Costa ACL Fitzjarrald DR Goulden ML Kruijt B Maia JMF Malhi YS Manzi AO Miller SD Nobrei AD von Randow C Sa LDA Sakai RK Tota J Wofsy S C Zanchi FB Saleskaa SR 2013 What drives the seasonality of photosynthesis across the Amazon basin A crosssite analysis of eddy flux tower measurements from the Brasil flux network Agric For Meteorol 182183 128144 httpsdoi org101016jagrformet201304031 Ribeiro JF Walter BMT 1998 Fitofisionomias do bioma cerrado In Sano SM Almeida SP Eds Ambiente e flora Planaltina EmbrapaCPAC Cerrado pp 89166 MS Biudes et al Agricultural and Forest Meteorology 307 2021 108456 10 Rodrigues TR Vourlitis GL Lobo FA Oliveira RG Nogueira JS 2014 Seasonal variation in energy balance and canopy conductance for a tropical savanna ecosystem of southcentral Mato Grosso Brazil J Geophys Res 119 1 113 httpsdoiorg1010022013JG002472 Rodrigues TR Vourlitis GL Lobo FA Santanna FB Arruda PHZ Nogueira JS 2016 Modelling canopy conductance under contrasting seasonal conditions for a tropical savanna ecosystem of south central Mato Grosso Brazil Agricultural and Forest Meteorology 218219 218229 Rosa R Sano EE 2013 Determinaçao da produtividade primaria liquida NPP de pastagens na bacia do rio Paranaíba usando imagens MODIS GeoFocus Rev Int Ciencia Tecnol Inf Geogr 13 367395 Ruimy A Jarvis PG Baldocchi DD Saugier B 1995 CO2 Fluxes over Plant Canopies and Solar Radiation a Review Adv Ecol Res 26 168 httpsdoiorg 101016S006525040860063X Samanta A Knyazikhin Y Xu L Dickinson RE Fu R Costa MH Saatchi SS Nemani RR Myneni RB 2012 Seasonal changes in leaf area of Amazon forests from leaf flushing and abscission J Geophys Res 117 G1 G01015 httpsdoi org1010292011JG001818 Santiago A Prado RJ Filho PM Alonso RV 2015 Caracterizaçao do material particulado suspenso no ar de CuiabaMT no período de queimadas Rev Mater 20 1 273283 httpsdoiorg101590S15177076201500010027 Santos LOF Machado NG Querino CAS Pedreira Junior AL Ivo IO Lotufo Neto N Biudes MS 2020 Hourly precipitation patterns in a Brazilian Tropical City Rev Bras Climatol 26 411424 httpsdoiorg105380abclima v26i071488 Seaquist JW Olsson L Ardo J 2003 A remote sensingbased primary production model for grassland biomes Ecol Modell 169 1 131155 httpsdoiorg 101016S0304380003002679 Sendall MM Vourlitis GL Lobo FA 2009 Seasonal variation in the maximum rate of leaf gas exchange of canopy and understory tree species in an Amazonian semi deciduous forest Braz J Plant Physiol 21 6574 httpsdoiorg101590S1677 04202009000100008 Silva FB Shimabukuro YE Aragao LEOC Anderson LO Pereira G Cardozo F Arai E 2013 Largescale heterogeneity of Amazonian phenology revealed from 26 year long AVHRRNDVI timeseries Environ Res Lett 8 2 112 httpsdoiorg 1010881748932682024011 Sims DA Rahman AF Cordova VD ElMasri BZ Baldocchi DD Bolstad PV Flanagan LB Goldstein AH Hollinger DY Misson L Monson RK Oechel W C Schmid HP Wofsy SC Xu L 2008 A new model of gross primary productivity for North American ecosystems based solely on the enhanced vegetation index and land surface temperature from MODIS Remote Sens Environ 112 4 16331646 httpsdoiorg101016jrse200708004 Souza MC Biudes MS Danelichen VHM Machado NG Musis CR Vourlitis G L Nogueira JS 2014 Estimation of gross primary production of the Amazon Cerrado transitional forest by remote sensing techniques Rev Bras Meteorol 29 1 112 httpsdoiorg101590S010277862014000100001 Vermote EF Kotchenova S 2008 Atmospheric correction for the monitoring of land surfaces J Geophys Res 113 D23 D23S90 httpsdoiorg101029 2007JD009662 Vourlitis GL da Rocha HR 2010 Flux dynamics in the Cerrado and CerradoForest Transition of Brazil In Hill MJ Hanan NP Eds Ecosystem Function in Global Savannas Measurement and Modeling at Landscape to Global Scales pp 97116 Boca Raton FL USA2011 Vourlitis GL Lobo FA Zeilhofer P Nogueira JS 2011 Temporal patterns of net CO2 exchange for a tropical semideciduous forest of the southern Amazon Basin J Geophys Res 116 G3 G03029 httpsdoiorg1010292010JG001524 Vourlitis GL Lobo FD Lawrence S de Lucena IC Pinto OB Dalmagro HJ Eugenia C Ortiz R Nogueira JS 2013 Variations in stand structure and diversity along a soil fertility gradient in a Brazilian savanna Cerrado in southern Mato Grosso Soil Sci Soc Am J 77 4 13701379 httpsdoiorg102136 sssaj20120336 Vourlitis GL Nogueira JS Lobo FA Pinto Jr OB 2015 Variations in evapotranspiration and climate for an Amazonian semideciduous forest over seasonal annual and El Nino cycles Int J Biometeorol 59 217230 httpsdoi org101007s0048401408371 Vourlitis GL Nogueira JS Lobo FA Pinto Jr OB 2014 Variations in evapotranspiration and climate for an Amazonian semideciduous forest overseasonal annual and El Nino cycles Int J Biometeorol httpsdoiorg 101007s0048401408371 Wang Z Xiao X Yan X 2010 Modeling gross primary production of maize cropland and degraded grassland in northeastern China Agric For Meteorol 150 9 11601167 httpsdoiorg101016jagrformet201004015 Webb EK Pearman GI Leuning R 1980 Corrections of flux measurements for density effects due to heat and water vapor transfer Quart J R Meteorol Soc 106 447 85100 httpsdoiorg101002qj49710644707 Willmott CJ Matssura K 2005 Advantages of the mean absolute error MAE over the root mean square error RMSE in assessing average model performance Clim Res 30 7992 httpsdoiorg103354cr030079 Wohlfahrt G Anfang C Bahn M Haslwanter A Newesely C Schmitt M Drosler M Pfadenhaeur J Cernusca A 2005 Quantifying nighttime ecosystem respiration of a meadow using eddy covariance chambers and modelling Agric For Meteorol 128 34 141162 httpsdoiorg101016jagrformet200411003 Wu C Chen JM Huang N 2011 Predicting gross primary production from the enhanced vegetation index and photosynthetically active radiation evaluation and calibration Remote Sens Environ 115 12 34243435 httpsdoiorg101016j rse201108006 Wu C Zheng N Gao S 2010 Gross primary production estimation from MODIS data with vegetation index and photosynthetically active radiation in maize JGeophys Res 115 D12 111 httpsdoiorg1010292009JD013023 2010 Xiao X Hollinger D Aber J Goltz M Davidson EA Zhang Q Moore III B 2004 Satellitebased modeling of gross primary production in an evergreen needleleaf forest Remote Sens Environ 89 4 519534 httpsdoiorg101016j rse200311008 Xiao X Zhang Q Hollinger D Aber J Moore III B 2005a Modeling seasonal dynamics of gross primary production of an evergreen needleleaf forest using MODIS images and climate data Ecol Appl 15 3 954969 httpsdoiorg10189004 0470 Xiao X Zhang Q Saleska S Hutyra L de Camargo P Wofsy S Frolking S Boles S Keller M Moore III B 2005b Satellitebased modeling of gross primary production in a seasonally moist tropical evergreen forest Remote Sens Environ 94 1 105122 httpsdoiorg101016jrse200408015 MS Biudes et al