^{1}

^{1}

^{*}

^{1}

^{2}

^{2}

^{2}

^{3}

This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (

In this paper, the results of a comparison between the soil moisture content (SMC) estimated from C-band SAR, the SMC simulated by a hydrological model, and the SMC measured on ground are presented. The study was carried out in an agricultural test site located in North-west Italy, in the Scrivia river basin. The hydrological model used for the simulations consists of a one-layer soil water balance model, which was found to be able to partially reproduce the soil moisture variability, retaining at the same time simplicity and effectiveness in describing the topsoil. SMC estimates were derived from the application of a retrieval algorithm, based on an Artificial Neural Network approach, to a time series of ENVISAT/ASAR images acquired over the Scrivia test site. The core of the algorithm was represented by a set of ANNs able to deal with the different SAR configurations in terms of polarizations and available ancillary data. In case of crop covered soils, the effect of vegetation was accounted for using NDVI information, or, if available, for the cross-polarized channel. The algorithm results showed some ability in retrieving SMC with RMSE generally <0.04 m^{3}/m^{3} and very low bias (^{3}/m^{3}), except for the case of VV polarized SAR images: in this case, the obtained RMSE was somewhat higher than 0.04 m^{3}/m^{3} (≤0.058 m^{3}/m^{3}). The algorithm was implemented within the framework of an ESA project concerning the development of an operative algorithm for the SMC retrieval from Sentinel-1 data. The algorithm should take into account the GMES requirements of SMC accuracy (≤5% in volume), spatial resolution (≤1 km) and timeliness (3 h from observation). The SMC estimated by the SAR algorithm, the SMC estimated by the hydrological model, and the SMC measured on ground were found to be in good agreement. The hydrological model simulations were performed at two soil depths: 30 and 5 cm and showed that the 30 cm simulations indicated, as expected, SMC values higher than the satellites estimates, with RMSE higher than 0.08 m^{3}/m^{3}. In contrast, in the 5-cm simulations, the agreement between hydrological simulations, satellite estimates and ground measurements could be considered satisfactory, at least in this preliminary comparison, showing a RMSE ranging from 0.054 m^{3}/m^{3} to 0.051 m^{3}/m^{3} for comparison with ground measurements and SAR estimates, respectively.

Soil moisture content (SMC), along with its temporal and spatial distribution, is widely considered as a key variable in numerous environmental disciplines, especially in climatology, meteorology, hydrology and agriculture. For hydrological and agricultural purposes, the SMC plays a fundamental role, since it controls the water available for vegetation growth [

Due to high variability of SMC in time and space, proper estimation of this variable is quite challenging. Ground measurements and remote sensing methods can be considered powerful tools for the SMC quantification. Ground measurements, such as those obtained by using calibrated probes (e.g., those based on Time Domain Reflectometry (TDR) techniques), can provide reliable point-scale measurements and, in case of distributed sensors, can also help in understanding the soil moisture patterns across-scales [

Remote sensing from active (SAR and scatterometer) and passive sensors (radiometers) have demonstrated to be good and flexible tools to detect spatial and temporal SMC [

Regarding the spatial resolution, the SMC estimates from microwave remote sensing can span from tens of meters up to 50 km, whereas, the highest temporal resolution can be achieved with monthly or bimonthly acquisitions. Low spatial resolution estimates can be instead available worldwide on a daily basis. Upcoming sensors such as Sentinel 1 and SMAP will represent a further step to overcome these limitations. Sentinel 1 will work at C-band with a rather high spatial resolution of 5 m × 20 m and the temporal repetition frequency of 5–6 days over the European continent and 12 days for global acquisitions. Moreover, recent studies based on Sentinel 1-like data indicated that the improved radiometric resolution of Sentinel 1 may also produce a reduction in the retrieval errors on SMC [

It is also worthwhile mentioning the NASA Soil Moisture Active Passive (SMAP) mission that will offer the uniqueness of radar and radiometric simultaneous observations at L-band, with a ground resolution of around 1 km and a temporal resolution of 3 days [

It should be noted that all microwave sensors are able to estimate SMC referring to the first few centimeters of soil only. One proposed solution to improve the spatial and temporal resolution of available SMC information and to simulate SMC for deeper soil layers is related to the assimilation of SMC, derived from remote sensing data, into hydrological and land surface models [

Crow and Ryu [

Draper

A necessary step before data assimilation is the comparison between SMC estimated by models and by remote sensing data, in order to verify the compatibility between these two sources of information [

Vischel [^{2}). The first estimates were derived from the physically-based distributed hydrological model TOPKAPI [^{2}, lying between 0.68 and 0.92. In [

In this paper, temporal evolutions of SMC measured on ground, and SMC derived from both SAR data and a hydrological model have been compared to each other, in order to mainly address the temporal compatibility of the two estimated SMC values. This multiple correlation between SMC estimated through SAR data, SMC obtained from the hydrological model, and SMC measured on ground, was carried out with the double purpose of, on one side, testing the ability of these two approaches in simulating the real SMC, and, on the other hand, checking the possibility of using a rather simple hydrological model for spatially and temporally extrapolating SMC, whenever SAR data are missing. The paper is organized as follows. In Section 2, the test sites and the available datasets are described. Section 3 describes the retrieval process used to estimate SMC from SAR images, while Section 4 introduces the proposed hydrological model. Comparison results are discussed in Section 5. Section 6 draws conclusions, possible applications and future works.

The investigation was carried out on the Scrivia test site, which is located in North-West Italy (central coordinates: 45°N, 8.80°E) (^{2}, situated close to the confluence of the Scrivia and the Po rivers. The site is characterized by large homogeneous agricultural fields of wheat, corn, sugarbeet, and potatoes [^{3}/m^{3}, and sunny and dry in summer, with average SMC < 0.15–0.20 m^{3}/m^{3}. According to the crop calendar of this area, in fall (October and November) most fields were bare and with SMC > 0.20–0.25 m^{3}/m^{3}, whereas in spring (March, April) almost half of the agricultural area was covered by growing wheat. The other half consisted of very rough bare fields, waiting for the seeding of corn. In spring SMC was usually rather high (>0.30 m^{3}/m^{3}) due to the frequent rainfall. In May corn was sowed in very smooth fields. These fields were irrigated and therefore their SMC was highly variable. In June–July, the SMC was usually very low (0.10–0.15 m^{3}/m^{3}) due to the absence of rainfall, except in the irrigated fields. Wheat was in the ripening phase in June and harvested at the beginning of July.

ENVISAT/ASAR images were mainly collected from 2003 to 2009 in both HH/HV and VV polarizations and at an incidence angle of 23°. In

The algorithm used for estimating SMC has already been described in [

Studies carried out in the past pointed out that the main constraint for obtaining a good accuracy with ANN approaches is the “robustness” of the training process, which has to be representative of a variety of surface conditions as wide as possible. In order to meet these requirements, the dataset implemented for the ANN training was obtained by combining experimental satellite measurements of backscattering coefficients (σ°) and corresponding ground parameters, derived from the archives available at IFAC and EURAC. Since these datasets were not sufficiently wide for training the ANN and completely setting the neurons and weights, data simulated using electromagnetic forward models have been included in the training set. The backscattering of the bare rough surfaces was obtained using the Advanced Integral Equation Model (AIEM) and Oh model [

After training, the ANNs were tested on a different dataset, obtained by re-iterating the model simulations. The most favorable results were obtained when co- and cross-polarizations were available, showing a determination coefficient (R^{2}) equal to 0.80, and RMSE < 0.04 m^{3}/m^{3} [

In ^{2} = 0.91, RMSE = 0.023 m^{3}/m^{3} (p < 0.05). The aim of _{SAR}, in m^{3}/m^{3}) and SMC measured (SMCmeas, in m^{3}/m^{3}) was obtained:

In this section, the results of a comparison between SMC SAR estimates and some hydrological model simulations are presented with reference to the test site of Scrivia.

We used the simple daily step model, described in [

The model computes soil volumetric water content variations (1/day) at daily time step as:

In the above expressions, θ_{FC} = soil volumetric water content at field capacity (−), θ_{WP} = soil volumetric water content at wilting point (−), PET = potential evapotranspiration (mm/day), L = soil thickness (mm), θ = soil volumetric water content (−), θr = residual soil volumetric water content (−), I_{ex} = infiltration excess, AET is the actual evapotranspiration, and K(θ) is the saturation-dependent hydraulic conductivity. The model assumes that drainage of the topsoil follows gravity only. Moreover, drainage is not allowed to exceed θ − θ_{FC} during one time step. Actual evapotranspiration, AET (mm/day), is:

K_{sat} = saturated hydraulic conductivity, mm/day,

n = exponent in Van Genuchten soil water retention curve model.

Infiltration excess is:
_{sat} and K(Θ) represents an infiltration capacity, which needs to be higher than K(Θ), in order to allow infiltration when soil is in dry conditions. Water in excess of θ_{s} is computed as:

Day by day, soil water content is updated on the basis of the above calculations. Besides the parameters representing physical characteristics of the soil, which can be in principle determined by experimental measurements, the model requires input of the L parameter (soil thickness).

The minimum set of parameters required as input includes precipitation, mean, minimum and maximum temperature at daily steps and an indication of soil texture. Potential evapotranspiration is estimated using the well-known Hargreaves-Samani formula [

Soils in the test area are predominantly loamy-sands (sand 51%, clay 13%, silt 36%) with a mean bulk density of 1.18 kg/L, according to the soil map of Regione Piemonte (

Knowing soil properties, hydraulic parameters can be indirectly estimated using pedotransfer rules or expert systems, such as the popular artificial-neural-network-based ROSETTA (

For the test site of Scrivia, 11 processed satellite images (see

The comparison was carried out by running the hydrological model with parameters for all soil textural classes present in the study area, and by considering the variability of the soil hydraulic parameters estimated by ROSETTA. The latter considered two soil depths: 30 and 5 cm, as depicted in

A comparison between the available data was carried out and shown in _{mod}) was directly compared to SMC measured on ground (SMC_{meas}). Subsequently, a comparison between SMC_{mod} and SMC estimated from SAR data (SMC_{SAR}) was also carried out (

Hydrological model: SMC_{mod} = 0.84SMC_{meas} + 0.044 (R^{2} = 0.55)

SMC_{SAR} = 0.638SMC_{mod} + 0.08 (R^{2} = 0.54)

Considering the low number of available measurements, these correlations have been found significant, with 95% confidence level (p-value).

In ^{2}, slope, RMSE, and p of all the correlation carried out between SMC measured on ground, estimated from SAR data and from the model at two depths (5 and 30 cm) are shown. We can note that the best correlation was obtained by directly comparing SMC_{SAR} and SMC_{meas} and the worst, at least in terms of RMSE, between SMC_{SAR} and SMC_{mod} at 30 cm. It can be observed that the SMC can be better approximated by the hydrological model at 5 cm. The RMSE values range from 0.051 and 0.054 m^{3}/m^{3} for SMC estimated with the hydrological model (at 5 cm depth) and SMC measured on ground and simulated from SAR data, respectively.

Although high spatial resolution products, such as SAR images, usually show a low revisit time, thus hampering their use for simulating soil moisture dynamics, they can be valuable for testing hydrological models and, in particular, hydrological patterns as well as basic assumptions of the model itself. In this view, the test of the soil moisture product with independent hydrological simulations can be considered an interesting result.

It is well known that microwave remote sensing techniques can provide rather accurate estimates of soil moisture content (SMC). However, the SMC obtained in this way only refers to the first centimeter layer of the soil, thus limiting its assimilation into hydrological models.

In this paper, a comparison between SMC obtained from SAR images, through an inversion algorithm based on an Artificial Neural Network (ANN) approach, and SMC estimated from a hydrological model was performed. The outputs of two models were subsequently compared with field measured SMC. The hydrological model estimated SMC of the two different depths: 30 cm and 5 cm. The first output tended to overestimate the SMC values obtained from SAR images, which, as expected, simulated a shallow SMC. The result of the hydrological model for the first 5 cm depth was instead much more in agreement with satellite data. The RMSE values of these comparisons were 0.052 m^{3}/m^{3} for the SMC estimated from the hydrological model and 0.023 m^{3}/m^{3} for the SMC estimated from SAR data.

It is highlighted that products derived from high-temporal frequency satellite images at low spatial resolution have already been used for the assessment of the temporal dynamics of soil moisture. On the other hand, high spatial resolution products, such as those considered in this work, which present lower temporal frequency (and consequently are of limited importance with respect to soil moisture dynamics), may be extremely valuable for testing hydrological patterns and basic assumptions of the models, such as hydrological connectivity and similarity. For these reasons, a deeper investigation on the reliability and compatibility of the soil moisture products derived from SAR images, by using independently derived hydrological simulations, have an important role in hydrological research.

A further comparison between SAR SMC estimates and hydrological model simulations over the Scrivia test site was carried out. The hydrological model reproduced similar values of SMC as compared to the ANN algorithm outputs and ground measurements, provided that the soil layer considered was of the order of only a few centimeters.

The found accuracies of the model simulations, the SAR estimates, and the ground measurements indicate that most of them are within the requested accuracies for satellite products of soil moisture, which, in case of GMES Sentinel-1, is ≤0.05 m^{3}/m^{3}. This result supports the idea that the model simulations may be used as a substitute in case of missing SAR data of dense temporal series or for extending the point-scale measurement of SMC to a more distributed and larger spatial scale.

The comparison conducted in this research can be considered a preliminary exercise, while comparisons with more complex spatially-explicit models should be expanded during future research.

This work was partially supported by the ESA/ESTEC contract n°4000103855/11/NL/MP/fk and by the Italian Space Agency (ASI) through the PROSA project.

The authors declare no conflict of interest.

Map of Northern Italy. The red star represents the test area of the Scrivia.

Soil moisture content (SMC) maps in (m^{3}/m^{3}) obtained through the Artificial Neural Network (ANN) algorithm by using ENVISAT/ASAR images collected on the Scrivia area (central coordinates: 45°N–8.80°E). Masked areas are: white = urban, magenta = forests, dark green = dense vegetation, blue = open water. The dimensions of the images are 20 × 20 km. In the red circle, the area where ground measurements were collected is shown.

SMC estimated by the algorithms (SMC_{SAR}, in m^{3}/m^{3}) on all the available fields of the Scrivia area as a function of the SMC measured on ground (SMCmeas, in m^{3}/m^{3}). The continuous line represents the regression equation of the dataset.

Temporal evolution of SMC (in m^{3}/m^{3}) derived from hydrological model (black points), ground truth (triangles) and ANN algorithm (diamonds) samples, for 30 cm-topsoil (

(_{mod}, in m^{3}/m^{3}) compared to SMC measured on ground (SMC_{meas}, in m^{3}/m^{3}). (_{SAR}, in m^{3}/m^{3}) compared to SMC estimated by the hydrological model (SMC_{mod}, in m^{3}/m^{3}).

ENVISAT/ASAR acquisitions over the Scrivia test site (APP: Alternate Polarization Precision Image, IMP: Image Mode Precision, IMS: Image Mode Single Look Complex).

1 | 7 November 2003 | APP | HH/HV | 2/23° |

2 | 30 April 2004 | APP | HH/HV | 2/23° |

3 | 4 June 2004 | APP | HH/HV | 2/23° |

4 | 22 October 2004 | IMP | VV | 2/23° |

5 | 26 November 2004 | APP | HH/HV | 2/23° |

6 | 11 March 2005 | IMS | VV | 2/23° |

7 | 30 May 2005 | APP | HH/HV | 2/23° |

8 | 31 March 2006 | IMP | HH | 2/23° |

9 | 26 September 2008 | IMS | VV | 2/23° |

10 | 24 April 2009 | IMS | VV | 2/23° |

11 | 29 May 2009 | IMS | VV | 2/23° |

Comparison between the average SMC values estimated from the backscatter of the images of _{SAR}, in m^{3}/m^{3}) and the corresponding SMC values measured on ground (SMCmeas, in m^{3}/m^{3}), averaged on 23 fields.

_{SAR} |
|||
---|---|---|---|

1 | 7 November 2003 | 0.297 | 0.3 |

2 | 30 April 2004 | 0.37 | 0.38 |

3 | 4 June 2004 | 0.12 | 0.15 |

4 | 22 October 2004 | 0.20 | 0.21 |

5 | 26 November 2004 | 0.28 | 0.28 |

6 | 11 March 2005 | 0.26 | 0.31 |

7 | 30 May 2005 | 0.20 | 0.22 |

8 | 31 March 2006 | 0.29 | 0.28 |

9 | 26 September 2008 | 0.17 | 0.17 |

10 | 24 April 2009 | 0.27 | 0.29 |

11 | 29 May 2009 | 0.24 | 0.21 |

Statistical parameters (R^{2}, Slope, RMSE, in m^{3}/m^{3}, and p) of all the performed regression equations between SMC estimated from SAR (SMC_{SAR}) data and from the hydrological model (SMC_{mod}) at two depths (5 and 30 cm), and the SMC measured on ground (SMC_{meas}).

^{2} |
^{3}/m^{3}) |
|||
---|---|---|---|---|

SMC_{SAR}/SMC_{meas} |
0.90 | 0.93 | 0.023 | <0.05 |

SMC_{mod}(5 cm)/SMC_{meas} |
0.55 | 0.84 | 0.051 | <0.05 |

SMC_{mod}(30 cm)/SMC_{meas} |
0.75 | 0.71 | 0.075 | <0.05 |

SMC_{SAR}/SMC_{mod}(5 cm) |
0.54 | 0.64 | 0.054 | <0.05 |

SMC_{SAR}/SMC_{mod}(30 cm) |
0.63 | 0.95 | 0.088 | <0.05 |