Rainfall erosivity classification.

## Abstract

To improve the use of soil and its conservation, precipitation data are necessary. With the Universal Soil Loss Equation (USLE), the study of historical precipitation series is a main factor, but in these series, there are gaps that need to be filled. This study had, as a basis, the methods of weighted likelihood, multiple regression, and weighted likelihood based on multiple regression to fill the gaps of the rainfall data for the rainfall gauges in the Brazilian biomes (Cerrado and Pantanal, municipalities of Campo Grande, Bandeirantes, Sidrolândia, Miranda, Fazenda Ponte, and Ribas do Rio Pardo). With this, it became possible to calculate the rainfall erosivity (R factor in the USLE). Therefore, the consistency of the filled rainfall data was analyzed by the double mass method. The value of the rainfall erosivity calculated varies from 2304.80 to 13562.10 MJ mm ha−1 h−1 year. With this data, it was possible to identify particular results that differed from the rainfall erosivity. Comparing all the gap-filling methods, numbers varying from 0–12% at the same rainfall gauge were obtained.

### Keywords

- water and soil conservation
- USLE
- water resources

## 1. Introduction

Climate changes are each day more and more notable throughout the world, and based on this fact, scientific studies are being developed, having as one of the main subjects studied being rainfall and its historical series [1].

In studying rainfall and its historical series, gaps in its data (these gaps can occur due to equipment failure or data observer’s mistakes, which are the most common reasons) were found. These gaps can occur in hourly, monthly, or annually collected data. In some situations, these gaps make it impossible to use the data in some studies [2].

With the need of filling these gaps, some methods were developed and are often used in studies. These methods include the artificial neural network (ANN) method, as it can be seen in [3, 4, 5], weighted likelihood method [6, 7], multiple regression method [8, 9], and weighted likelihood based on multiple regression method [1, 2].

Consequently, with the development of these methods, it was necessary to create a method to analyze the data consistency when its gaps were filled. According to this, [10] developed a method called double mass.

A continuous historical rainfall series, with filled gaps and analyzed consistency, can be applied in many studies such as urban drainage, soil conservation, and water conservation. In the soil conservation field, many studies have been developed about soil loss due to water erosion [11, 12, 13], which is described by the equation proposed by [14], which considers variables like the soil erodibility, topographic factor, soil use and management, conservation practices, and rainfall erosivity.

The rainfall erosivity (Rc) is calculated based on historical rainfall series, and to obtain these continuous historical series, certain methods are used, where the resultant data can be different depending on the used method. Based on this, this study was developed aiming to analyze the differences obtained in the rainfall erosivity results calculated with filled rainfall data using the methods—weighted likelihood, multiple regression, and weighted likelihood based on multiple regression—and to obtain a better correlation coefficient between different hydrological data sources (radar, satellite, and local).

## 2. Area of study

In order to fill the gaps in historical series, an auxiliary rainfall gauge is used. For each gap in the series, it is advised to use at least three other values from an auxiliary rainfall gauge [2]. Therefore, six rainfall gauges (Figure 1) were used: Campo Grande, Bandeirantes, Sidrolândia, Miranda, Fazenda Ponte, and Ribas do Rio Pardo. These stations are located in the Paraná River Basin, in the central area of the state of Mato Grosso do Sul, Brazil. With the stations defined, values of historical series from the Agência Nacional de Águas (ANA) and Instituto Nacional de Meteorologia (INMET) data base were collected. The range with the fewest gaps was from January 1, 2001, to December 31, 2014, and for filling the gaps, three methods were used.

## 3. Methods used to fill gaps

### 3.1. Weighted likelihood method

As it can be seen in [2], the month without data is filled with Eq. (1):

where:

P_{x} = data to be filled (mm),

n = number of auxiliary stations,

N_{x} = annual average rainfall at the station without data (mm),

N_{i} = annual average rainfall at the auxiliary station (mm),

P_{i} = rainfall at the auxiliary station in the month to be filled (mm).

### 3.2. Multiple regression method

As pointed out in [15], the multiple regression method is based on applying multiple regression, establishing a relation among the auxiliary stations and the station with the data gap, and it uses Eq. (2):

where:

y_{c} = data to be filled (mm).

n = number of auxiliary stations.

a_{n} = coefficient to be estimated by multiple regression.

x_{1i}, x_{2i} = data in auxiliary stations (mm).

### 3.3. Weighted likelihood based on multiple regression (mixed)

As it can be seen in [16], weighted likelihood based on multiple regression is based on mixing both of the previous methods. To make it easier to understand the name of this method, it will be denoted as mixed. First a multiple regression between the gap station and each auxiliary station is calculated separately, and then the weight of each station related to the gap station is calculated, using Eq. (3):

where:

W_{yxj} = weight factor between the station (y, gap station, and xj auxiliary station),

r_{yxj} = correlation coefficient between the gap station and auxiliary station (linear regression coefficient),

n = number of auxiliary stations.

For each auxiliary station, a value of W_{yxj} was obtained. With all the weight factors calculated, their sum must be equal to 1. After this is calculated, the data for the gap is calculated with Eq. (4):

where:

y_{c} = station data to be filled (mm),

x_{n} = rainfall data at the auxiliary station in the month of the data to be filled (mm),

W_{xn} = weight factor between gap station and auxiliary station.

### 3.4. Double mass method

With the data gaps filled by all three methods, it was necessary to analyze the data consistency. To achieve this, the double mass method was used as it can be seen in [17, 18, 19], described by [10], which consists of comparing two rainfall gauges using the amount of rainfall during the period of the study, developing a chart using a reliable station on the x-axis and the station to be compared on the y-axis. The chart developed tends to be a straight line; the inclination of the straight line represents the correlation between the reliable station and the station to be analyzed.

In some cases, the dots of the chart may not tend to be a straight line, or it may appear as ranges in the chart with different inclinations from the straight line found in the linear regression. These differences may be because of the failure in obtaining the data, changing the observer at the station, changing the environment nearby the station, or changing the location of the monitoring equipment [10].

To analyze the consistency, the station in Sidrolândia was determined as reliable and, therefore, used as a reference, since during the period of study, there was only one value gap.

## 4. Rainfall erosivity

To obtain the value of rainfall erosivity, Eq. (5) was used described by [20]

where:

Rc = rainfall coefficient (mm),

p^{2} = monthly rainfall (mm),

P = annual average rainfall (mm) according to [21] who has defined these parameters in Eq. (5) to be applied in Campo Grande, as it can be seen in [11]. These parameters are shown in Eq. (6):

where:

R = rainfall erosivity factor (MJ mm ha^{−1} h^{−1} year),

p = monthly rainfall,

P = annual average rainfall.

With this equation the rainfall erosivity for each method and for each rainfall gauge was calculated. After that, the rainfall erosivity for each station was classified for each method used. The following chart was used to classify the rainfall erosivity level (Table 1).

Erosivity (MJ mm ha^{−1} h^{−1}) | Erosivity classification |
---|---|

R < 2.452 | Low erosivity |

2.452 < R < 4.905 | Moderate erosivity |

4.905 < R < 7.357 | Moderate to high erosivity |

7.357 < R < 9.810 | High erosivity |

R > 9.810 | Very high erosivity |

## 5. Gap filling

All the rainfall gauges had gaps in their historical rainfall series; with the total data having 156 values, the stations had the following gaps: Campo Grande 17.3%, Bandeirantes 3.2%, Sidrolândia 0.6%, Miranda 2.6%, Fazenda Ponte 1.9%, and Ribas do Rio Pardo 3.2%.

For the gap filling using the weighted likelihood method, the values were analyzed with the double mass method, and the dots close to the linear regression were satisfactory with R^{2} values above 0.9936.

With the multiple regression method, the dispersion of the dots tended to be a straight-line regression with R^{2} values above 0.9938.

And with the weighted likelihood based on multiple regression the dispersion of the dots tended to be a straight-line regression with R^{2} values above 0.9938.

Even if the dispersion of the dots in all methods used had a satisfactory R^{2} value, the double mass method shows consistency in all stations but for the Fazenda Ponte station, because the dispersion of the dots at this station for all methods showed a different inclination along the straight-line regression.

These differences are explained by [10]; these differences can happen because of many factors like changing the monitoring equipment operator, changing the environment nearby the station, or changing the location of the monitoring equipment.

## 6. Calculating the rainfall erosivity

With the results filled and the continuous historical rainfall series, the rainfall erosivity was calculated for each station and for each method. The values obtained can be seen in Table 2, for Campo Grande.

Year | Weighted likelihood | Multiple | Mixed |
---|---|---|---|

2002 | 7390.3 | 7835.8 | 7498.0 |

2003 | 8699.5 | 9254.8 | 9953.1 |

2004 | 8115.8 | 8115.8 | 8115.8 |

2005 | 8988.5 | 9228.9 | 9045.0 |

2006 | 8054.3 | 8085.0 | 8059.2 |

2007 | 9680.6 | 9757.8 | 9696.3 |

2008 | 8394.8 | 8394.8 | 8394.8 |

2009 | 9849.2 | 9849.2 | 9849.2 |

2010 | 8437.2 | 8539.1 | 8479.4 |

2011 | 9795.6 | 10324.2 | 10134.4 |

2012 | 9507.7 | 9507.7 | 9507.7 |

2013 | 9857.1 | 9857.1 | 9857.1 |

2014 | 9480.0 | 9848.4 | 9671.3 |

Total | 116,250 | 118,598 | 118,261 |

Dif. % | −2% | 0% | 2% |

Maximum | 9857.10 | 10324.20 | 10134.40 |

Minimum | 7390.30 | 7835.80 | 7498.00 |

Average | 8942.40 | 9123.00 | 9097.00 |

The period with higher rainfall erosivity at the Campo Grande station was in 2013 according to the weighted likelihood method and in 2011 for the other two methods. The year with the lowest rainfall erosivity was in 2002 and was equal for all methods. The consistency analysis can be seen in Figure 2.

It can be noticed that for all methods, the dispersion of the dots tends to be a straight-line regression; this implies that the inclination of all methods tends to be a straight line. The rainfall erosivity in Bandeirantes can be seen in Table 3.

Year | Weighted likelihood | Multiple | Mixed |
---|---|---|---|

2002 | 7207.4 | 7207.4 | 6822.3 |

2003 | 9844.8 | 9844.8 | 10997.9 |

2004 | 8140.5 | 8140.5 | 8103.9 |

2005 | 9988.7 | 9988.7 | 10952.6 |

2006 | 9802.7 | 9802.7 | 11484.1 |

2007 | 8953.5 | 8953.5 | 8.317.3 |

2008 | 8655.5 | 8321.0 | 8894.2 |

2009 | 10440.0 | 10440.4 | 10807.4 |

2010 | 10214.0 | 10214.5 | 11654.5 |

2011 | 11396.0 | 11396.3 | 12128.6 |

2012 | 10469.0 | 10469.0 | 11168.3 |

2013 | 9492.0 | 9492.8 | 8812.0 |

2014 | 9268.0 | 9822.6 | 9864.4 |

Total | 123,874 | 124,094 | 130007.6 |

Dif. % | 0% | −5% | 5% |

Maximum | 11396.3 | 11396.3 | 12128.6 |

Minimum | 7207.40 | 7207.40 | 6822.30 |

Average | 9528.80 | 9545.70 | 10000.60 |

For the Bandeirantes station, the years of maximum and minimum rainfall erosivity were the same for all methods, in 2011 and 2002, respectively. Furthermore, the amount of the annual rainfall erosivity differs 5% (between weighted likelihood and weighted likelihood based on multiple regression). The consistency analysis can be seen in Figure 3.

For the double mass comparison for all three methods, the same dispersion pattern of the dots can be noticed. They tend to be a straight-line regression, showing a consistency in the filled gaps. The rainfall erosivity in Sidrolândia can be seen in Table 4.

Year | Weighted likelihood | Multiple | Mixed |
---|---|---|---|

2002 | 6959.2 | 6959.2 | 6430.5 |

2003 | 10078.8 | 10078.8 | 11049.6 |

2004 | 8525.7 | 8525.7 | 9011.1 |

2005 | 8479.7 | 8479.7 | 8133.8 |

2006 | 8043.5 | 8043.5 | 7797.5 |

2007 | 8676.2 | 8676.2 | 8401.6 |

2008 | 7330.7 | 7330.7 | 6718.1 |

2009 | 8521.0 | 8521.0 | 7545.5 |

2010 | 9491.9 | 9491.9 | 10599.2 |

2011 | 8927.3 | 8927.3 | 8170.2 |

2012 | 8607.5 | 8607.5 | 7826.6 |

2013 | 9686.7 | 9686.7 | 9271.5 |

2014 | 9513.2 | 9688.3 | 10015.3 |

Total | 112,841 | 113,016 | 110,970 |

Dif. % | 0% | 2% | −2% |

Maximum | 10078.80 | 10078.80 | 11049.60 |

Minimum | 6959.20 | 6959.20 | 6430.50 |

Average | 8680.10 | 8693.60 | 8536.20 |

The year 2002 had the lowest rainfall erosivity for all methods, and 2003 had the highest rainfall erosivity, also for all methods. The rainfall erosivity in Miranda can be seen in Table 5.

Year | Weighted likelihood | Multiple | Mixed |
---|---|---|---|

2002 | 6221.7 | 6221.7 | 4736.0 |

2003 | 8780.3 | 8780.3 | 9272.5 |

2004 | 8533.0 | 8533.0 | 8798.5 |

2005 | 9185.4 | 9185.4 | 8589.9 |

2006 | 6520.9 | 6566.2 | 5745.0 |

2007 | 7139.4 | 7139.4 | 5382.3 |

2008 | 8047.4 | 8047.4 | 7246.6 |

2009 | 9359.5 | 9351.6 | 8434.3 |

2010 | 7978.7 | 7978.7 | 7239.7 |

2011 | 9506.9 | 9506.9 | 8547.2 |

2012 | 8201.3 | 8201.3 | 6987.4 |

2013 | 8160.9 | 8160.9 | 6650.1 |

2014 | 8142.1 | 8778.2 | 8415.1 |

Total | 105,777 | 106,450 | 96,044 |

Dif. % | −1% | 10% | −10% |

Maximum | 9506.90 | 9506.90 | 9272.50 |

Minimum | 6221.70 | 6221.70 | 4736.00 |

Average | 8136.70 | 8188.50 | 7388.10 |

For the Miranda station, the period with the highest rainfall erosivity was in 2001 according to the weighted likelihood and multiple regression methods and in 2003 according to the weighted likelihood based on multiple regression method. The period with the lowest rainfall erosivity was in 2002 for all methods. The percentage difference between the weighted likelihood and weighted likelihood based on multiple regression methods was 10%. The double mass analysis can be seen in Figure 4.

The data consistency of the Miranda station can be analyzed through its dispersion of dots compared to a straight-line regression, showing the same inclination. The rainfall erosivity at the Fazenda Ponte station can be seen in Table 6.

Year | Weighted likelihood | Multiple | Mixed |
---|---|---|---|

2002 | 8228.4 | 8228.4 | 8784.0 |

2003 | 8961.6 | 8961.6 | 10735.4 |

2004 | 5677.4 | 5677.4 | 3777.0 |

2005 | 8042.9 | 8042.9 | 6903.9 |

2006 | 7737.0 | 7886.2 | 8070.0 |

2007 | 9977.0 | 9977.0 | 8967.4 |

2008 | 5207.0 | 5341.9 | 3296.7 |

2009 | 5458.3 | 5458.3 | 2304.8 |

2010 | 9229.6 | 9229.6 | 9228.5 |

2011 | 9580.5 | 9580.5 | 8822.1 |

2012 | 8481.8 | 8481.8 | 7609.0 |

2013 | 7937.3 | 7937.3 | 6229.6 |

2014 | 7290.4 | 7290.4 | 5918.4 |

Total | 101,809 | 102,093 | 90,647 |

Dif. % | 0% | 11% | −12% |

Máximo | 9977.0 | 9977.0 | 10735.40 |

Mínimo | 5207.00 | 5341.90 | 2304.80 |

Média | 7831.50 | 7853.30 | 6972.90 |

The period with the highest rainfall erosivity at the Fazenda Ponte station was in 2007 (weighted likelihood and multiple regression) and in 2003 (weighted likelihood based on multiple regression), and the period with lowest rainfall erosion was in 2008 (weighted likelihood and multiple regression) and in 2009 (weighted likelihood based on multiple regression). For this method the largest percentage difference was 12%. The double mass analysis can be seen in Figure 5.

Analyzing the dispersion of the data dots from this station, it is noticed that in some parts, the dispersion of the dots has different inclinations along the chart, even if the statistic approach is satisfactory (R^{2} = 0.9936). These differences along the straight-line regression show us inconsistencies described by [10]. These different inclinations can be explained by facts such as changes of monitoring station operators, environmental changes nearby the location of the station, and changing of the location of the monitoring equipment. The rainfall erosivity in Ribas do Rio Pardo can be seen in Table 7.

Year | Weighted likelihood | Multiple | Mixed |
---|---|---|---|

2002 | 8564.3 | 8564.3 | 9544.8 |

2003 | 8549.5 | 8549.5 | 9251.4 |

2004 | 9306.9 | 9306.9 | 10365.7 |

2005 | 8835.9 | 8835.9 | 8506.0 |

2006 | 9246.2 | 9246.2 | 10475.0 |

2007 | 9750.6 | 9750.6 | 9898.2 |

2008 | 7252.5 | 6992.6 | 6352.9 |

2009 | 8998.5 | 8998.5 | 8213.7 |

2010 | 9541.0 | 9541.0 | 10217.7 |

2011 | 12165.6 | 12165.6 | 13562.1 |

2012 | 7957.2 | 7957.2 | 6592.5 |

2013 | 8270.9 | 8270.9 | 6724.7 |

2014 | 8506.9 | 8779.3 | 8773.6 |

Total | 116,946 | 116,958 | 118,478 |

Dif. % | 0% | −1% | 1% |

Máximo | 12165.60 | 12165.60 | 13562.10 |

Mínimo | 7252.50 | 6992.60 | 6352.90 |

Média | 8995.90 | 8996.80 | 9113.70 |

The period with the highest rainfall erosivity at the Ribas do Rio Pardo station was in 2013 and with the lowest rainfall erosivity was in 2008. This rainfall gauge had the lowest percentage difference among all the stations (1%). The double mass analysis can be seen in Figure 6.

The data of the Ribas do Rio Pardo station is noticed to be consistent; the dispersion of the dots tends to be a straight-line regression, with no change in inclination along the line.

The values of erosivity that were found in the study are in agreement with [24].

However, in the author’s study, erosivity is classified only as high, and the gap filling can range from moderate to very high. This can be explained by the fact that the author [24] did not mention gap filling and used fewer stations in the study area (Cerrado and Pantanal biome).

## 7. Rainfall erosivity classification

After filling the gaps and analyzing the consistency of all rainfall gauges, the rainfall erosivity was classified according to Table 1. The rainfall erosivity in all stations was filled by the weighted likelihood method, which can be seen in Table 8.

Station | Average | Classification |
---|---|---|

Campo Grande | 8942.4 | High erosivity |

Bandeirantes | 9528.8 | High erosivity |

Sidrolândia | 8680.1 | High erosivity |

Miranda | 8136.7 | High erosivity |

Fazenda Ponte | 7831.5 | High erosivity |

R. do Rio Pardo | 8995.9 | High erosivity |

The classification of rainfall erosivity using the multiple regression method to fill gaps can be seen in Table 9. The weighted likelihood based on multiple regression method and its classification results can be seen in Table 10.

Station | Average | Classification |
---|---|---|

Campo Grande | 9123.0 | High erosivity |

Bandeirantes | 9545.7 | High erosivity |

Sidrolândia | 8693.6 | High erosivity |

Miranda | 8188.5 | High erosivity |

Fazenda Ponte | 7853.3 | High erosivity |

R. do Rio Pardo | 8996.8 | High erosivity |

Station | Average | Classification |
---|---|---|

Campo Grande | 9097.0 | High erosivity |

Bandeirantes | 10000.6 | Very high erosivity |

Sidrolândia | 8536.2 | High erosivity |

Miranda | 7388.1 | High erosivity |

Fazenda Ponte | 6972.9 | Moderate to high |

R. do Rio Pardo | 9113.7 | High erosivity |

After the classification was finished for each method, it can be noticed that the Bandeirantes and Fazenda Ponte stations had different rainfall erosivity classifications between the methods. Bandeirantes had a rainfall erosivity classification of very high erosivity (weighted likelihood mean based on multiple regression) and of high erosivity (weighted likelihood and multiple regression). The Fazenda Ponte station had a rainfall erosivity classification of moderate to high erosivity (weighted likelihood based on multiple regression) and high erosivity (weighted likelihood and multiple regression).

The results reported here may help to identify a more adequate methodology to fill the gaps in the region. It will subsidize appropriate plans and projects to infer the best land-use strategies to improve water and soil conservation and quality and promote agriculture sustainability. Mainly, this is important because the Cerrado is one of the Brazilian biomes that has been subjected to the highest agronomic pressure, according to [25], with high interaction with the Pantanal biome.

## 8. Rainfall data sources combined: satellite, radar, and local

When it is necessary to study rainfall, a couple of variables need to be determined, for example, the size of the area, duration of the historical rainfall series, and data source. Nowadays there are some kinds of sources available: satellite, radar, and local. How to choose among them? A very important detail when choosing the data source is to identify the size of the study area. When the study area is a state, a country, or a continent, the satellite data achieves a better accuracy. On the other hand, if the study area is a state or an area with a couple of cities, radar data is more advised. Finally, if the study area is a city, a small watersheds, or a couple of cities, the local data (rainfall gauges) provides a better result.

Satellite data is possible to be obtained using Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks-Climate Data Record (PERSIANN- CDR) [26], Tropical Rainfall Measuring Mission (TRMM) [27], and Gravity Recovery and Climate Experiment (GRACE) [28]. An important characteristic is the rainfall amount; some studies show that the same satellite data can provide different accuracy depending on the rainfall amount. Some characteristics are important to observe; in developing countries the number of local rainfall gauges is low, and some studies need to use two data sources combined.

Satellite data are widely used to calculate hydrological parameters for areas that are sparsely equipped with rain gauges; thus it is possible to obtain data for a large area. On the other hand, its accuracy for high rainfall quantities is low, for example, [26] describes a correlation coefficient of 0.62 between PERSIANN-CDR and data from local gauges, in a heavy rainfall event.

Consequently, different studies for each data set are required to obtain a better combination according to each study. This paper provides an approach to obtain more accurate data through different gap-filling methods. In this way, more studies are needed to sensor gap filling, providing future studies with bettered methods and combinations.

## 9. Conclusions and recommendations

The value of rainfall erosivity calculated varies from 2304.80 to 13562.10 MJ mm ha

^{−1}h^{−1}year. It was possible to identify variations in rainfall erosivity classification, comparing all the gap-filling methods; numbers varying from 0–12% at the same rainfall gauge were obtained.In the double mass analysis, even if statistic approaches are satisfactory and tend to be a straight-line regression, the inclination along the straight-line regression should be considered.

The consistency analysis can explain the different results obtained. The Fazenda Ponte station was an example where a break in the slope was found and the results obtained diverged 12%.

The weighted likelihood mean and multiple regression methods had similar performances in filling gaps; the rainfall erosivity values had a 2% maximum difference.

The weighted likelihood based on multiple regression was not often found in scientific articles even if it is adopted in books.

For future studies the use of the weighted likelihood based on multiple regression gap-filling method combined with a satellite data source is recommended.

## References

- 1.
Ishihara JH, Fernandes LL, Duarte AAAM, Duarte ARCI, Ponte MX, Loureiro GE. Quantitative and spatial assessment of precipitation in the Brazilian Amazon (Legal Amazon) — (1978 to 2007). Revista Brasileira de Recursos Hídricos. 2014; 19 :29-39. DOI: 10.21168/rbrh.v19n1.p29-39 - 2.
Oliveira LFC, Fioreze AP, Medeiros AMM, Silva MAS. Comparison of gap filling methodologies of annual historical series of rainfall. Revista Brasileira de Engenharia Agrícola e Ambiental. 2010; 14 :1186-1192 - 3.
Detzel DHM, Oening AP, Souza ARR, Cerminaro SLC. Completing hourly river stage data via ARIMA models. Revista Brasileira de Recursos Hídricos. 2013; 18 :281-292 - 4.
Oliveira GG, Pedrollo OC, Castro NMR, Bravo JM. Hydrological simulations with different proportions of controlled area in the watershed. Revista Brasileira de Recursos Hídricos. 2013; 18 :193-204. DOI: 10.21168/rbrh.v18n3.p193-204 - 5.
Paiva LFG, Montenegro SMGL, Valença MJS. Runoff forecasting model based on the neutral networks technique for the São Francisco River basin. Revista Brasileira de Recursos Hídricos. 2010; 15 :05-15. DOI: 10.21168/rbrh.v15n1.p5-15 - 6.
Castro FS, Pezzopane JEM, Cecílio RA, Pezzopane JRM, Xavier AC. Evaluation of the performance of the different methods of interpolators for parameters of the climatologic water balance. Revista Brasileira de Engenharia Agrícola e Ambiental. 2010; 14 :871-880. DOI: 10.1590/S1415-43662010000800012 - 7.
Silva RM, Silva LP, Montenegro SMGL, Santos CAG. Analysis of the space-time variability and identification of rainfall pattern within the Tapecurá River basin, Pernambuco state. Sociedade and Natureza. 2010; 22 :357-372. DOI: 10.1590/S1982-45132010000200010 - 8.
Costa HC, Marcuzzo F, Ferreira OM, Andrade LR. Seasonality and spatial distribution of rainfall in the state of Goiás and Federal District. Revista Brasileira de Geografia Física. 2012; 06 :1275-1291 - 9.
Swarowsky A, Heldwein AB, Buriol GA, D'avila RF, Estefanel V. Homogeneity and descriptive statistics of monthly and annual rainfall for Santa Maria, RS. Revista Brasileira de Recursos Hídricos. 2006; 11 :89-97. DOI: 10.21168/rbrh.v11n4.p89-97 - 10.
Searcy JK, Hardison CH. Double-mass curves. Manual of Hydrology: Part 1. General surface-water techniques. Washington; 1960. pp. 66 - 11.
Magalhães Filho FJC, Ayres FM, Sobrinho TA. Integrating GIS and USLE for soil loss mapping in environmental protection area. Revista Agrarian. 2014; 7 :552-559 - 12.
Chaves HML, Braga B Jr, Domingues AF, Santos DG. Quantificação dos benefícios ambientais e compensações financeiras do “Programa do Produtor de Água” (ANA): I.Teoria. Revista Brasileira de Recursos Hídricos. 2014; 9 :5-14 - 13.
Kinnell PIA. Event soil loss, runoff and the universal soil loss equation family of models: A review. Journal of Hydrology. 2010; 385 :384-397 - 14.
Wischmeier WH, Smith DD. Predicting Rainfall Erosion Losses: A Guide to Conservation Planning. U.S. Department of Agriculture, Handbook; 1978. 537 - 15.
Depiné H, Castro NZR, Pinheiro A, Pedrollo OC. Filling hourly data gaps in rainfall time series using artificial neural networks. Revista Brasileira de Recursos Hidrícos. 2014; 19 :51-63. DOI: 10.21168/rbrh.v19n1.p51-63 - 16.
Bertoni JC, Tucci CEM. Precipitação. In: CEM T, editor. Hidrologia: Ciência e Aplicação. 3th ed. Porto Alegre: UFRGS; 2002. p. 943 - 17.
Wissmann JA, Tampelini LG, Feil AC, Sampaio SC, Suszek M. Ferramenta computacional para análise de consistência de dados pluviométricos. Revista Varia Scientia. 2006; 06 :99-106 - 18.
Sanches FO, Balen DS, Silva RV, Rosa KK, Radunz AL. Rains in Rio Grande do Sul: A study of intense rainfall accumulated in the upper Uruguay Gaúcho. Revista Brasileira de Climatologia. 2014; 15 :143-162. DOI: 10.5380/abclima.v15i0.38074 - 19.
Chechi L, Sanches FO. Using the rainfall anomaly index (RAI) in evaluating the El Niño southern oscillation (ENSO) in the upper Uruguay gaucho from 1957 to 2012. Revista Brasileira de Geografia Física. 2013; 06 :1586-1597 - 20.
Lombardi Neto F, Moldenhauer WC. Rainfall erosivity–Its distribution and relationship with soil loss at Campinas, state of São Paulo, Brazil. Bragantia. 1992; 51 :189-196 - 21.
Oliveira PTS, Wendland E, Nearing MA. Rainfall erosivity in Brazil: A review. Catena. 2012; 100 :139-147. DOI: 10.1016/j.catena.2012.08.006 - 22.
Carvalho NO. Hidrossedimentologia Prática. 2nd ed. Rio de Janeiro: Interciência; 2008. p. 602 - 23.
Foster GR, McCool DK, Renard KG, Moldenhauer WC. Conservation of the universal soil loss equation to SI units. Journal of Soils and Water Conservation. 1981; 36 :355-359 - 24.
Trindade ALF, Oliveira PTS, Anache JAA, Wendland E. Spatial variability of rainfall erosivity in Brazil. Pesquisa Agropecuária Brasileira. 2016; 51 :1918-1928 - 25.
Kaschuk G, Alberton O, Hungria M. Quantifying effects of different agricultural land uses on soil microbial biomass and activity in Brazilian biomes: Inferences to improve soil quality. Plant and Soil. 2011; 338 :467-481. DOI: 10.1007/s11104-010-0559-z - 26.
Ashouri H, Hsu KL, Sorooshian S, Braithwaite DK. PERSIANN-CDR: Daily precipitation climate data record from multisatellite observations for hydrological and climate studies. Bulletin of the American Meteorological Society. 2015; 96 :69-83. DOI: 10.1175/BAMS-D-13-00068.1 - 27.
Yong B, Liu D. Global view of real-time Trmm multisatellite precipitation analysis: Implications for its successor global precipitation measurement mission. Bulletin of the American Meteorological Society. 2015; 96 :283-296. DOI: 10.1175/BAMS-D-14-00017.1 - 28.
Yi S, Song C, Wang Q, Wang L, Heki K, Sun W. The potential of GRACE gravimetry to detect the heavy rainfall-induced impoundment of a small reservoir in the upper Yellow River. Water Resources Research. 2017; 53 :6562-6578. DOI: 10.1002/2017WR020793