Pitro Mnzs Snhz Mo ,Pulo Trso Snhs Olivir ,Muro Antonio Homm Antuns ,Vlmir Luio Durigon ,Elin Cristin Croso Filgo ,Dnil Fons Crvlho ,*

a Agronomy-Soil Science Post Graduate Program,Federal Rural University of Rio de Janeiro,BR 465,km 7,23897-000,Seropédica,RJ,Brazil

b Federal University of Mato Grosso do Sul,CxP.549,Campo Grande,MS,79070-900,Brazil

c Federal Rural University of Rio de Janeiro,Engineering Department,BR 465,km 7,23897-000,Seropédica,RJ,Brazil

d Technical College of the Federal Rural University of Rio de Janeiro,BR 465,km 7,23897-000,Seropédica,RJ,Brazil

e National Center for Soil Research,Brazilian Agricultural Research Corporation,Rio de Janeiro,Brazil

Keywords:Soil erosion Land cover Soil use and management factor Vegetation seasonality NDVI

A B S T R A C T We present a new approach for calculating the C-factor of RUSLE considering the effect of low-reflectance vegetation cover areas on the reduction of the effects on erosion caused by rainfall seasonality.For this,we propose the coefficients Cr2(rescaled 2)and C-PC(Precipitation Correction),which represent the Cfactor,and an adaptation in NDVI calculation,according to the seasonality of precipitation(NDVI-PC).The Cr2 factor is used when there is no seasonal effect of rainfall on vegetation,while the C-PC factor is calculated for localities under the influence of seasonality,from NDVI-PC.The proposed approaches were tested using different satellites images in the Palmares-Ribeirão do Saco watershed,Rio de Janeiro,Brazil.The values of Cr2 and C-PC factors were compared to the Cr factor(rescaled)and to mean values from the literature for different land covers.Our results indicated that the Cr2 factor represents an improvement in accuracy in relation to Cr by considering specific values of the studied area to normalize the data without generalizations.Furthermore,the C-PC factor is able to simulate the effect of seasonality,providing more realistics values of soil loss by the RUSLE as a function of the proportion of area affected by the rainfall seasonality obtained from NDVI-PC.We conclude that both Cr2 and C-PC factors generate values similar of the C-factor observed in the literature,and therefore are able to provide better soil loss estimation than that using the Cr factor.

1.Introduction

Remote sensing data have been used in studies on water erosion in watersheds,enabling the development of solutions to problems of soil and water resources degradation(Alewell,Borrelli,Meusburger,& Panagos,2019;Benavidez et al.,2018).One of the most used methodologies in remote sensing is the application of vegetation indices,which possibilities a quali-quantitative analysis analises of growth dynamics and vigour of vegetation cover.There are several models that make it possible to study soil cover as the NDVI(Normalized Difference Vegetation Index)(Durigon et al.,2014;Van Der Knijff et al.,2000),NDII(Normalized Difference Infrared Index)(Joiner et al.,2018),NDWI(Normalized Difference Water Index)(Guha et al.,2020),EVI(Enhanced Vegetation Index)(Gurung et al.,2009)and SAVI(Soil Adjusted Vegetation Index)(Huete,1988),among others.NDVI is the most commonly used and it is more suitable for the study of natural vegetation(Bannari et al.,1995).It enables the evaluation of spatial and temporal changes occurring on the Earth’s surface associated with reflectance in the red and near infrared bands of the electromagnetic spectrum(Rouse et al.,1974,p.371).

Several studies use spatial data in models that estimate soil loss(Batista et al.,2019)in watersheds but,depending on their characteristics,the results obtained differ from reality due to limitations and uncertainties associated with the models(Chavez,2010).Both in USLE(Universal Soil Loss Equation)(Wischmeier &Smith,1978)and in its revised version(RUSLE)(Renard et al.,1997)do not include factors as sediment deposition along the slopes(Cecílio et al.,2009)and the presence of dry vegetation,affected by the effect rainfall seasonality,as an agent capable of reducing soil loss.

The C-factor represents the characteristics of soil management and cover in RUSLE and is usually obtained in watersheds studies,through data from the literature,by using NDVI or in a hybrid manner,including remote sensing techniques associated with field studies.The obtaining of C-factor through satellite images has been studied because it is one of the most significant sources of uncertainty in the model(Almagro et al.,2019;Durigon et al.,2014;Phinzi & Ngetar,2018;Tanyas et al.,2015).In this context,images provided by the TM/Landsat 5,OLI/Landsat 8 and LISS3/Resourcesat-1 sensors represent a good option for vegetation studies in watersheds(Phinzi & Ngetar,2018).

In vegetation targets under water deficit,NDVI may assume values close to zero,thus it can be associate with bare soil targets.Besides affecting cultivated areas and pastures,this response of the electromagnetic spectrum can occur in places with predominance of larger species,especially those with thin crowns or deciduous,where in many cases the understory vegetation,which is more susceptible to climatic variations,affects the values of NDVI(Pisek et al.,2015;Schmidt et al.,2018;Silva & Sano,2016).

Incorrect values of soil loss can be obtained depending on how rainfall seasonality influences vegetation cover due to a physiological issue of each species,which shows a reduction in leaf water content during drought periods,affecting biomass and,consequently,its reflectance(Johansen and Tøvikmmer,2014).This aspect makes it possible to infer that rainfall seasonality has an influence on the NDVI response and that it can be corrected to generate a C-factor that differentiates areas with bare soil from areas with dry vegetation.This procedure would reduce uncertainty in soil loss estimates by RUSLE,since the model disregards the effect of rainfall seasonality(Benavidez et al.,2018).From the correlation between amount of rainfall and vegetation reflectance it is possible to infer the effect of seasonality on soil loss quantitatively.This correlation also varies according to the type and physiological characteristics of the plant species that serve as land cover(Chamaille-Jammes et al.,2006;Fabricante et al.,2009;Nanzad et al.,2019;Richard&Poccard,2010).However,to the best of our knowledge,there is no approach considering the rainfall seasonality on the C-factor estimation by using vegetation indexes.

Considering the hypothesis that the presence of low-reflectance vegetation can influence the soil erosion process,we propose a new approach for calculating the C-factor of RUSLE,that take into account the effect of low-reflectance vegetation cover areas on the reduction of the effects on soil erosion caused by rainfall seasonality.Thus,we proposed the coefficients Cr2 and C-PC,which represent the C factor,and an adaptation in NDVI calculation,according to the rainfall seasonality in the study region(NDVI-PC).

2.Material and methods

2.1.Seasonality-conditioning variables(pptx and Lv)

To define the influence of rainfall seasonality in vegetation reflectance from the satellite images,we used the variables:i.Pptx which corresponds to the total precipitated in“x”days before the passage of the satellite;and Lv,which represents the average precipitation of the day accumulated in periods of“x”days in the rainfall series studied.The“x”variable represents the period of total rainfall precipitated with the best correlation with NDVI values in 0-0.5 class,which corresponds to areas interpreted as bare soil and low-reflectance vegetation.

When Lv is lower than or equal to Pptx,the rainfall from the period preceding the satellite passage is equal to or higher than the historical average of the region and,therefore,it is considered that there is less presence of dry vegetation with low reflectance(Fig.1).In this case,the Cr2 factor(Eq.(1))should be used in the calculation of RUSLE.On the other hand,if Lv is greater than Pptx,the average rainfall in the historical series is higher than the rainfall of the period preceding the evaluated image and,therefore,the vegetation is expected to be drier because of seasonality.Thus,the C-PC factor should be used(Eq.(4)).

2.2.C factor rescaled(Cr2)

The Cr2 factor is an adaptation of the Cr factor presented by Durigon et al.(2014)and enables a specific adjustment in the conversion of NDVI into C-factor for each image(Eq.(2)).With this adaptation,the cover index adequately fits to the range between 0.0 and 1.0(saturated variation),in which values close to 0.0 are associated with vegetation,and values close to 1.0 are associated with bare soil.

where z is the variable that represents the maximum pixel value of NDVI.

2.3.Modification of NDVI

A modification in NDVI,called NDVI-PC(Normalized Difference Vegetation Index-precipitation correction)(Eq.(2)),is suggested in order to enable the identification of areas with low-reflectance vegetation in periods of absence of rainfall.Its main function is to increase the NDVI values as a function of precipitation,allowing the reclassification of dry vegetation targets erroneously considered as bare soil(Fig.2).

where.

ρnir-spectral reflectance of the surface in the near infrared band(dimensionless);

ρred-spectral reflectance of the surface in the red band(dimensionless);

Pptx-accumulated precipitation in the x period prior to the date of the image used in the calculation of the NDVI(mm);

Lv-leveling variable,equal to the average accumulated precipitation in periods of x days in the studied time series(mm).

The response of dry vegetation in NDVI-PC,with the reduction in near infrared values and increase in the red band,enables the separation of pixels with a higher percentage of vegetation with low values of NDVI.This vegetation is able to reduce the impact of rainfall on the soil,protecting it from erosion.

2.4.C-PC factor

The C-PC factor(Eq.(3))simulates the proportions of vegetation areas subject to reflectance changes due to rainfall regimes.This proposal considers dry vegetation as a cover capable of reducing the impact of raindrops when reaching the soil,considering the vegetation with lower reflectance in the near infrared during the conversion from NDVI to C-factor.

Fig.1.Representative flowchart of the steps to obtain C-factor by the Cr2 and C-PC method.

Fig.2.Example of the functioning and classification of NDVI(a)and NDVI-PC(b)values,indicating the separation of the vegetation class affected by the effect of rainfall seasonality.

where H is the percentage of pixel area in situation of low NDVI due to seasonality(Eq.(4)).

The H variable is calculated for NDVI within the 0.3-0.5 class,which represents the percentage of area occupied by vegetation affected by the drought season,that is,by the seasonality effect(Fig.2B).The final result of C-PC factor is attributed to the state of vegetation as a function of time that is characterized by the Pptx/Lv ratio in the affected physical space(H variable).The multiplication of the Cr2 factor by(Pptx/Lv)Htends to reduce its values,since its calculation is performed under the condition of Pptxbeing lower than Lv(Fig.1).

The variation in Cr,Cr2 and C-PC factors was evaluated considering H=1(equivalent to an area whose soil cover is represented by 100%of affected vegetation of reduced NDVI)and Lv of 5.0 mm.The C-PC factor was also simulated through variations in H values,keeping Pptxat 1.0 mm and subsequently varying Pptxbelow 5.0 mm with fixed values of H(1.0).In both simulations,Lv was equal to 5.0 mm.

2.5.Study area

The Palmares-Ribeirão do Saco watershed(22°30′33′′and 22°22′53′′S and 43°30′15′′and 43°20’50”W)was used as a case study to evaluate the factors Cr,Cr2 and C-PC.The area is located in the Paraíba do Sul river basin,considered one of the regions most affected by water erosion in Brazil despite its environmental and social relevance(Durigon et al.,2014).

The total watershed area is 85.85 km2where 71%,17%and 12%of the area has a uniform,convex and concave shape,respectively.The compactness coefficient is 1.63 and a form factor is 0.31,indicating a low tendency to flood peaks.The average altitude in the region is 780.43 m,varying from 1210 m to 580 m(Durigon et al.,2014).The relief is very rugged,with 2.21% of area classified as flat(slope 0-3%),6.14%as smoothly wavy(slope 3-8%),17.48%as wavy(slope 8-20%),42.79% as strongly undulating(slope 20-45%),28.5% as mountainous(slope 45-75%)and 2.88% as strongly mountainous(slope＞75%).The region has predominantly pasture cover(68.2%)and environmental preservation area at specific points.It has also forests(13.5%),tropical supernifolia above 750 m of altitude and sucaducifolia vegetation in the others,capoeiras(4.4%),planted forests eucalyptus(4.4%)and cultivated areas with short cycle crops(3.4%).The urbanized area is only 5.7% of watershed.

The climate is Cw according to the climate classification of Köppen-Geiger(Durigon et al.,2014).The average annual rainfall is 1100 mm,with values ranging from 1700 mm in rainier years and around 740 mm in dryer years.December and January correspond to the rainy period(200 mm)while July and August is the dry period(19.8 mm).The highest and lowest average temperatures are recorded,respectively,from January to March(30°C)and June to August(12°C),with a slightly varying thermal amplitude.

In the studied watershed,the Oxisols(Latossolos)appears in places of high altitude,Acrisol(Argissolos)and Inceptsol(Cambissolos)in steep slope areas,and Inceptsol(Cambissolo),Acrisol(Argissolo)and Gleyssoil(Gleissolo)in the lower parts of the landscape.Due to the characteristics of relief and erodibility,this watershed has a high risk and natural potential for soil erosion,where approximately 98.5%of the basin area located in the risk of erosion class which amounts to values greater than 10 t ha-1mm-1(Durigon,2011).

Land cover and management in the studied area were characterized using images from TM/Landsat 5,OLI/Landsat 8,REIS/RapidEye,ILSS3/Resourcesat-1 sensors from 2009 to 2015(Table 1).The images were subjected to atmospheric corrections through the radiative transfer program 6 S(Second Simulation of Satellite Signal in the Solar Spectrum)(Antunes et al.,2012)(Datum WGS84)and cropped at the edge of the basin.In the field,points were recorded with a GPS(Garmin/GPSMAP 62)in January 2016 for the georeferencing of images(except OLI/Landsat 8,which are already orthorectified when obtained)and visual investigation of the presence of erosion and vegetation under water deficit that characterized the seasonal effect.

The daily precipitation data used to obtain the Cr2 factor and CPC factor were obtained during 2009 thought 2015 by the PESAGRO weather station,located in the district of Avelar,municipality of Paty do Alferes-RJ,Brazil.In order to define Pptxand the mean period of Lv,the precipitation periods of 365,180,30 and 15 days were correlated with the NDVI values in the class from 0 to 0.5.The best correlation obtained in the study area corresponds to the total precipitated in the fifteen days preceding the date of satellite passage(r=0.64),therefore“x”is equal to 15 days.For the case study,the Lv used was 4.19 mm and in cases where Pptxwas equal to 0.0,a value of 0.1 was adopted so that the calculation of C-PC factor(Eq.(3))could be performed.

The ideal range of 0.3-0.5 used as a parameter to obtain H was defined from a test with different NDVI responses for three identified targets with the presence of bare soil and identified with presence of vegetation under effect of seasonality.For this,NDVI of the OLI/Landsat 8 image of 05/22/2016 was calculated and compared with the index calculated for Google Earth Pro image of 05/29/2016,which provided visual details,due to its high spatial resolution.In each selected target,the means and standard deviations of the reflectance values of 9 pixels were used.NDVI values of the central pixels of the coordinates selected as a function of time were also compared using the images in Table 1.These results were subjected to correlation with rainfall data from the 15-day period prior to image acquisition and the respective erosivity index(EI30)obtained by the equations used by Carvalho et al.(2014)(Table 2).Once the a range of values was justified and established,the H variable was obtained using equation(4).

Table 1 Images selected to be used in soil loss estimation in the period from 2009 to 2015.

Table 2 Equations used to obtain the average monthly erosivity index(EI30)in each 15-day period from the sum of precipitation of each 15-day period(p),in mm.

2.6.Difference in soil cover and comparison of results

As a criterion of seasonality effect,the variation in the response of pixels for the range from 0.3 to 0.5 of NDVI and NDVI-PC was adopted.To verify differences in soil cover status over time,the time series data applied to the Cr,Cr2 and C-CP factors were subjected to descriptive analysis,Shapiro-Wilk normality test and Kruskal-Wallis analysis of variance(p<0.05).The results of Cfactor obtained from the literature,the Cr factor and the integration between the C-PC factor and the Cr2 factor in the time series were also compared.In the literature,the following values wereconsidered:0.02 for Atlantic Forest and pastures;0.51 for cultivated areas without conservation techniques(considering intense management and plowing down the hill);and 1.0 for urban areas and bare soil(Almagro et al.,2019;Oliveira et al.,2015).The value 1.0 was attributed to urban areas so that there were no discrepant results in comparison with the others.

The annual soil loss in the Palmares-Ribeirão do Saco watershed from 2009 to 2015 was estimated by USLE/RUSLE(Eq.(5))from the Cr2 and C-PC factors associated with each image in Table 1,considering them independent of each other.

Table 3 Result of the subtraction of the classes with greatest influence of the seasonality effect between 0 and 0.5 of the NDVI by NDVI-PC.

where.

SL-average annual soil loss(t ha-1year-1);

R-rainfall erosivity factor(MJ mm ha-1h-1year-1);

K-soil erodibility factor(t ha h MJ-1ha-1mm-1);

L-slope length factor(dimensionless);

S-slope steepness factor(dimensionless);

C-soil use and management factor(dimensionless);

P-conservation practices factor(dimensionless).

The variables K,L and S were generated from the same maps used by Durigon et al.(2014)and Carvalho et al.(2014).We assigned a value of 1.0 to the P-factor assuming the non-existence of conservation practices in the study area.The digital elevation model(MDE)used to generate the LS maps was developed at a scale of 1:50,000,while the spatial resolution used in the images was 23.5 m for LISS3/Resoursesat 1 and 30 m for TM/Landsat images.5 and OLI/Landsat 8.

3.Results and discussion

Of the seven images used,only the one of June 05,2010 and September 05,2011 generated NDVI-PC,because only for them Ppt15was lower than the leveling variable(Lv).The displacement of values between NDVI-PC classes in relation to NDVI(Fig.3)in the image of June 05,2010 was higher than that of September 05,2011 as the values of Ppt15were 1.7 and 3.4 mm,respectively.The low rainfall in the period suggests that there is presence of vegetation affected by seasonality,reducing the reflectance in the near infrared band for these targets.

The greatest differences of the subtraction between NDVI and NDVI-PC are presented in the classes 0.3-0.4 and 0.4-0.5 for both images(Table 3),so the 0.3-0.5 range is considered the one that best represents the vegetation affected by drought.

Table 4 Area(ha)occupied by each class of soil loss estimated by the Cr-factor and C-PC factor and the respective differences between the two factors for the same classes.

Table 5 Results of EI30(MJ mm ha-1 h-1)for each 15-day period from 2009 to 2015 and the respective erosivity factors(R)in MJ mm ha-1 h-1 year-1 for the Palmares-Ribeirão do Saco watershed.

Fig.3.NDVI(a,b)and NDVI-PC(c,d)images for Palmares-Ribeirão do Saco watershed.

Fig.4.Minimum,maximum and mean values of the C-Lit,Cr,Cr2 and C-PC factors in the Palmares-Ribeirão do Saco watershed,for the selected dates.

Fig.5.Maps with maximum and minimum values of soil loss:(a)02/28/2009;(b)02/15/2010;(c)June 05,2010;(d)11/19/2010;(e)September 05,2011;(f)06/17/2012;(g)January 03,2015.

The classification and separation of images without being based on field data is very subjective for targets of similar values such as bare soil and dry vegetation.For soil cover area classification,the greatest difficulty is the separation between herbaceous targets of dry areas and of wetlands using,among other tools,NDVI.Thus,by using additional data to NDVI it is possible to improve the final response of RUSLE.Similarly,by adding the effective rate of contributing area calculated for different types of land use,Wei et al.(2018)improved the results of the slope factor(L)in RUSLE,and Tanyas et al.(2015)suggested an approach to obtain the C factor based on data from remote sensing and field,from geographic information systems(GIS).

In general,the results that most approached the C-Lit factor(obtained from literature data)were obtained by the factors C-PC and Cr2,respectively(Fig.4).This aspect can be observed from the mean values relative to the proportion at which each pixel value occurs in the images.Despite that,the differences between Cr,Cr2 and C-PC factors are small.Although the C-Lit factor was used as a reference,its use considers soil cover and management as static,not admitting the natural variations for each pixel individually(Almagro et al.,2019;Borrelli et al.,2018).Therefore,the use of calculation methodologies for variable C factor throughout the year becomes a good strategy in the use of soil loss estimation models.

Due to the use of the proposed adjustments for the Cr2 and C-PC factors,we observed that the minimum value in the images was zero(Fig.4).For the years of 2010 and 2011,we noted a difference between the Cr and C-PC factors of 1.64%and 1.26%for average and maximum values,respectively.We also found that the maximum Cr2 values tend to be greater and the average smaller than the Cr factor.

To obtain the annual soil losses in the study area,the C-PC factor was used for the dates of June 05,2010 and September 05,2011 and the Cr2 factor was used for the others.The result of the descriptive analysis of the percentage of area occupied by each class indicated a tendency of the data to non-normality according to analysis of kurtosis and skewness,confirmed by the Shapiro-Wilk test.The result of the Kruskal-Wallis test(equivalent to ANOVA I)for nonparametric data(Fantin-Cruz et al.,2010;Waltner et al.,2018)resulted in the absence of a difference between the percentages of soil cover in the period from 2009 to 2015(p<0.01),suggesting that the rainfall seasonality effect on the vegetation in the region is small for the images on the dates obtained and does not generate as much influence on the final response in soil loss.

In the studied region,it is possible to observe a trend of reduction in areas of bare soil and an increase in the area of vigorous vegetation,which is more evident in the time series(2009-2015)that integrates data from C-PC factor and Cr2 factor,because it transfers part of the values of the percentage of area occupied by vegetation affected by seasonality to the classification of more vigorous vegetation.

Comparative analysis of the results obtained from the Cr factor(Durigon et al.,2014)in comparison to the Cr2 factor with the C-PC indicated no difference for the studied area in the 2009-2015 period.Most estimates of soil loss in the Palmares-Ribeirão do Saco watershed were lower than 20.0 t ha-1.

Regardless of the methodologies for calculating the C-factor,the estimate of soil loss obtained for the region(Fig.5)from the images worked is very small.This result is associated with the period of obtaining the images,which corresponds to the drought season in the region.In periods with greater occurrence of rainfall and with higher erosive potential,the images showed a large amount of clouds,making it impossible to use them to evaluate soil cover.

The highest rates of soil loss were observed in November 2010 and March 2015 with maximum peaks in points reaching values of 126 t ha-1yr-1.On the other hand,we noted large areas with values ranging from 25 to 75 t ha-1yr-1.

The comparison of the Cr factor with the C-PC factor,observed in the responses of the dates June 05,2010 and September 05,2011,shows a small reduction in the values of C-PC factor compared to Cr.This relationship is not relevant because the H variable was below 5%,in the C-PC factor.The difference between the methodologies becomes evident in the classes worked with values below 2.0.t ha-1(Table 4).

The observed difference denotes the tendency of the C-PC factor to reduce the estimated values of soil loss for both dates,indicating that there was a migration of approximately 13 and 19 ha to the lowest class of soil loss,respectively for June 05,2010 and September 05,2011.Despite being relatively small,the difference in soil loss of 1.0 t ha-1may be associated with the fertile surface layer in shallow soils.Thus,the estimation of soil loss from the C-PC factor becomes more relevant under the conditions of the images worked.Moreover,the values of rainfall erosivity(Table 5)have a great influence on the final values of soil loss.

Schmidt et al.(2018)highlight the relationship of complementarity between rainfall erosivity(R-factor)and soil cover(C-factor)in RUSLE.Thus,the use of precipitation data associated with the obtaining of a C-factor more representative of soil cover conditions is an important increment to be taken into account in the calculation of soil loss,especially in areas where the effect of rainfall on the reflectance of vegetation cover occurs at greater intensity.

4.Conclusions

The innovation proposed in obtaining the C-factor applied to RUSLE points to a path of routines that improve the accuracy of information obtained by satellite images from the correction using meteorological data of the studied region.Thus,the conclusions of the present study are:

-NDVI-PC enables the separation of classes affected by the effect of rainfall seasonality in relation to NDVI.

-Cr2 and C-PC factors are variables that reduce soil loss values in RUSLE,being closer to the C factor generated from literature data when compared to the Cr factor.

-Cr2 represents an improvement in accuracy in relation to Cr by considering specific values of the studied area to normalize the data without generalizations.

-The proposed Cr2 and C-PC factors provide values ranging from 0.0 to 1.0 and appears as suitable alternatives to compute the RUSLE’s C-factor on multiple spatial and temporal scales.

-The C-PC factor was able to simulate the effect of vegetation cover seasonality for the Palmares-Ribeirão do Saco watershed,making it possible to consider the presence of dry vegetation affected by rainfall regimes and improving the estimate by RUSLE.

The impact of this study expands the possibilities of improving the techniques to obtain data from remote sensing in large territorial spaces,representing an advance in the way in which the data used in soil loss equations can be worked in time series.

Declaration of competing interest

The authors declare that there are no financial interests or personal relationships that may have influenced the research reported in this article.

Acknowledgments

Paulo Tarso S.Oliveira was supported by the Brazilian National Council for Scientific and Technological Development(CNPq)(grants 441289/2017-7 and 306830/2017-5)and the Coordination of Superior Level Staff Improvement-Brazil(CAPES)(Finance Code 001).