The amount of water stored in the soil is a key parameter for the energy and mass fluxes at the land surface-atmosphere boundary and is of fundamental importance to many agricultural, meteorological, biological and biogeochemical processes [1
]. For these reasons, soil moisture (SM) has been identified as an Essential Climate Variable (ECV) by the Global Climate Observing System (GCOS) secretariat [4
]. Monitoring such a complex phenomenon over wide areas is not trivial. In fact, it has been observed that particular meteorological conditions, geological characteristics, topography and land cover can affect the soil moisture variation in a small area as much as in a large region [5
]. Moreover, the amount of water stored in the top layer of the soil can change significantly within a few hours [8
] due to the influences of the atmosphere.
Spaceborne remote sensing has shown itself to be a suitable tool to monitor soil moisture over large regions at regular time intervals. Great progress has been made by the scientific community in the last three decades aiming at developing soil moisture retrieval techniques by using optical, thermal infrared (TIR) and microwave (MW) sensors [9
]. Since the late 1970s, coarse resolution (25–50 km) soil moisture products derived from past and present microwave radiometers (Advanced Microwave Scanning Radiometer (AMSR-E) [11
] and WindSat [12
]) and scatterometers (European Remote Sensing satellites (ERS) scatterometer (SCAT) [13
] and Meteorological Operational satellite (MetOp) Advanced Scatterometer (ASCAT) [14
]) have been available on an operational basis. A first global soil moisture product meeting the requirements set by GCOS was created within the framework of the European Space Agency (ESA) Water Cycle Multi-mission Observation Strategy (WACMOS) project [16
], by merging soil moisture products derived from multi-frequency radiometer and C-band scatterometer observations into a single dataset covering the period from 1979 to 2010 [17
]. The WACMOS soil moisture product is currently being extended and enhanced in the framework of the ESA-funded Climate Change Initiative (CCI) program [20
]. Despite the advantageous high temporal frequency (up to daily data available) of such a product, its relatively coarse spatial resolution (0.25 × 0.25 deg) may not be suitable to represent the soil moisture variation within a quite large area. Increasing confidence in the use of the CCI ECV SM product (we will refer to it as “ECV SM” in this paper) can be achieved by assessing its quality through inter-comparisons with independent soil moisture datasets. Commonly, ground measurements, models or other satellite acquisitions are used to provide validation soil moisture datasets ([21
data-based validation has generally been achieved over small temporal and spatial scales but has been significantly advanced since the establishment of the Global Soil Moisture Data Bank [23
] and the International Soil Moisture Network [24
]. Such an approach was used in [25
], where three global soil moisture products, including the WACMOS time series, have been validated using a combination of 196 in situ
stations taken from five soil moisture networks across the world. Similarly, in [26
] and [27
], over 600 in situ
stations have been used for validating ASCAT and ECV SM products, respectively, finding general good agreement between the satellite-derived and in situ
observations. However, soil moisture records provided by the in situ
networks represent only single point locations and usually cover limited observation periods. The necessity of a comprehensive characterization of in situ
representativeness errors when considering satellite-derived and in situ
soil moisture inter-comparison has been highlighted in [28
], where the quality of over 1400 in situ
stations of the ISMN for representing soil moisture at satellite footprint scales (~25 km) has been investigated on a global basis by adopting a triple collocation approach.
The higher spatial resolution and the regular coverage provided by spaceborne Synthetic Aperture Radars (SARs) make them a promising additional data source for measuring seasonal and long-term variations in surface soil moisture content and for a better understanding of coarse scale soil moisture products ([29
]). For instance, the advanced synthetic aperture radar (ASAR) instrument onboard the ENVISAT satellite was capable of providing global measurements at 1 km and 150 m spatial resolution every four to seven days, depending on the acquisition plan. However, the comparison of time series of soil moisture datasets acquired by different sources and representing different spatial scales is challenging due to the scale differences between products and/or observations [31
]. However, given the temporal stability of soil moisture patterns, their inter-comparisons are useful where soil moisture values at smaller scales are representative of the mean soil moisture content over larger areas [32
This study is focused on investigating the capability of the coarse scale ECV SM product in capturing the temporal and spatial variations in surface soil moisture, as recorded by in situ
instruments and retrieved from ASAR Wide Swath (WS) acquisitions. It is an extension of the work presented in [33
], where the first released version of the global ECV SM product was validated over three sites in South Ireland. This former study proved that despite the adopted validation method do not make use of dense in situ
station networks, nor hydrological models, it has the potential to be an efficient and cost-effective approach, whose reliability was proved by the consistency of the achieved results with those reported in other papers using different sensors and classical methods. Although a quite good quality of the first version of the ECV SM product in South Ireland has been observed in [33
], the study highlighted also its poor capability in capturing the driest and wettest soil conditions, as well as a decrease in its reliability in the presence of particular types of soil and at higher altitudes. Because this former work was carried out over a limited and quite homogeneous region, the influence of other factors (e.g., land cover, complex topography, climate zone) on soil moisture behavior and on the accuracy of the global ECV SM dataset could not be investigated. However, the actual advantage for climate change studies, which can be derived from the availability of such a long, temporally frequent and global SM product, has to be further tested. Aiming at a more comprehensive understanding of the ECV SM product, which would lead to an increase in the confidence in its use, the study presented in [33
] needs to be extended to other areas worldwide, especially focusing on those which could be representative of specific climate zone and characterized by a variety of land cover, soil type, and different topography.
Recently, the ECV SM dataset has been temporally extended and enhanced, and a new version has been made available in July 2014. Continuing and extending the validation activity of this global SM product, a more comprehensive analysis is presented in this study, which shows the results of the quality assessment of the latest released ECV SM product carried out over three different countries characterized by contrasting climate conditions: Spain, Ireland and Finland.
2. Test Sites Description
The quality assessment of the ECV SM product has been focused on three different areas located in the Duero basin in Spain, in southern Ireland, and in Finland (see Figure 1
). The choice has been driven by the interest in investigating the capability of the ECV SM data in describing the soil moisture dynamics in different scenarios especially in terms of climate and land cover.
Areas and sites under investigation in Spain (REMEDHUS soil moisture network), Ireland (soil moisture network from AEON project) and Finland (FMI and GTK soil moisture network).
Areas and sites under investigation in Spain (REMEDHUS soil moisture network), Ireland (soil moisture network from AEON project) and Finland (FMI and GTK soil moisture network).
The Duero basin is characterized by a semi-arid continental Mediterranean climate, with an average annual precipitation of 385 mm and a mean temperature of 12 °C [34
]. This quite flat region (slope: <10%; altitude: 700–900 msl) is mainly covered by cereal fields and vineyards, but patchy areas of forest and pasture can also occur. The soil texture is mainly sandy, with a mean sand content of about 71%.
The Irish region under investigation is characterized by a humid mild temperate climate, with a mean annual precipitation of ~1200 mm·yr−1
and a mean temperature of 10 °C. The in situ
stations are installed in grassland areas, which represent almost 80% of the agricultural area of Ireland (4.4 million hectares) [35
]. The region is typically low lying, with altitudes ranging between 15 m and 104 m above sea level, and relatively flat (slope lower than 6°). On the basis of the United States Department of Agriculture (USDA) classification, the soil texture is classified as sandy loam in Kilworth and as loam in Pallaskenry and Solohead.
The Finnish sites are mainly distributed along the eastern edge of the country, with the exception of Pori, on the south-west, and Sodankylä, which is located in Northern Finland, north of Arctic Circle [36
]. All the sites are in low-lying regions (altitude: <400 msl), typically covered by boreal forests, open and forested bogs and tundra highlands. Due to the boreal climate characterizing this area, winters are very cold and snow precipitations are very common. Along its eastern border with Russia, and in the northern areas the snow coverage is often deep, with some remaining on the ground into early May, and much later to the north of the Arctic Circle. The annual amount of rain precipitation varies between 500 mm in Northern Finland and 650 mm in south-east of the country. The annual mean temperature varies from more than 5 °C in Southwestern Finland, to a couple of degrees below zero in Northern Finland.
This work has focused on the quality assessment of the latest released version of the global soil moisture product provided through the ESA CCI program. The long SM time series (up to five years of observations) collected over three regions in Spain, Ireland and Finland have been temporally and spatially compared to the finer spatial resolution SM product retrieved from ASAR WS acquisitions, and to in situ soil moisture datasets.
The suitability of using ASAR WS data for the purpose of this study has been proved by comparing the retrieved SM time series with the ground measurements. In fact, high correlation values have been observed both in Spain and Ireland. In contrast, poor agreement has been found by comparing the ASAR SM datasets and in situ
measurements recorded at the Finnish sites. Such outcomes can be possibly due to the unsuitability of the soil moisture retrieval algorithm in accurately processing satellite acquisitions taken at latitudes higher than 60° north. In fact, data are resampled on a regular grid, regardless of the geolocation of the observed areas. Such an approach leads to a less precise backscattering estimation at the Northern latitudes, and therefore it introduces a further error in the retrieval of soil moisture. Indeed, all the dataset inter-comparisons carried out in this study provided the weakest correlations at the Finnish sites. Beside the algorithm related issues, several other possible reasons could be associated to this poor performance. Firstly, the area contained within these ECV cells is dominated by forests. Dense vegetation cover attenuates the backscattered signal and decreases the sensitivity of the radar backscatter to soil moisture [69
]. Secondly, the GTK in situ
soil moisture sensors are buried at a depth of 0.1 m which is beyond the depth at which the satellite is sensitive to surface soil moisture. Nevertheless, no significantly improved correlation values have been found for the FMI Sodankylä ECV cell, where the in situ
sensor is buried at a depth of 0.02 m. Similar low correlations have been observed in [26
] using ASCAT data and in [21
] for the northern latitudes. Conversely, in [70
] relatively high correlations have been evaluated in Norway between both ASCAT (0.68 < R < 0.72) and AMSR-E (0.52 < R < 0.64) SM retrieved data and in situ
measurements taken by sensors buried at a depth of 0.1 m. However, the analysis carried out in the present work displayed reasonable agreement between the ASAR and ECV SM datasets, with R values varying between 0.41 and 0.51 in all the Finnish study areas. In [26
], the authors hypothesize that the poor correlation found between SM measurements taken at the FMI in situ
station and ASCAT retrieved soil moisture values may be partly explained by improper freeze/thaw flagging. They achieved increased correlation by excluding from the dataset comparisons of those measurements recorded in early spring, when soil moisture is still artificially low due to frozen soil. In order to verify whether an inaccurate flagging of the ECV SM product occurred, we exploited the soil temperature information provided together with the soil moisture data in situ
, and we excluded from the validation exercise those observations taken over frozen soil (temperature below 0 °C). However, such an approach did not lead to any significant improvement or variation in terms of correlation. A possible reason could be the small number of observations taken over frozen soil that did not lead to any major change of the SM datasets.
Quite high correlation values between the ECV SM and ASAR SM time series in the Spanish and Irish regions demonstrated the capability of the ECV SM product in describing the soil moisture temporal dynamics, despite its coarser spatial resolution. Even better agreement has been observed between the ECV SM datasets and the ground measurements recorded at the Irish Solohead station and at many of the Spanish sites. Such results are consistent with the outcomes provided in previous works dealing with the validation of satellite SM products by using ground measurements taken at the REMHEDUS stations. For instance, in [25
], the authors compared the earlier version of the ECV SM dataset and in situ
SM measurements recorded at a number of REMHEDUS stations. In [25
], the average correlation was evaluated equal to 0.63 (±0.036). In [27
] results are provided in terms of Spearman correlation, which varies within the interval 0.6–0.7. It is worth noting the fact that the quality assessment of the latest released version of the merged ECV SM product exhibited higher correlations in all the Spanish ECV size pixels. In fact, we found correlation values generally higher than 0.7. Similar results have been observed in previous works where the SM time series recorded at the REMHEDUS stations have been compared to only ASCAT [26
], or both ASCAT and AMSR-E SM products [71
The actual improvement of the ECV SM product has been observed also by carrying out the analysis at the Irish sites, where the SM datasets inter-comparisons provided enhanced results in terms of correlation, with respect to the findings published in [33
]. In [60
], correlation values calculated between ASCAT SM and in situ
values taken in humid regions are similar to those calculated at the Irish sites between ECV SM data (in our work, generated by using only ASCAT acquisitions) and in situ
measurements. Our results are also consistent with those presented in [52
], where the soil moisture retrieved from ASAR GM acquisitions has been compared with that derived from ERS scatterometer data and in situ
The typical seasonal variation of soil moisture has been quite well captured by the in situ
SM time series. By carrying out the comparison between the three SM datasets on a seasonal basis, in [33
] it was observed that the ECV SM product failed in capturing the wettest and driest conditions in Ireland, as the poorest agreement between satellite derived SM time series and ground measurements was found in winter (wettest season) and summer (driest season). Despite the present study has shown an actual improvement of the ECV SM dataset in this regard, the seasonal based analysis confirmed what was previously observed. This is possibly due to the limited number of available satellite images for the retrieval of the soil moisture by applying the change detection algorithm, which is likely to lead to the underestimation of the sensitivity of the microwave signal to soil moisture (the difference between the driest and wettest signal) [73
]. This can explain the low correlation between ASAR and in situ
SM data in winter and summer observed in most of the sites analyzed in this study. On the other hand, the highest correlation values between the satellite SM time series and the in situ
datasets have been achieved in spring and/or autumn, both in Ireland and in Spain. In spite of being located in different climate zones, Southern Ireland and Northern Spain are both characterized by wet winters and dry summer. As discussed in [33
], when heavy and/or continuous precipitation occur over a poorly-drained soil, a water layer could persist on the surface, reducing the satellite microwave backscattering sensitivity to soil moisture and, hence, providing incorrect estimates of the moisture content. Typical poor drained soils are those with a high clay percentage content, such as Solohead in Ireland [33
] (22%) or the REMHEDUS stations M09 (26%) and L07 (33%) within the ECV-A [38
]. In contrast, summer is the driest season, during which the vegetation reaches the maximum growth in Ireland. This may affect the quality of the retrieved soil moisture from SAR and scatterometer images. The seasonal correlation values evaluated for each Finnish site under study are generally very low or even negative. However, there is some consistency with the results achieved in the other areas, as the best agreements have been observed in spring at Kuusamo station, and in autumn in the other sites. Aiming at the understanding of the extent to which geophysical factors, such as soil texture, terrain composition and altitude, affect the retrieved ECV SM product values, the spatial distribution of soil moisture has been investigated at the ASAR scale (1 km) within the ECV size pixels. The approach here adopted makes use of actual observations covering the whole area, differently from classical methods which analyze the soil moisture spatial variability through geostatistical analysis by using hydrological models and in situ
networks over wide regions.
], it was observed that heavier rains and higher mean moisture contents are often associated with lower spatial variability (CV). In principle, the spatial distribution of surface soil moisture content is controlled by environmental attributes, such as land use and topography. In this work we firstly analyzed the possible relationship between the average of soil moisture retrieved from ASAR WS acquisitions over the ECV cells and the CV. In [66
], the authors observed that the SM spatial variability increases over sandy soil as the soil dries, reaching the maximum CV near the residual moisture content. Such behavior is well described by a decreasing power function. However, above SM values of 0.2 m3
, a decreasing linear function is an effective approximation of the relationship between soil moisture and CV values. A linear relationship was found also in [67
], where the authors focused their study about the spatial variability of soil moisture over humid grassland. The more recent work presented in [33
] confirmed what has been previously observed: by focusing on the spatial variability of soil moisture over the Irish grasslands, a decreasing power function was found to be a very good approximation of the relationship between SM and CV (R2
> 0.87). Results achieved by replicating the study over the sandy soil of the semi-arid Duero Basin region are consistent with those described above, with a coefficient of determination higher than 0.8 in all the ECV cells. By comparing the soil moisture variability in wet and dry catchments in New Zealand and Australia, respectively, the study published in [76
] highlighted that the decreasing variability associated to the increasing moisture content, and the increasing variability exhibited at the drier locations, are due to differences in the seasonal patterns of controlling processes associated with seasonal changes in spatial mean soil moisture. While, in [33
], the ability of ASAR WS retrieved SM data to track this full spectrum of varying moisture content and seasonal behavior was proved, in the Spanish sites no evidence of a specific seasonal spatial variability could be observed. Indeed, the driest soil conditions associated to the highest spatial variability (CV = 1.0–1.4) occurred in spring. However, because such extreme low soil moisture occurrences occur only twice along the multi-year period of observation, we could hypothesize that these dry soil conditions were unusual. While winter soil moisture observations are mainly associated to wetter soil conditions and lower CV, summer and autumn ASAR SM values are quite variable as well as the associated SM spatial variability over the ECV size pixels (0.19 < CV < 0.65).
A decreasing power function has been found to estimate quite well the spatial variability of soil moisture associated to its amount over the Finnish sites as well. In Finland soil moisture values are generally lower than 0.2 and more homogenously distributed over the ECV cell (0.19 < CV < 0.4).
The spatial variability of soil moisture has also been investigated by comparing SM time series in each ASAR pixel to the corresponding ECV SM dataset. High correlation (R > 0.5) values have been evaluated all over the four Spanish regions, proving the capability of the coarser spatial resolution product in capturing the soil moisture behavior within quite large areas. Patterns of pixels exhibiting poorer agreement between the datasets (low R) are due to artificial surfaces which have not been accurately masked out in the pre-processing phase (north-west corner), or correspond to areas covered by forests (North and Middle East side), whose presence hinder the accurate retrieval of soil moisture from microwave satellite acquisitions. Similarly, vineyard and fruit trees lead to quite large patches of ASAR pixels characterized by low correlation (R < 0.3) in the ECV-A cell. Correlation maps generated by comparing ASAR SM and in situ
SM datasets in each 1 km × 1 km pixel provided a picture of the representativity of the ground measurements over the ECV size pixels. Despite the larger spatial scale difference led to a slight worsening of the performance, correlation maps are consistent with those generated by using the ECV SM dataset. The correlation maps generated through the ASAR and ECV SM time series comparison at the SAR spatial scale over the Irish sites further demonstrated the enhancement of the CCI SM product with respect to the previous released version [33
]. Although poorer agreement has been noted between ASAR and in situ
SM datasets, the highest and lowest correlation patterns correspond well to those observed by comparing the satellite SM time series.
Issues related to the adopted SM retrieval algorithm, as well as the presence of forests all over the Finnish regions led to generally lower correlation values between the satellite SM products at the ASAR spatial scale. Nevertheless, a rather good agreement between the SM time series is achieved in the FMI cell, where few lower correlation patterns of ASAR pixels occurred in the proximity of the water course crossing the ECV cell and over peat bogs. The correlation map derived from the comparison between ASAR and in situ SM datasets taken at Sodankylä station resulted to be almost complementary to the one generated from the satellite SM time series comparison. The geographic coordinates of the FMI station pinpoint the site in proximity of the river and in the middle of the ECV cell. Therefore, the ground measurements are likely to better represent the soil moisture conditions in those ASAR pixels characterized by the same features as the station site.
The effect of specific land covers on the representativity of the ECV SM product has been highlighted also within the GTK cells, where the presence of peat bogs and the proximity to water bodies lead to lower correlation values associated to the corresponding ASAR pixels.
This paper presented an inter-comparison study of two satellite derived surface soil moisture products with in situ measurements in three European countries belonging to different climate zones. The main objective of this work was to assess the quality of the latest released CCI ECV SM product by investigating its ability to capture the same relative temporal behavior as the finer spatial resolution ASAR SM dataset. Regional (at ECV cell spatial scale) and pixel (at ASAR spatial scale) based analysis have been carried out. In situ soil moisture observations were also used as a reference to verify the accuracy of both the satellite SM products.
Without using any hydrological model or dense in situ networks, the validation activity presented in this work provided results consistent with those published in previous papers using different sensors and classical methods. Such an outcome demonstrated that our approaches are efficient and cost-effective validation techniques for low-resolution SM products.
The study proves that the coarse scale ECV product is representative of the temporal soil moisture variations observed through finer scale ASAR-derived and in situ soil moisture observations at the selected study sites. Strong correlations were observed over humid and semi-arid sites. Specifically, the satellite-derived SM products inter-comparison provided correlation values ranging between 0.70 and 0.73 in the Irish sites, and between 0.58 and 0.80 in all the analyzed Spanish ECV cells. Even better agreement has been observed between the ECV SM datasets and the ground measurements recorded at the Irish Solohead station (R = 0.83) and at many of the Spanish sites (0.70 < R < 0.86). Weaker correlations between ASAR and ECV SM time series were observed over the Finnish sites (R < 0.5), irrespective of the method used for excluding frozen soil condition observations. Poorer results, highlighted by even negative correlation values, have been observed when comparing satellite SM products with ground measurements.
The quality assessment of the ECV SM product through the ASAR pixel-based analysis exhibited R values larger than 0.55 all over the Spanish and Irish ECV cells, where also very high correlation patterns have been observed. Poorer agreement generally occurred over the FMI and GTK regions, where R lies below 0.5, with the exceptions of Ilomantsi and Kuusamo where quite large patterns of ASAR pixels characterized by very high correlation values (0.7 < R < 0.95) have been observed.
The effect of geophysical factors, such as soil type, topography and land cover, on the spatial variability of soil moisture and on the accuracy of satellite-derived soil moisture products, has been also investigated. In terms of soil type, it has been found that less accurate SM values are estimated over soil with higher clay content. Although the areas under study are characterized by quite low complex topography, it has been observed that the quality of the satellite-derived SM datasets decreases over the Irish regions located at higher altitudes [33
]. Concerning the influence of specific land cover on the ECV SM quality, we found that the presence of forests (tall or dense vegetation), peat bogs or the proximity to water bodies, lead to a poorer representativity of the satellite-derived SM product.
On the basis of the overall outcomes of this work, we can state that the ECV SM product is a good representation of the soil moisture condition over the 0.25° × 0.25° cell. Moreover, an improvement of the quality of the latest released version of the ECV SM dataset has been proved by comparing the results reported in the present manuscript with those published in [33
]. However, although providing confidence in the use of the ECV SM product, results and observations presented in this work highlighted also the need of further investigating on the source of error related to SM retrieval algorithms, particularly over high latitude areas (i.e.
, north of 60 deg), as well as on the sensitivity of the accuracy of satellite-derived SM datasets to the soil type and clay content, forest coverage and complex topography. Future studies addressing these issues may benefit from the results presented in this work to be used as a benchmark for a better understanding of soil moisture products.