Event-based hydrology and sedimentation in paired watersheds under commercial eucalyptus and grasslands in the Brazilian Pampa biome

Érson Diniz Eling ,José Migul Richrt ,*,Jhon Jiro Zulug Pláz ,Mirim Frnn Rorigus ,Mirin Lgo Vlnt ,Rosn Bros Lops Cvlcnt ,Polo Rggini ,Rghvn Srinivsn

a Soils Department,Universidade Federal de Santa Maria,Avenida Roraima,1000 Bairro Camobi,Santa Maria,RS,97105-900,Brazil

b Corporación Colombiana de Investigación Agropecuaria(AGROSAVIA),Centro de Investigación Nataima,Espinal,Colombia

c CMPC Celulose Riograndense,Rua São Geraldo,1680,Guaíba,RS,92500-000,Brazil

d Department of Civil Engineering,University of Siegen,Paul Bonatz Strasse 9-11,Siegen D,57068,Germany

e Department of Ecology and Conservation Biology,Texas A&M University(TAMU),534 John Kimbrough Blvd.,Room 305,College Station,TX,77843-2120,USA

Keywords:Southern grasslands Paired watersheds Runoff Erosion LISEM

A B S T R A C T Ecological and hydrological impacts are expected when watersheds are under commercial eucalyptus.In this study we evaluated event-based hydrology and sedimentation in paired watersheds under commercial eucalyptus and grasslands cover.Flow,suspended sediment concentration,and sediment production were measured and simulated with the Limburg Soil Erosion Model(LISEM)in paired hydrographic watersheds located in southern Brazil;one occupied with 7-years old Eucalyptus saligna plantation(forest watershed-FW;0.83 km2)and another Pampa biome with native and exotic grasses used for beef cattle production(grassland watershed-GW;1.10 km2).The effect of rainfall on hydrological and sedimentological variables was more evident in GW compared with FW.The shape of the hydrograph was steepest in GW,with rapid,marked increase and decrease in flow during rainfall.The parameters changed for model calibration in FW and GW were soil saturated hydraulic conductivity,initial soil moisture,and Manning roughness coefficient,whereby the statistical performance was classified as good(NSE＞0.75).Fourteen events different from those used in the calibration process were validated for FW and GW,with NSE index classified as“good”and“very good”for hydrology.Our study shows limitations of LISEM in calibrating and validating the sediment yield for individual events due to the spatial variability of factors controlling soil erosion.The validated modeling results of this study may be applied in simulating soil and crop scenarios in watersheds cultivated with commercial forest and grassland for grazing.

1.Introduction

Afforestation by eucalyptus of land previously used for cattle raising and agriculture(Oliveira et al.,2017)in the Pampa biome in southern Brazil has raised concerns because of potential detrimental changes in hydrology and soil erosion(Reichert et al.,2017;Rodrigues et al.,2018;Valente et al.,2020).Forests have significant effects on ecology(Evans & Turnbull,2004),hydrology(Andréassian,2004;Ferreto,Reichert,Lopes Cavalcante,&Srinivasan,2020;Oki & Kanae,2006;Reichert,Cechin,Reinert,Rodrigues,& Suzuki,2018;Reichert et al.,2017),soil erosion(Rodrigues et al.,2014,2018;Valente et al.,2020)and edaphic aspects(Cavalli,Reichert,Rodrigues,& de Araújo,2020;França,Reichert,Holthusen,Rodrigures,& de Araújo,2021;Holthusen,Brandt,Reichert,& Horn,2018;Reichert et al.,2015,2016,2018;Reichert,Rodrigues,Bervald,& Kato,2016).

With eucalyptus afforestation,there are connections among ecology,hydrology and erosion.Land use changes affect the relationships of living beings with the environment and with each other,influencing the functioning of local environment due to the ecophysiological plant relocation.Hydrology is a field of science that examines the movement,distribution and physical and chemical properties of water,and water relates with the ecology of all living forms in the environment.Furthermore,ecology,hydrology and erosion processes are interconnected since soil erosion,promoted by hydrological processes,may cause a progressive deterioration of the landscape,especially on steep agricultural slopes,affecting ecological functions.Afforestation with eucalyptus,however,can be an important resource for reducing soil erosion,by controlling runoff energy in the long run(Körner,2018;McCabe et al.,2017;Pijl,Reuter,Quarella,Vogel,& Tarolli,2020;Valente et al.,2020).

The effects of large-scale eucalyptus production in southern Brazil are not known and there are uncertainties about the capacity of the Pampa biome in supporting large-scale forestry production without jeopardizing soil and water resources.In this biome,eucalyptus stands are being planted even in sandy soils,which are susceptible to erosion and have low water retention capacity,fertility,and organic matter content(Reichert,Amado,Reinert,Rodrigues,& Suzuki,2016).Further,in many areas of the world,large-scale eucalyptus plantations have been set up because of the rapid financial return.In some countries,such as Spain,these plantations already occupy more space than native forests(Goded et al.,2019).These large-scale plantations did not demonstrate a clear effect on soil erosion and river flow in four river basins in southeastern Australia(Dresel et al.,2018).Occurrence of erosion was related to the reduction of eucalyptus forests in north-central Portugal(Malvar et al.,2017),and runoff and sediment concentration decreased by more than half after the implantation of eucalyptus forests in the Ethiopian plateau(Mhiret et al.,2019)and in southern China(Zhao,Zhang,Lan,Tongway,& Freudenberger,2020).

Soil degradation is augmented by animal trampling due to extensive livestock activity,which increases compaction(Holt,Bristow,& McIvor,1996)particularly near the soil surface(Ambus,Reichert,Gubiani,& de Faccio Carvalho,2018;Cecagno et al.,2016;Collares,Reinert,Reichert,& Kaiser,2011;Kunz et al.,2013;Martins et al.,2003;Mesquita & Moraes,2004;Watung,Sutherland,& El-Swaify,1996).Cattle change the relationship between soil and plant in tree regeneration in the Pampa(Redin et al.,2017),modify land surface forms and accelerate degradation processes(Thomaz & Dias,2009).

Studies on rainfall-flow-erosion modeling mainly seek to understand processes related to the spatio-temporal variability of sediment yield at the watershed scale.In Brazil,there are few studies that have monitored watersheds with the purpose of evaluating soil and water losses through physically-based modeling(Dambrós,2011;de Barros,Minella,Dalbianco,& Ramon,2014;Peláez,2014;Rodrigues et al.,2014).Long-term data are scarce,and costly to obtain.Therefore,models are important tools to provide insights into hydrological,sedimentological,and sediment yield processes that are relevant in the context of using and managing natural resources.

The Limburg Soil Erosion Model(LISEM)simulates streamflow and erosion processes during a single rainfall event(De Roo,Wesseling,& Ritsema,1996;Jetten,2002).LISEM was calibrated and used extensively with satisfactory results for rural watersheds with intense soil use and degradation in southern Brazil(Gomes,2008;Moro,2011)and also in watersheds in Belgium(Takken et al.,1999),Kenya(Hessel & Tenge,2008),Netherlands(Sheikh,van Loon,Hessel,& Jetten,2010),Ethiopia(Grum et al.,2017),Spain(Grum et al.,2017)and Norway(Starkloff,Stolte,Hessel,Ritsema,& Jetten,2018),with promising results,but information regarding the use of LISEM in cultivated forest watersheds is incipient(Rodrigues et al.,2014).

Our hypothesis was that,for single rainfall events,watershed under eucalyptus plantation generates less sediment concentration,sediment yield,and streamflow compared with grassland watersheds in the Pampa biome.Our objective was to evaluate the ability of LISEM to simulate the effect of vegetation on the flow dynamics,water flow and sediment production in paired watersheds,and to compare sediment yield and flow in watersheds with eucalyptus plantation and grasslands.We believe that through strategies of continuous monitoring and subsequent application of modeling techniques,it is possible to discriminate the effect of different types of vegetation and soil cover,such as eucalyptus plantation and pasture,on the influence on the dynamics of surface runoff formation,being that runoff is directly related to the slope of the landscape and soil texture.

2.Material and methods

2.1.Paired watersheds

The study was conducted in a watershed with Eucalyptus saligna(FW)and another with grassland for beef cattle grazing(GW),in southern Brazil(Fig.1).Climate is humid subtropical with no drought,an average annual temperature of 18.6°C(Moreno,1961),mean annual precipitation of 1356 mm,and mean monthly precipitation above 60 mm(Alvares,Stape,Sentelhas,De Moraes Gonçalves,& Sparovek,2013).

The FW watershed has an area of 0.83 km2,average time of concentration of 1:47 h,mean slope of 8%,and soils Ultisols,Ultisols,Inceptisols,Entisols,and Entisols(USDA-Soil Survey Staff,1999).Besides Eucalyptus saligna planted in 2006 with a 3.0×3.3-m spacing(Fig.2),grasslands in surrounding areas consists of grasses and shrubs,and riparian vegetation characterized by arboreal stratum of native species,as described in Valente(2018).The GW watershed(Fig.2)has an area of 1.10 km2,average time of concentration of 2:52 h,mean slope of 3.08%,and soils Ultisols,Inceptisols,and Entisols(USDA-Soil Survey Staff,1999).Pasture consists of native grasses and sown oats.There are arboreal clumps or isolated individuals of native trees,along with native riparian vegetation as described in(Valente,2018).

2.2.Hydrology and erosion monitoring at watershed scale

Hydrosedimentometric monitoring was carried out in two periods.The first period,from January 2012 to September 2013,was used to obtain data for model calibration.During the second period,between January 2019 and May 2019,data were obtained for model validation.Monitoring flumes where equipped with limnigraphs,turbidimeters and pluviometers,and connected to dataloggers that recorded every 10 min.

2.2.1.Sediment concentration and sediment yield

Suspended sediment was sampled with a USDH-48 sampler during rainfall events,whereas turbidity was measured continuously and concentration of suspended sediment estimated with a calibration curve.Total sediment yield was determined from integrated bed load and suspended sediment concentration obtained during the automatic monitoring period.Hydrographs and sedimentographs were examined based on their shape and in relation to streamflow and sedimentation occurrence.

2.3.Hydrological and sedimentological modeling

Fig.1.Location of the studied watersheds.

Fig.2.Land use and soil classes of the forest and grassland watersheds.

Hydrological processes were modeled using the Limburg Soil Erosion Model(LISEM)model(De Roo & Jetten,1999),which describes infiltration,runoff,deposition and sediment yield,a distributed-model recommended for small(10-300 ha)watersheds.Hydrological variables were runoff volume,maximum streamflow,peak time,and hydrograph shape.Sedimentological variables under observation were sediment yield,maximum concentration of suspended sediment,peak time,and sedimentograph shape.These data were obtained through monitoring of quantities corresponding to the variables adjusted in the calibration procedure.

Calibration and validation were performed by manual optimization of sensitive model parameters.The optimized sensitive parameters were soil saturated hydraulic conductivity(ks),total porosity(θs),initial soil moisture(θi),Manning’s n coefficient(n),median particle size distribution(D50),cohesion of the wet soil(Coh),cohesion by plant roots(Coah),and aggregate stability(As).The optimization of each parameter was uniform.In other words,as an example,when ks was reduced by 10%,this percentage was added for all land uses.Land use and soil classification information(Table 1)was assigned to digital maps with a 5x5-m spatial resolution.Since the model does not simulate total streamflow but total surface runoff,a separation between baseflow and surface runoff from the total runoff volume is needed.This was achieved by hydrograph separation using the constant-slope approach proposed by(Vente,1964).

Since rainfall events with higher magnitudes generate larger soil and water losses,they were selected for calibration and validation of the LISEM model.Initial soil moisture was defined as a fraction of total porosity occupied by the water phase,and estimated by using the ROSETTA pedotransfer functions based on sand,silt and clay fractions(Twarakavi,Sakai,&Šim◦unek,2009),and available in Hydrus 1-D(Šimunek,Šejna,& van Genuchten,1998).The initial conditions were given by total precipitation volume of the antecedent rainfall event and the modeled event as well as the interstorm period elapsed between those two events expressed in days.

The model was parameterized using data measured in the field or the laboratory or with the aid of information published in the literature(Table 1).Surface random roughness RR is the standard deviation of surface elevation along a small transect using a frame with bars as reference;ground cover by vegetation was verified in the field,rock cover was estimated locally,vegetation height and leaf area index(LAI;m2m-2)were obtained from(Rodrigues et al.,2014),maximum canopy storage(Smax;mm)was estimated based on the LAI using equation(Smax(crops)=0.935+0.498 LAI-0.00575 LAI2or Smax(eucalypt)=0.0918 LAI1.04),cohesion by plant roots was taken from the literature(Morgan,Quinton,Smith,R.E.,Govers,Poesen,Auerswald,Chisci,Torri,Styczen,&Folly,1998;Morgan&Rickson,2005),aggregate stability from McCalla(1944),modified by Imeson and Vis(1984),saturated hydraulic conductivity Ks was estimated with a permeameter(Gubiani,Reinert,Reichert,Gelain,& Minella,2010),and total porosity was determined according to Danielson and Sutherland(1986).

Table 1 LISEM numeric input and output parameters for eucalyptus and grassland watersheds.

Soil cohesion was determined in the field with a Torvane,immediately after a rainfall event,when the soil was at nearsaturation level.Soil particle size was quantified after dispersion(Suzuki,Reichert,Albuquerque,Reinert,& Kaiser,2015)by granulometry analysis,namely sand fraction by sieving,and silt and clay fractions by sedimentation(Klute,1986)to define the median particle size distribution(D50).The tension of the wetting front was estimated(Rawls,Brakesnsiek,& Soni,1983),and Manning’s coefficient n was calculated as in(Haan,Barfield,& Hayes,1993).We note that all parameters were used deterministically,whereby no parameter uncertainty was considered(in essence parameter values were assumed as mean with zero variance).This choice can be considered as justified,given the small size of the watershed of application,but should be reconsidered for applications with larger scale analysis.

Simulations were executed with a temporal resolution of 20s.During calibration,the most sensitive parameters of the model(DeRoo,Wesseling,et al.,1996;De Roo,Offermans,et al.,1996),namely Ks,RR,n and the initial water content,were modified,whereas all other parameters were held constant.Graphical and statistical analyses were used to benchmark measured against predicted values.The PBIAS statistical test was calculated to evaluate model efficiency in estimating runoff volume,sediment yield,maximum streamflow,maximum suspended sediment concentration,and hydrographs and sedimentographs peak times.We chose the Nash-Sutcliffe efficiency(Nash&Sutcliffe,1970)NSE＞0.50 and percent bias(PBIAS)≤25%as standard for acceptable flow simulations,and NSE＞0.50 and percent bias(PBIAS)≤50%for sediments(Moriasi et al.,2007).

The data analysis included the calculation of total precipitation and maximum precipitation intensities,average and maximum flows and hydrograph separation.Subsequently,data related to sediment-related variables were grouped and organized,and finally represented by the concentration of sediments in medium and maximum suspension and total sediment production(Table 1).After tabulation and grouping of hydrological and sedimentological variables,data normalization was applied as a strategy to restructure the tables and attributes,with the aim to reduce redundancies and allowing to resize the database.

3.Results

3.1.Rainfall hydrosedimentological events used for calibration

The 22 rainfall events that generated runoff both in the FW and GW,occurred between August 2012 to September 2013,produced a total runoff volume ranging from 13.3 to 107 mm and were characterized by a maximum rainfall intensity from 1.7 up to 36.9 mm h-1(Tables 2 and 3).For most of the monitored rains,considering the slopes of the regression equations,the pastureoccupied watershed(GW)showed greater runoff and,consequently,greater maximum flow and sediment production,compared with the forest watershed(FW)(Fig.3).

3.1.1.Forest watershed events

Rainfall events at the FW generated runoff(R)ranging from 69.2 up to 5245.0 m3,maximum streamflow(Qmax)from 5.2 up to 345.5 L s-1,and runoff coefficient(C)from 0.1 up to 6.0%.The maximum concentration of suspended sediment varied between 2.3 and 599.6 mg L-1,and the observed suspended sediment yield between 0.0002 and 2.0252 Mg km-2(Table 2).

The highest rainfall intensities generally provided highest runoff,maximum streamflow,sediment concentration,and yield.The rainfall event of November 04,2013 had the highest maximum rainfall intensity(35.8 mm h-1)and total volume(107.2 mm),but did not generate the highest runoff,maximum streamflow,runoff coefficient,total sediment yield,and suspended sediment concentration.

Rainfall events occurring on wet soil,such as those recorded on 09/18/12 and 10/09/12,provided the highest values of maximum streamflow(345.5 and 303.9 L s-1,respectively),runoff(2684.7 m3and 5245.0 m3,respectively),maximum suspended sediment concentration(599.6 and 252.2 mg L-1,respectively),and suspended sediment yield(1.84 and 2.01 Mg km-2,respectively).The maximum suspended sediment concentration observed on 09/18/12 was incurred by the collapse of stream-channel banks,visually observed in the field after the event.

The FW hydrographs showed smooth increases and decreases as a response to rainfall events(Fig.4).In general,the maximum suspended sediment concentration arose after the maximum streamflow,where sedimentographs were characterized by a steep increase and smooth decrease thereafter with the exception of the event which occurred on 09/18/12 and saw a steep increase and decrease in suspended sediment concentration.

3.1.2.Grassland watershed events

The amplitude for runoff was high in GW,ranging from 228.1 up to 31207.4 m3,with a runoff coefficient between 0.8 and 36.0%.Highest rainfall intensities associated with highest rainfall volume were responsible for the highest maximum streamflow(1212.7 and 1878.7 L s-1km-2)and,consequently,for the highest sediment yield(3.17 and 9.08 Mg km-2;Table 3).

Table 2 Hydrosedimentological variables of monitored events in the watershed with eucalyptus.

Table 3 Hydrosedimentological variables of monitored events in the watershed with grassland(GW).

Fig.3.Sediment yield(SY),maximum flow(Qmax)and surface runoff(SR)varying with total rainfall(TR),for watersheds with eucalyptus(FW)and with grassland(GW).

Maximum sediment concentrations ranged from 30.6 to 494.6 mg L-1(Table 3),resulted in suspended sediment yield varying from 0.02 to 9.09 Mg km-2.The sediment yield was dependent on the increase in flow and particle transport capacity,as the storms increased.The intensity of the rain on the wet soil and the consequent runoff volume showed a higher correlation coefficient(r=0.98)than the duration of the rain(r=0.18)or total precipitation(r=0.52).This can be verified in the event on 09/18/12,which presented total precipitation of 87.8 mm and maximum sediment concentration of 599.6 and 223.7 mg L-1for FW and GW,respectively.However,one day before the occurrence of this event,another precipitation event with a total accumulation of 19.0 mm was observed.High-magnitude rainfall events generated elevated maximum streamflow,and hydrograph increase and decrease were considerably steep due to the fast streamflow response to rainfall(Fig.4).In general,asynchronous maximum concentration of suspended sediment was observed prior to maximum streamflow in the GW.

3.1.3.Forest watershed versus grassland watershed

Rainfall effect on hydrological and sedimentological variables was more evident in GW compared with FW.Average runoff was higher in GW(28.3 mm)than in FW(4.76 mm)and average runoff coefficient was 0.9% for FW and 5.9% for GW.The largest recorded maximum streamflow was about six times higher in GW than in FW.The highest suspended sediments concentrations during hydrosedimentological events were higher in GW than in FW(Tables 2 and 3).The shape of the hydrograph was steepest in GW,with fast,sharp increase and decrease,while multiple peaks occurred for some events in GW.The onset of the increase of the hydrograph and maximum streamflow occurs in less time in GW and the fall is faster than in FW.

Fig.4.Precipitation(P),observed(Q obs)and simulated(Q sim)flows,observed suspended sediment(Css obs)and simulated(Css sim)varying with time(T)for the event of day 09/18/2012,calibrated for the forest(FW)and grassland watershed(GW).

Maximum suspended sediment concentration generally arose ahead of the maximum streamflow recordings for GW and after for FW.The increase in suspended sediment concentration was less steep in FW,and remained constant for some events after reaching its maximum value.In GW there was a rapid increase in suspended sediment concentration with streamflow increase,and the sedimentograph decrease was generally steep.

3.2.Rainfall hydrosedimentological events used for model validation

The 14 rainfall events used for model validation generated total rainfall in FW and GW ranging from 28.7 to 48.3 mm,respectively,and maximum precipitation intensity from 1.7 to 36.9 mm h-1(Tables 4 and 5).

3.2.1.Forest watershed events

Surface runoff(SF)in FW varied from 39.8 to 593.9 m3,maximum flow(Qmax)from 15.9 to 157.7 L s-1,and flow coefficient(C)from 0.1 to 1.4.The concentration of suspended sediments ranged from 13.3 to 786.5 mg L-1,and yield of suspended sediment ranged from 0.0025 to 0.1323 Mg km-2(Table 4).

Although the events validated in the months of January and February in FW presented the highest accumulation of precipitation,the event monitored on 09/05/19 had the highest volume of runoff.This event occurred in a period of higher soil moisture,a condition that favors the anticipation of flow formation.The event of 05/27/19 showed total precipitation of 28.7 mm,which provided the highest concentration of suspended sediments and,consequently,the highest sediment yield.These higher concentrations are probably associated with the initial soil moisture condition,which is directly related to the previous event that occurred on 09/05/19,with a total precipitation of 40.1 mm(Table 4).

3.2.2.Grassland watershed events

Surface runoff(SR)in GW ranged from 1041.0 to 7140.4 m3,maximum flow(Qmax)varied from 81.9 to 805.6 L-1,and flow coefficient(C)from 3.1 to 13.0.The concentration of suspended sediments ranged from 85.6 to 383.5 mg L-1and the production of suspended sediments ranged from 0.3789 to 3.0193 Mg km-2(Table 5).

Higher intensities of precipitation generally provide higher runoff,maximum flows,concentration of suspended sediments and sediment yield.The event on the 01/17/19 presented greater accumulated precipitation,which resulted in higher runoff(7140.4 m3)and sediment yield(3.0193 Mg km-2).This behavior may be associated with the antecedent event that occurred on 12/01/19,with a total precipitation of 28.9 mm,which possibly contributed to the higher antecedent soil moisture.In general,GWshowed higher flows than FW(Table 5).

Table 4 Hydrosedimentological variables of events validated in the watershed with eucalyptus(FW).

Table 5 Hydrosedimentological variables of events validated in the watershed with Grassland(GW).

3.2.3.Forest watershed versus grassland watershed

The effect of precipitation on water and solid discharge from events validated by the model was greatest in GW.The average flow was higher in GW(3218.7 m3)than in FW(257.3 m3),and the average flow coefficient was 0.8% for FW and 7.9% for GW.The highest maximum flow was about twelve times larger in GW than in FW.The highest concentrations of suspended sediment from hydrosedimentological events used in the validation were higher in GW,approximately 45%,than in FW.On the other hand,the average sediment production was 137%higher in GW than in FW(Tables 4 and 5).

3.3.Hydrosedimentologic modeling

3.3.1.Model calibration

For some input parameters,changes in values were required for all simulated events.For the sediment variables,the adjustments made to the input values were to attenuate peak oscillations in the sedimentographs.The altered parameters for the FW modeling were soil saturated hydraulic conductivity,initial soil moisture,and Manning roughness coefficient,with percentage decreases in relation to the input values of 88.3%,34% and 74.4%,respectively.For the GW,these parameters were decreased by 92.1%,4.5% and 8.0%,respectively.The changes in initial water content and saturated hydraulic conductivity influenced the maximum flow rate and the peak flow,providing a reduction in the simulated maximum flow for both watersheds.Of the calibrated events(Fig.4)to represent surface runoff in the two watersheds,five events presented statistical indicators classified as good in terms of the NSE(NSE＞0.75;Table 6).

For the sedimentogram representation of the same rainfall events,the NSE coefficient presented unsatisfactory results.In GW,the percentage bias(PBIAS)indicator presented acceptable results(PBIAS≤50%)for sediment yield in three rainfall events;seven rainfall events showed acceptable results for peak time of the simulated sedimentograms.For FW,the PBIAS percentage showed acceptable results for sediment yield in one precipitation event,and for sedimentograms peak time in two other events(Table 7).

3.3.2.Uncalibrated rainfall events

The hydrographs of the ten events measured in FW and seven in GW were not satisfactorily calibrated by the model.In view of LISEM’s difficulty in representing the runoff for this group of events,we did not simulate the concentration of suspended sediments for these events.

The statistical performance for the simulated and observed hydrograph,determined by the NSE,varied between-0.01 and-29.3 in FW and between 0.19 and-9.23 in GW,representing unsatisfactory values.The PBIAS for the maximum flow rate(Qmax)was between 258.9% and-46.5% in FW and 356.2% and-69.3% in GW.The comparison of the peak time of the simulated and measured hydrograph showed a variation between 162.5%and 4.6%for FW.However,for GW,the variation of peak times between events presented a smaller amplitude,between 110.0% and-2.8%.The greatest variation in estimating surface flow volume occurred in FW,between 108.5%and 96.6%,whereas in GW,where the flow rates were higher,the variation was lower,roughly between 70.8%and-64.7%(Table 8).

3.3.3.Model validation

Six events in the watershed with eucalyptus(FW)and eight events in the watershed with grasslands(GW)were validated,using a set of events different from those used in the calibration process,but with similar characteristics,such as total precipitation and rainfall intensity,flows,peak times and sediment production.For validation,the output parameters of the events calibrated in the years 2012 and 2013 were applied,as described in section 3.2.1.The statistical results of the validation were determined by the NSE index to compare the shape of the observed and simulated hydrographs and through the percentage bias(PBIAS)for maximum flow,peak time,and flow(Table 9).The events validated in FW and GW represent total precipitation between 26.9 and 48.3 mm.Fig.5 show the rain events with the best model validation performance,for FW and GW,respectively.

Of the set of six events validated for the FW,all presented an NSE index classified as“good”and“very good”.The event measured on 09/09/19,with a total precipitation of 39.9 mm,presented the best NSE validation index,classified as“very good”.Of the set of eight rain events validated for GW,seven had an NSE index classified as“good”and“very good”.The events measured on 02/15/19 and 03/16/19 showed an almost perfect fit for the shape of the measured and simulated hydrograph,with an NSE index of 0.96 and 0.93,respectively.

The sedimentological parameters validated by the model were concentration of sediments in maximum suspension and production of sediments.The statistical results of the validation were determined by the NSE index for observed and simulated sediment production and by means of the bias percentage(PBIAS)(Table 8).Only two events measured in GW,on the 04/04/19 and 05/27/19,presented satisfactory statistical results for the production of sediment.In general,most of the simulated sedimentograms in FW and GW showed unsatisfactory results,that is,they were not properly validated(Table 10).

Table 6 Observed and simulated values and statistical performance of the LISEM model for the simulation of surface runoff,peak flow,and peak time for rainfall events in the eucalyptus and grassland watersheds.

Table 7 Observed and simulated values and statistical performance of the LISEM model for simulating of sediment yield,maximum suspended sediment concentration and peak time events in the eucalyptus and grassland watersheds.

Table 8 Observed and simulated values and statistical performance of the LISEM model for the unsatisfactorily calibrated events for the simulation of surface runoff and peak time for precipitation events in the eucalyptus and pasture watersheds.

Table 9 Validated values and LISEM statistical performance for simulating runoff,peak flow and peak time for rain events in eucalyptus and grassland watersheds.

Table 10 Statistical performance for validating variables related to sediment production.

4.Discussion

4.1.Hydrosedimentological events

The highest flow values observed in the study areas were associated with total precipitation.The effect of rainfall events on hydrological and sedimentological variables was more pronounced in GW.This behavior was also shown when consecutive rainfall events occurred,especially in the winter-spring transition period,when soil water content remains high for several days and thus rapidly saturating the soil for runoff generation.Rodrigues et al.(2014)reported the importance of the initial soil moisture on runoff formation,where low total rainfall associated with low maximum rainfall intensity was responsible for the highest maximum streamflow and surface runoff in the watershed,because of rapid soil saturation.When a given volume of rain generates greater runoff volume when compared with larger rainfall,the time interval between the previous rain and the effect of the initial soil moisture should be taken into account.The initial moisture is a controlling variable of hydrological processes and is highly variable in space and time.Furthermore,this variable is influenced by vegetation,land use,topography and climate(Su & Shangguan,2019).The preceding high moisture condition of the soil promotes a significant decrease in the saturated hydraulic conductivity.In other words,the hydraulic and structural properties of the soil during the precipitations govern the occurrence of erosion and the formation of runoff(Zambon et al.,2021),justifying the higher values of maximum flow and sediment yield that occur on wet soil.

Maximum flow was dependent on total precipitation.This condition is because of event characteristics,i.e.,when rainfall intensity is higher than water infiltration rate in the soil,runoff is favored and provides higher maximum streamflow.Although the FW(0.83 km2)is smaller than the GW(1.10 km2),responses to high rainfall arose slower and with lower maximum streamflow in FW,which may be related to vegetation cover with eucalyptus,a condition that contributes to the increase in soil water infiltration,as observed in watersheds with eucalyptus stands(Reichert et al.,2017;Rodrigues et al.,2014).Moreover,rainfall is intercepted by the canopy and litter(Reichert et al.,2017),where intercepted rainfall flows along tree trunks and reaches the soil surface with less energy to disaggregate and transport soil particles.By contrast,the highest maximum streamflow in GW indicates a lower capacity for the watershed to retain the rainfall volume of intense rainfall events.Didoné et al.(2014)observed that soil management systems employed in grassland watershed in southern Brazil were ineffective in reducing runoff and erosion caused by intense rainfall events.

The effect of precipitation on water discharges was least in FWGW.This result may be related to the possible benefits of forest crops to the infiltration dynamics and consequent reduction of runoff,mitigating the occurrence of soil erosion and loss of surface water(Ferraz,Rodrigues,Garcia,Alvares,&Lima,2019).The Pampa biome is characterized by smoothly wavy areas,where hydrology takes on a particular dynamic,especially in view of the agricultural intensification that promotes the most intense hydrological responses,promoting accelerated flow and accumulation of sediments generated by preferential paths,which may justify the greater GW discharges(Ares,Varni,& Chagas,2020).

Fig.5.Precipitation(P),observed(Q obs)and simulated(Q sim)flows,observed sediment yield(SY obs)and simulated(SY sim)varying with time(T)for the event of day 01/17/2019,validated for the forest(FW)and grassland watershed(GW).

Suspended sediment concentration and sediment yield,closely dependent on streamflow,were generally higher in GW than in FW.Reduced sediment yield indicates an improvement in soil management(Minella,Merten,Walling,&Reichert,2009).The highest sediment concentration and yield is possibly a result of reduced interception and dissipation of rainfall energy by the degraded grassland and to soil surface exposure to raindrop impact of tilled soil for pasture seeding.Furthermore,wet areas adjacent to creeks favor runoff due to low infiltration conditions,even if rainfall intensity is lower than the soil infiltration capacity,and lateral flow may occur.

Maximum concentration of suspended sediment taking place after maximum streamflow was more evident in FW.When the former happens after the maximum streamflow,the sedimentograph indicates the sediment source is at a larger distance from the stream channel,such as sediments mobilized in the watershed and then transferred by runoff(Minella,Merten,Reichert,&dos Santos,2007).When the maximum suspended sediment concentration precedes the maximum streamflow,as observed in GW,sediments are rapidly mobilized,transported,and deposited(Brito,2006;Minella,Merten,& Magnago,2011;Minella,Merten,& Reichert,2007;Pinheiro,Kaufmann,Schneiders,& Gotardo,2013;Pires,Naves Silva,Curi,Leite & de Figueiredo Brito,2006;Rodrigues et al.,2014;Rodríguez & Camargo,2009).Sediments near or inside the stream are mobilized,respectively,by surface runoff on the landscape or by streamflow in the creek.Flow connectivity and particle size probably explain major changes in sediment source apportionment(Rodrigues et al.,2018).

Absence of a significant increase in suspended sediment concentration with increasing streamflow after the highest rainfall intensity is possibly due to transport capacity and the exhaustion of the readily available sediments,which were transported at rainfall inception(Minella et al.,2011;Rodrigues et al.,2014),with continuous exhaustion of the sediment supply in subsequent rainfalls(Minella et al.,2009).

High flow rates,suspended sediment concentration and sediment production in GW and FW were probably influenced by the occurrence of moderate to high intensity El Niño climate phenomenon in southern Brazil between 2009 and 2016(Berlato &Cordeiro,2018;Kayano,Andreoli,Souza,Garcia,& Calheiros,2016;Matzenauer,Radin,& Maluf,2018).Furthermore,during this period some eucalyptus stands were cut in 21.4% of the FW area,followed by replanting with Eucalyptus saligna species,promoting greater soil exposure to runoff and erosion processes.

The runoff coefficient in GW was significantly higher than in FW,demonstrating land use has a strong influence on the hydrological response.Similarly,Sriwongsitanon and Taesombat(2011)demonstrated lower runoff coefficient in forest area is than in non-forest area,due to greater soil moisture storage capacity and evapotranspiration in the forested area.

4.2.Hydrosedimentological modeling with the LISEM

4.2.1.Calibration

The changes in saturated hydraulic conductivity output values in the model in relation to the input values can be justified by the difficulty in spatially representing soil properties due to spatial variability,thus influencing infiltration predictions(Weiss &Gulliver,2015).In modeling runoff in eucalyptus forest areas(Mello,Lima,&Silva,2007),observed an increase in flow when the soil moisture increased slightly,similar to that observed in most calibrated events in our watersheds.

Although LISEM performed well in representing the shape of the hydrograph and the hydrological variables for low and high magnitude events,it was not possible to obtain a single calibrated set of parameters for all events.Each event had to be calibrated and validated separately(new parameter values),because the magnitude of the flow and the shape of the hydrograph depend on the intensity of the rain,previous soil moisture and water infiltration in the soil,as demonstrated by(Karssenberg(2006)and Léonard,Ancelin,Ludwig,and Richard(2006).In addition,the dynamics of water infiltration in the soil column changes during wetting and saturation cycles,thus changing the intensity of the water repellency phenomenon and the water infiltration in native grasslands soils(Vogelmann,Reichert,Prevedello,Awe,&Cerdà,2017),which probably explains the need to adjust the previous soil moisture in the calibration process of each event individually.

The wide range of NSE values for erosion indicate the difficulty in modeling the erosion process for the selected study areas.The simulations indicate that the equation that represents the erosion process was not sensitive to changes in aggregate stability.We tried to calibrate the parameters related to sediment production by changing the D50sediment and the cohesion of the soil and root,but changing these parameters did not show much sensitivity.

Another drawback found in the calibration is that LISEM does not consider sediment readily available for the drainage channel or the sediment exhaustion.This means that the hysteresis between the peak of the sediment and the peak of the flow is not simulated correctly,that is,the model makes the two peaks artificially synchronous.

4.2.2.Validation

The LISEM model provides a simplified representation of the processes,which is directly dependent on the availability of observations,knowledge of the phenomena involved,computational capacity and,mainly,application purposes.There are few modeling studies presenting the validation component for rain events using LISEM.However,the validation process is essential in providing an assessment of the adequacy of the model to represent real-world hydrological processes(Biondi,Freni,Iacobellis,Mascaro,&Montanari,2012).

Using the set of output parameters from the calibration process,we validated fourteen rainfall events for FW and GW,all with results of satisfactory statistical performance for the hydrology component.On the other hand,for the component related to sediment yield,only two GW events showed statistical results within acceptable limits.Our study shows LISEM’s limitations in calibrating and validating the sediment yield for individual events.Thus,our results corroborate with the results obtained by Barros(2012),who observed limitations of LISEM to validate erosion processes due to the great spatial variability of factors that control erosion,such as soil detachment,transport sediment,deposition and location of sediment sources.Moreover,De Roo and Jetten(1999)found that the spatial and temporal variability in initial soil water content and hydraulic conductivity in the watershed are reasons for limitations in calibration and validation processes,mainly related to the erosion aspect.The difficulty in modeling the production of sediments in watershed scale may also be related to the need to understand the dynamics of potential changes in the flow and load of sediments.These changes may be related to variations in land use and spatial variability scenarios,changing the distribution of average sediment production(Sharafati,Pezeshki,Shahid,& Motta,2020;Shrestha,Cochrane,Caruso,& Arias,2018).Furthermore,climate variability,such as wet and dry periods,can alter the initial soil moisture(Fortesa,Latron,García-Comendador,Company,& Estrany,2020)impacting the flow formation dynamics during rain events(Verma et al.,2020).These factors can be intensified by the effect of topography,watershed morphology and representativeness of the model’s parameters(Gull&Shah,2020).The performance of sediment modeling is also dependent on uncertainties in terms of location and spatial distribution of potential sediment sources(Rodrigues et al.,2018)and particle deposition(Wilkes et al.,2019)that may promote an increase or decrease in peak of sediments(Ares et al.,2020)in different rainfall events,being characterized as one of the biggest difficulties of LISEM in representing mathematically the production of sediments.

4.3.Implications for land management

Eucalyptus plantations,besides being of importance for the production of raw materials like paper,packaging and lumber for civil construction,have important implications for land management,which can directly impact on hydrological and erosive dynamics.Land occupied by eucalyptus forests is growing significantly in several parts of the word.Brazil,Argentina and Chile comprise the largest areas of eucalyptus plantations in South America(Mattos,de Oliveira,Lucas,&Wendland,2019).In addition to the implications for ecosystems and historical species,land occupied by eucalyptus plantations still raise concerns on the hydrological behavior of forested watersheds,related to surface water flow,sediment transport,soil erosion,soil water retention,and soil fertility(Mhiret et al.,2019).

Brazil is recognized worldwide for its vast forest cultivation covering millions of hectares,representing an option to supply the commercial demand for wood(Colodette,Gomes,Gomes,&Cabral,2014;Ferraz et al.,2019).However,it is emphasized that forestry practices generally influence water quality when compared to other land uses(Brown,Western,McMahon,& Zhang,2013;Rodrigues,Taniwaki,Lane,de Paula Lima,&de Barros Ferraz,2019),due to the different dynamics in flow and sediment transport,where planted forests,native or exotic,are able to protect the soil and minimize the effects of sediment exports.Forests have high rainfall interception by because of canopy,promoting increased infiltration and,consequently,reducing surface runoff and sediment exports(Ogasawara,Santos,Cassiano,Wemple,& Ferraz,2020;Wichert,Alvares,Arthur,& Stape,2018).On the other hand,in forest watersheds with a high density of unpaved roads,when land management practices are inadequate,there might be greater sediment exports due to the increase in vehicle circulation during the forest harvesting.In addition,unpaved roads can represent adverse effects to water courses due to increased flow and sediment discharge.Therefore,the construction of forest roads requires proper planning,by considering land slope and hydrological connectivity(Arismendi et al.,2017;Erdem,Enez,Demir,&Sariyildiz,2018).

It is of utmost importance to adopt strategies to understand and mitigate the impacts of agricultural practices on erosion and runoff in watersheds occupied with eucalyptus plantations and pastures for livestock,that are representative of various regions of southern Brazil in terms of climate,soil classes,geomorphology,and land use.These strategies can be based on the use of robust monitoring combined with modeling techniques,as demonstrated in our study.We are aware that the length of the investigation can impact the results of this study,but we believe it contributes to instigate opportunities for new research.

5.Conclusions

Rainfall events studied in the calibration stage produced maximum flow rate and sediment yield up to six times larger in GW than in FW.Higher concentration and yield of sediments in GW are possibly related to reduced rain by the degraded grasslands and because of soil surface exposure.Although the FW is smaller than the GW,the hydrologic responses to intense rainfall were slower and with a lower maximum flow rate in FW,which may be related to the eucalyptus vegetation cover,a condition that contributes to increasing water infiltration in the soil and decreasing runoff generation.Modeling with LISEM showed that the results for the calibration of the runoff were acceptable for 22 of the 37 simulated events in FW and GW.The averages of the output parameters for each calibrated event differed from the average input values in the model.The calibrated output parameters allowed validating a set of events from a different period,where the hydrological response presented satisfactory statistical performance.

In the validation stage,the effect of rain on water and solid discharge from events validated by the model was greater in GW than in FW.The highest concentrations of sediment in suspension of hydrosedimentological events used in the validation were 45%higher in GW than in FW.All validated events showed results of satisfactory statistical performance for the hydrology component,GWbut only two events for the erosion component.This result denotes limitations of LISEM in calibrating and validating the sediment yield due to the spatial variability of factors that control the dynamics of flow and suspended sediment load,such as variations in land use and scenarios of spatial variability,changing the distribution of average sediment yield.Moreover,climatic variability alter the initial soil moisture,impacting the flow formation dynamics in all rain events.The performance of erosion modeling also depends on uncertainties in terms of location and spatial distribution of potential sources and deposition of sediments in different rainfall events,being characterized as one of the greatest difficulties of LISEM in representing mathematically the erosive aspect of watersheds.

The validated LISEM model for hydrology is a useful tool to evaluate soil and crop scenarios in watersheds cultivated with commercial forest and grassland for beef cattle,both activities of significant economic importance.

Acknowledgments

This study was financed in part by the“Coordenação de Aperfeiçoamento de Pessoal de Nível Superior”(CAPES)-Finance Code 001,Fapergs(“Fundação de Amparoà Pesquisa do Estado do Rio Grande do Sul”),and CMPC Celulose Riograndense.JMR thanks CNPq(Brazilian Council for Scientific and Technological Development)for the research fellowship.