Soil moisture is an important parameter of the hydrologic cycle, which has hydrological, ecological, environmental, and agricultural impacts. This occurs because soil moisture is directly or indirectly related to processes, such as surface runoff and groundwater recharge, as well as ecosystem behavior [1
]. Soil moisture monitoring supplies fundamental information about interactions between soil, vegetation, and atmosphere to improve the accuracy of meteorological forecasting [3
]. These data also help improve agricultural productivity and flood and drought risk management, contributing to supporting actions that mitigate the effects of water scarcity [1
]. Soil moisture has been considered an important variable for establishing drought severity. Soil moisture has been applied for drought monitoring by the United States Drought Monitor, which uses data from the U.S. National Weather Service’s Climate Prediction Center soil moisture model. Soil moisture has been particularly useful for agricultural drought monitoring using the water balance model [1
] and remotely sensed soil moisture [8
]. Soil moisture can be obtained by in situ measurements, which is expensive for large areas [4
], or by sensors on board satellites for global monitoring. Given the limitations of in situ measurements, remote sensing has become an efficient alternative for collecting data on a large scale for studies where time and space are relevant aspects [2
Soil moisture studies using remote sensing gained momentum during the 1990s with the launching of three satellites missions that transported Synthetic Aperture Radar (SAR) Radar Satellite (RADARSAT), European Remote Sensing (ERS), and Japanese Earth Resources Satellite (JERS). SAR data have potential for estimating soil moisture due to the ability of electromagnetic pulses to penetrate the soil and modify its properties when interacting with water [10
Brightness temperature (TB) is another method used to estimate soil moisture using remote sensing. TB is acquired with passive sensors that operate in the range of microwaves. The estimation is possible due to the direct relation between soil moisture and soil emissivity [10
]. The Soil Moisture Ocean Salinity (SMOS) and Soil Moisture Active and Passive (SMAP) missions are examples of sensors with these characteristics. Both sensors use the L-band to collect soil moisture. These missions have conducted campaigns to assess the remotely-obtained products compared with in situ data in different climates in order to study drought effects [14
In November 2009, the European Space Agency (ESA) launched the SMOS satellite carrying on board a passive sensor. This microwave imaging radiometer using aperture synthesis (MIRAS), used in the present work, operates on the L-band at frequency 1.4 GHz and polarization horizontal-vertical (H-V) [2
]. These products are available with a daily time resolution for three-day, nine-day, monthly, and yearly periods [19
] and have an average ground resolution of 43 km [13
], but can be better depending on the level of processing (40 km for L2, 25 km for L3, and 1 km for L4).
A limited number of soil moisture monitoring networks exist around the world [20
]. Some of these networks include: Soil Moisture Measurement Stations Network of the University of Salamanca (REMEDHUS), Spain [5
]; Wales Soil Moisture Network (WSMN), UK [21
]; and OzNet in Australia [14
]. South America is an example of a region that lacks in situ data, mainly due to its large territory where data collection is often unviable. A soil moisture monitoring network was established in the Brazilian semiarid region with the objective of obtaining relevant information for use in agriculture productivity modelling. The in situ data are also important for remote sensing product validation [21
] for the confident use of these data. However, maintaining a monitoring network is costly and a major challenge, mainly for large areas. Owing to the few studies developed with the goal of assessing the SMOS products in Brazil and South America, the present study seeks to validate these products, indicating which would be useful for future applications as drought monitoring in the semiarid regions. Northeast Brazil is considered the most densely populated semiarid region in the world with a population reaching 23.5 million inhabitants. The climate affects the life of the population that has to face long drought periods with consequences for water supply, irrigation and agriculture, among other activities. The impacts of droughts can be attenuated with tools that allow monitoring of the variables associated to this phenomenon, especially for large areas as is the case of Brazilian semiarid whose area is almost 1 million km2
The SMOS products represent the heterogeneity of a surface with only one pixel, complicating validation [6
], whereas the in situ observation networks provide data for validation on several scales [6
]. The density of the in situ stations is crucial for the satisfactory representation of wet and dry periods in the region represented by the pixel. The coarse resolution of the soil moisture remote sensing products also creates challenges for comparisons with observation networks. Even for large networks, the sampling rate is not high enough to provide at least one station per grid cell [6
The validation of products of passive microwaves can be performed using in situ data, models, and other satellite products. Several processes have been developed by the scientific community to validate soil moisture remote sensing products from satellites, such as the Advanced Microwave Scanning Radiometer for the Earth Observing System (AMSR-E) [10
], SMAP [26
], and SMOS [2
]. These remotely sensed data have been used to obtain drought indices with monitoring goals [10
], including SMOS soil moisture data [1
]. This kind of application is relevant for the aim of our study.
Since 2012, the semiarid region of Northeast Brazil has been affected by drought encompassing a large area of about 1 million km2
. By the end of 2017, precipitation was still below the historical average. The assessment of this drought showed it was the most extreme and longest drought ever registered in the area [31
]. According to Alvalá et al. [32
], in 2015–2016, 184 municipalities in the State of Pernambuco were affected by the drought, of which 76 had more than 50% of their area impacted with negative consequences on family agriculture. An estimate of family farming establishments susceptible to drought impact during 2015 and 2016 was equivalent to 141,143 establishments in Pernambuco State alone.
Considering the difficulties related to SMOS data validation and the importance of studies in different parts of the world, this paper reports on the validation of soil moisture data from MIRAS-SMOS using two in situ databases collected in Pernambuco State, located in Northeast Brazil. The extent of the study area ensures heterogeneity in terms of land cover, topography, and climate characteristics. The validation process considers two approaches, pixel-station comparison and areal average, for three regions in Pernambuco State with different climate characteristics. After validation, the SMOS data are used for drought assessment to calculate soil moisture anomalies for the period of data available.
The performance of the SMOS data varied according to the climate characteristics of the study area. The APAC network has six stations located in the Mata region (humid tropical climate) and six in the semiarid region. The correlations of the Pearson’s r in the Mata region varied from very weak to moderate with daily data, and from moderate to strong for eight-day data. The Willmott index varied between 0.36 and 0.69, and 0.42 and 0.71 for the one-day and eight-day data, respectively. In the six stations located in the semiarid region, the Pearson’s r varied from moderate to strong correlations for the daily time interval and moderate to strong correlations for the eight-day time interval (five stations with strong correlation). For the Willmott index, the correlations varied from 0.51 and 0.86 with the daily time intervals to 0.52 and 0.95 with the eight-day time interval (four stations between 0.83 and 0.95). The SMOS data exhibited the strongest correlations with the CEMADEN network because all stations were located in the semiarid region. The assessment based on the areal average also showed that the stations located in the semiarid region performed the best. The correlations varied from strong to very strong in both time intervals. Considering the three regions, Agreste had the best correlation values. The Student’s t test was applied to all correlations to verify if the results were statistically significant. The results showed that SMOS and in situ data had a statistically significant correlation with just one exception. The station with the lowest number of samples for the eight-day interval (n = 11) had a p-value greater than 0.05.
Despite the difference in performance, the SMOS satellite was able to satisfactorily capture the time variation in the in situ data in both dry and wet regions, accurately capturing the seasonality due to the rainfall [44
]. The SMOS soil moisture product supplied consistent results for all surfaces varying from very dry to wet. Similar performance was observed by Molero et al. [2
], for which the best agreement between SMOS and the in situ data were obtained for a semiarid region. Other studies also reported better results for validating soil moisture remote sensing in semiarid land in comparison to temperate regions [45
]. The validation of SMOS data accomplished by Molero et al. [2
] was completed in four study areas with different climate characteristics. These data were processed by the DISaggregation based on Physical and Theoretical scale Change (DISPATCH) algorithm, which generates soil moisture data with a resolution of 1 km. The results showed that the product improved the space-time correlation with in situ measurements for semiarid regions with considerable soil moisture space variability due to precipitation and irrigation. In subhumid regions, the performance of the algorithm was poor, except in summer, for which the results were better.
The set of BIAS calculated presented positive and negative values. The results were mainly positive, which reveals a slight overestimation of the SMOS data in relation to both observation networks for areal average and pixel-station validations. In the latter, both networks had a BIAS that was more than 50% positive, and the overestimation was more evident when satellite data were validated with the APAC stations, for which 75% of the daily data had a positive BIAS and 66.7% for the eight-day average. The stations were located on terrain with a slope varying between 0% and 20%. The 0–3% slope interval (comprising 25 stations) tended to present a proportion of positive bias greater than the overall proportion, representing a positive bias in 76% of the stations and negative bias in 24% of the stations. The 3–8% interval (31 stations) had a proportion near the overall value (58% positive and 42% negative). Finally, the 8–20% interval (eight stations) presented a positive bias lower than the overall proportion (12.5% positive and 87.5% negative). Underestimation was identified by some authors during the validation procedures [1
]. He et al. [49
] verified that the sun-glint, which is the reflected solar radiation from the land surface near the specular direction, can affect the brightness temperature by increasing emissivity and decreasing soil moisture, which result in negative BIAS. The sun-glint is stronger for greater terrain slopes, lower solar incidence angles, and over wetter soils. The method developed by He et al. [49
] showed that the inclusion of the terrain slope will result in stronger sun-glint for the SMAP radiometer, which means greater brightness temperature and, consequently, lower soil moisture.
The average RMSD calculated with daily time series in the pixel and areal assessment was 0.071 m3·m−3 and 0.04 m3·m−3, respectively. These values are near to the expected 0.04 m3·m−3 accuracy of the MIRAS-SMOS sensor. The RMSD was particularly low in the areal average for the Sertão (0.025 m3·m−3) and Agreste (0.033 m3·m−3) regions.
Some of CEMADEN’s stations presented inferior performance during the dry period, as shown in Figure 5
and Figure 6
. González-Zamora et al. [6
] observed underestimation of the SMOS data particularly in the dry periods, but this behavior has not yet been well characterized or understood.
The statistical criteria values showed that SMOS data fit the in situ data well. Both the pixel and areal assessments had RMSD values around the expected accuracy of the MIRAS sensor. Similar to other validation studies, better results were observed in the semiarid part of the study area. In general terms, the behavior of soil moisture satellite and in situ agreed, despite the slight overestimation verified in the assessment.
The assessment presented in this paper is one of the first accomplished in South America in an area the size of the state of Pernambuco. Another particular aspect of this study is the evaluation of the SMOS data in terms of areal average. The satisfactory results are relevant for their application to large areas for drought monitoring, as Northeast Brazil is almost 1 million km2.
The results obtained with the validation of the SMOS data in Pernambuco State encourage its use in other applications. The drought period showed the need for maintaining regular monitoring of the hydrological and climate variables related to this phenomenon. Soil moisture data from satellites can be used together with other data, such as precipitation, streamflow, and water storage in reservoirs to better characterize the state of drought over a period of time. This may be particularly relevant for planning in agriculture and supporting decision makers and farmers. One future application will be for the calculation of a drought index to monitor potential impacts for water resources and agriculture in Northeast Brazil. The application of the SMOS data can be extended to the entire Northeast of Brazil, which has been impacted by a long period of drought. The climate characteristics of Pernambuco State are similar to those for the entire Northeast.