Estimating high resolution evapotranspiration from disaggregated thermal images

Remote Sensing of Environment 187 (2016) 423–433

Contents lists available at ScienceDirect

Remote Sensing of Environment journal homepage: www.elsevier.com/locate/rse

Estimating high resolution evapotranspiration from disaggregated thermal images M. Bisquert a,⁎, J.M. Sánchez a, R. López-Urrea b, V. Caselles c a b c

Applied Physics Department, University of Castilla-La Mancha, Polytechnic School and IDR, University Campus, 16071 Cuenca, Spain Instituto Técnico Agronómico Provincial de Albacete and FUNDESCAM, Polígono Industrial Campollano, Avda. 2ª-42 B, 02007 Albacete, Spain Earth Physics and Thermodynamics Department, University of Valencia, C/Dr. Moliner 50, 46100 Burjassot, Spain

a r t i c l e

i n f o

Article history: Received 14 January 2016 Received in revised form 12 October 2016 Accepted 30 October 2016 Available online xxxx Keywords: Disaggregation LST Evapotranspiration Sentinel 2 Spot 5 Take 5 MODIS

a b s t r a c t Accurate evapotranspiration (ET) estimations based on surface energy balance from remote sensing require information in the thermal infrared (TIR) domain, normally provided with an insufficient spatial resolution. In order to estimate ET in heterogeneous agricultural areas, we inspect in this paper the use of disaggregation techniques applied to two different sensors, such as MODIS (daily revisit cycle and 1 km spatial resolution in the TIR domain) and Spot 5 (5 days revisit cycle and 10 m spatial resolution in the VNIR bands but no TIR band). Spot 5 images were used as a proxy for upcoming Sentinel-2. The Simplified Two-Source Energy Balance (STSEB) model was used for the estimation of ET. Since no Sentinel-2 images were available yet, images from the Spot 5 Take 5 experiment were used for testing this approach. Results assessment was conducted at two different levels: field scale (using ground data), and scene scale (using Landsat 7-ETM+ images as a reference). Validation of both disaggregated land surface temperature (LST) and derived surface energy fluxes was performed. Mean absolute deviations of ~2 °C in disaggregated LST were observed at both field and scene scales. At field scale, relative errors of 22% and 19% were obtained for ET at instantaneous and daily scales, respectively. At scene scale, the four components of the surface energy balance equation were obtained with relative errors of 3, 14, 11 and 8% for net radiation, evapotranspiration, sensible heat flux and soil heat flux, respectively, compared to Landsat. The results obtained were compared to the use of the MODIS LST at its original resolution (1 km), which was used also to obtain the surface energy fluxes. As the surface heterogeneity increases the errors in both MODIS LST and ET become more and more significant, compared to the use of the disaggregated images. Although reference images at 10 m spatial resolution were not available at this stage for a more robust comparison, this paper shows the potential of the use of disaggregated LST to estimate ET at 10 m spatial resolution, which is especially attractive in highly heterogeneous areas. © 2016 Elsevier Inc. All rights reserved.

1. Introduction Evapotranspiration (ET) is a key component in hydrological and energy balances models. It is essential for water resources management, as well as in many other disciplines such as hydrology, ecology and climate change (Anderson et al., 2007; Wang and Dickinson, 2012; Katul et al., 2012). Good knowledge of ET can improve the water resources management at different scales; in particular, detailed ET estimations in agricultural areas can improve detection of water stress and help in the irrigation scheduling (French et al., 2015). There exist several techniques to measure in situ ET: eddy covariance systems, weighing lysimeters, Bowen stations, etc. All these are punctual data on a specific location and the extrapolation to large regions is difficult; moreover the instruments needed require an important economic investment.

⁎ Corresponding author. E-mail address: [email protected] (M. Bisquert).

http://dx.doi.org/10.1016/j.rse.2016.10.049 0034-4257/© 2016 Elsevier Inc. All rights reserved.

Remote sensing offers a suitable alternative for regional to global scale applications. Different approaches have been developed for ET estimation from satellite images and meteorological variables. The PenmanMonteith equation is used in the MODIS (Moderate Resolution Image Spectroradiometer) ET product MOD16 (Mu et al., 2007; Mu et al., 2011). The inputs used in this product include: land cover, fraction of vegetation cover, albedo and meteorological variables. Other methods estimate the ET based on the surface energy balance (SEB) equation. Approaches based on the energy balance usually obtain the ET as a residual of the SEB equation. These methods require information in the thermal infrared (TIR) domain, and are then limited by the availability and accuracy of these data. Some other approaches do not use TIR data, although several studies have shown that the use of TIR data can better monitor the short-time changes in the canopy conditions and consequently in ET, than the use of only visible and near infrared (VNIR) data (French et al., 2015). A wide variety of studies have shown the feasibility of using the SEB approach together with thermal remote sensing to estimate surface energy fluxes, including ET (Norman et al., 1995;

424

M. Bisquert et al. / Remote Sensing of Environment 187 (2016) 423–433

Anderson et al., 1997; Bastiaanssen et al., 1998; Allen et al., 2007; Sánchez et al., 2008a; French et al., 2015). The estimation of ET in agricultural areas from remote sensing data is mainly limited by the sensors spatial resolution. This is especially remarkable when using TIR data, because the images in the TIR domain are provided with a lower spatial resolution than in the VNIR domain onboard the same sensor. In order to solve this problem, several fusion and disaggregation techniques have been proposed to downscale the TIR data to the VNIR spatial resolution within a particular sensor (Kustas et al., 2003; Agam et al., 2007). Beyond this, some recent works have explored the possibility to downscale TIR data from low resolution sensors (SEVIRI, MODIS, AATSR, or the coming Sentinel-3) to the spatial scale of medium-high resolution sensors (ASTER, Landsat, Sentinel-2) (Bindhu et al., 2013; Bisquert et al., 2016). Bisquert et al. (2016) analyzed several disaggregation methods applied to Landsat and MODIS images in central Spain. The assessment included methods using classical approaches (linear and quadratic regression), data mining and neural networks; a simple method based on the linear adjustment between land surface temperature (LST) and NDVI (Normalized Difference Vegetation Index) led to the best results. An analysis per land cover occupation was also performed. The Landsat TIR bands were used as the reference data for the validation. The use of disaggregated temperature to estimate ET has been analyzed in several works (Bindhu et al., 2013; Corbari et al., 2015). However, the disaggregation method used in Bindhu et al. (2013) is better suited for areas with high vegetation cover, which is not the case of our study area. This method was already tested in Bisquert et al. (2016) leading to poor results. In Corbari et al. (2015) the disaggregation was applied to MODIS and SEVIRI sensors separately, so the disaggregated images had spatial resolutions of 250 m and 1000 m, respectively. These spatial resolutions are not suited for agricultural areas with small plots. An alternative to the use of disaggregation techniques are the fusion methods, such as the widely used Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM) (Gao et al., 2006). STARFM is based on the spectral and spatial relationship between an existing high/low resolution image pair. This relationship is later used to complete the time series of high resolution images for dates when only low resolution images are available. This method has been widely used for fusing VNIR images (Senf et al., 2015; Schmidt et al., 2015; Doña et al., 2015). Recently, Cammalleri et al. (2013) investigated the use of the STARFM for fusing daily ET retrieved from multiple TIR sensors. The performance of this methodology was further assessed in different fields, distinguishing between irrigated and rainfed fields (Cammalleri et al., 2014). The same methodology was applied to two vineyards in Semmens et al. (2015). Results from these studies were validated at daily timescale using ground data. An optimum remote sensing system for agricultural applications would provide data as often as twice per week for irrigation scheduling and once every two weeks for general crop damage detection or crop production (Moran et al., 1997; Moran, 2000). Many authors have dealt with this issue in recent literature (Zarco-Tejada et al., 2005; Dorigo et al., 2007; Roy et al., 2008; Amorós-López et al., 2013; Inglada et al., 2015; etc.). The recently launched Sentinel-2a enhances the temporal and spatial resolution of the current satellites, offering a 10-day revisit cycle with 10–30 m spatial resolutions. With the coming Sentinel-2b the combination of both sensors will offer a 5-day revisit cycle. The constraint is that no thermal sensor is present in this satellite. Thus, the fusion methods cannot be applied to this satellite; contrarily it is possible to apply the disaggregation techniques, as shown in Bisquert et al. (2016). The main objective of this paper is to assess the use of disaggregated LST combining two different satellites to estimate ET from a high resolution sensor provided with no thermal band. There is interest in the scientific community in the application to Sentinel-2 because of its temporal and spatial resolution. While waiting for the availability of Sentinel-2 scenes, images from the Spot-5 Take 5 experiment simulating Sentinel-2 during summer 2015 were used in this work. A first

assessment of the disaggregated LSTs obtained from MODIS and these Spot-5 images was performed at field scale by means of ground data measured in different plots. Also, a comparison between MODIS LST and ground data was performed to see the improvement of the disaggregated LST face to the use of MODIS LST. Then, a scene scale assessment was conducted using Landsat TIR data as a reference. Subsequently, the ET was estimated using the disaggregated LSTs on one hand, and using the MODIS LST on the other hand, and results were validated with ground data from two different plots at field scale, and using Landsat ET images as a reference at scene scale.

2. Materials and methods 2.1. Study site and measurements The study area is located in Barrax, central Spain, including “Las Tiesas” experimental farm (39°03′35″N, 2°06′W). This is a traditional ESA (European Space Agency) test site used in different international campaigns: SEN2FLEX (SENtinel-2 and Fluorescence Experiment, Sobrino et al., 2008), SPARC (SPectra bARrax Campaign, Moreno et al., 2004), ImagineS (Implementing Multi-scale Agricultural Indicators Exploiting Sentinels, Latorre et al., 2014), and DAISEX (Digital Airborne Imaging Spectrometer EXperiment, Berger et al., 2001). This is a patchy agricultural flat area and is frequently used in agriculture water management studies (López-Urrea et al., 2006; López-Urrea et al., 2009; Sánchez et al., 2011; López-Urrea et al., 2012; Sánchez et al., 2014). Two weighing lysimeters are available in this experimental farm, installed in a grass plot and in a vineyard, with continuous data acquisition. The grass plot is used to monitor the reference evapotranspiration and is maintained in optimum growth conditions (López-Urrea et al., 2006). A meteorological station is set up in the grass plot, recording air temperature, relative air humidity, long and short wave radiation and atmospheric pressure. More details on the study area, the plot, the lysimeter and the meteorological station can be found in López-Urrea et al. (2006). The second lysimeter used in this work was installed in a nearby vineyard, where an additional meteorological station was assembled, together with a Hukseflux NRO1 sensor measuring net radiation at 4-m height. All data were collected and stored in 15-min averages. Ground radiometric temperatures were measured in the grass and vineyard fields, besides three additional test plots (poppy, barley and maize), concurrently with daytime, cloud-free MODIS overpasses during the summer of 2015 (Fig. 1) (DOYs: 127, 132, 134, 141, 173, 182, 189, 198, 214 and 237). Overpass time for all dates is around 11 am (UTM). A resume of the meteorological conditions for each date is shown in Table 1. The barley and maize plots occupy a 40 ha pivot field covering each crop 50% of the area. The vineyard, grass and poppy fields cover areas of 3, 1 and 0.6 ha, respectively. LST transects were carried out at the five different plots (Fig. 1). A set of portable infrared thermal radiometers (IRTs) Apogee MI-210 were used. These are single band (8–14 μm) instruments with an accuracy of ± 0.3 °C and a field of view of 22°. In order to capture the spatial variability of the surface within each field, the radiometers were carried back and forth along transects covering large areas within each plot, and finally the average LST and its standard deviation are obtained for each field. In the vineyard field, which is a row crop, special care was taken in the measurements, pointing the radiometer to an area including both, vine plants and soil surrounding. As expected, higher standard deviations are observed in the vineyard LST measurements. Measurements were made at a rate of N5 measurements per minute and data were collected during periods of 15–20 min centered at the satellite overpass time to assure a good stability of the IRTs response. Measurements of the downward sky radiance, required for the atmospheric correction, were performed with each radiometer at the start and end of the temperature transects. Also, information on irrigation events was registered.

M. Bisquert et al. / Remote Sensing of Environment 187 (2016) 423–433

425

Table 2 Spectral range of the common bands in Spot-5 and Sentinel-2.

Green Red NIR SWIR

Fig. 1. Overview of “Las Tiesas” experimental farm, together with pictures of the different field plots taken early in the season.

2.2. Satellite images Data from the Spot-5 Take 5 project, provided by CNES and ESA (https://spot-take5.org/) were used in this work to simulate Sentinel2 images. This is a second experiment following the Spot-4 Take 5 experiment (Hagolle et al., 2015b) with the same aim. The four VNIR bands of the Spot sensors are similar to four bands present in Sentinel-2 (see Table 2). The Spot-5 level 2A images are provided atmospherically corrected with 10 m spatial resolution and 5-day revisit cycle. A cloud mask is also included. The atmospheric correction and the cloud mask are performed by the CESBIO (Centre d'études spatiales de la biosphère) using the MAACS (Multi-sensor Atmospheric Correction and Cloud Screening) algorithm (Hagolle et al., 2015a). Spot images have four VNIR bands but no TIR band. MODIS images were used to disaggregate LST from its 1 km spatial resolution to the Spot 10 m resolution (the disaggregation procedure is described in detail in Section 2.3). The selected dates were those with low cloud coverage and also low MODIS viewing angle in order to avoid errors linked to angular effects. MODIS images are provided atmospherically corrected and include a quality band with information on each pixel quality. VNIR data were obtained from the MOD09GA (250 m) and MOD09GQ (500 m) products (http:// reverb.echo.nasa.gov). TIR data were obtained from MOD11_L2 product provided at 1 km spatial resolution. All the images were masked using their corresponding quality band (QA). In the MOD09 images, pixels with low quality and the medium quality pixels affected by clouds, shadows and high aerosol levels were discarded. In the MOD11 images, pixels not produced or affected by cirrus, clouds or nearby clouds were discarded. The accuracy of LST in the MOD11_L2 product is better than 1 K (Coll et al., 2005; Coll et al., 2009).

Spot 5 (nm)

Sentinel 2 (nm)

500–590 610–680 780–890 1580–1750

543–578 650–680 785–900 1565–1655

The disaggregation method applied to a pair of two different sensors needs equivalent spectral data from both sensors. In order to make Spot and MODIS VNIR bands spectrally equivalent, a normalization procedure was applied (Bisquert et al., 2016) prior the disaggregation of the MODIS LST to the Spot spatial resolution. Spot and MODIS images were first aggregated to 1 km spatial resolution and a linear regression between both was obtained. This equation was then applied to the Spot 10 m spatial resolution images. Landsat 7 ETM+ images were used as a reference for the validation of the disaggregated LSTs and the surface energy fluxes at full scene scale. Landsat TIR band (band 6) was aggregated to 60 m spatial resolution (this is its original resolution although it is provided at 30 m). The aggregation of thermal data was based on the Stefan-Boltzmann law with the assumption of similar emissivity values for adjacent pixels as suggested by Gao et al. (2012). Then, band 6 was atmospherically corrected using the Atmospheric Correction Tool of (Barsi et al., 2003; Barsi et al., 2005) (http://atm-corr.gsfc.nasa.gov/), and the LST was obtained following the procedure described in Sánchez et al. (2008b). For the VNIR bands the Landsat Surface Reflectance (CDR) product was used. In order to minimize the difference due to the time lapse between both sensors (Bisquert et al., 2016) Landsat LSTs were also normalized using the MODIS LST, applying the same procedure described before for MODIS and Spot. For the comparison of Spot and Landsat, Spot results were aggregated to Landsat spatial resolution by averaging all the Spot pixels within each Landsat pixel. The aggregation of LST was based on the Stefan-Boltzmann law with the assumption of similar emissivity values for adjacent pixels. 2.3. Disaggregation method In Bisquert et al. (2016) several disaggregation methods were assessed with pairs of Landsat and MODIS images in the same area of Barrax and using Landsat TIR data for validation. A modification of a method presented by Agam et al. (2007) showed the best results. This is a simple method based on the linear relationship between NDVI and LST at the coarse resolution image. The disaggregation was applied here as follows. First, the coefficient of variation (CV, computed as the standard deviation divided by the mean) is obtained from the 4 × 4 pixels belonging to each 1000 m pixel as suggested by Kustas et al. (2003). The 25% of the most homogeneous pixels (those with a lower CV) were used to perform the linear adjustment between LST and NDVI (Eq. (1)) for each image. LST ¼ a0 þ a1 NDVI

ð1Þ

A residual correction was included in order to take into account local conditions and to correct the possible deviations produced by the equation (Kustas et al., 2003). This correction consisted in aggregating the

Table 1 Resume of the meteorological conditions observed for the measuring dates at the MODIS overpass time: air temperature (Ta), relative humidity (Hr), wind speed (u) and solar global radiation (S). DOY (date)

127 (05/07)

132 (05/12)

134 (05/14)

141 (05/21)

173 (06/22)

182 (06/29)

189 (07/08)

198 (07/17)

205 (07/24)

214 (08/02)

237 (08/25)

Ta (°C) Hr (%) u (m s−1) S (W m−2)

23.6 40 3.9 999

27.7 36 1.6 984

32.9 22 6 960

15.2 52 1.7 1021

29.3 39 1.3 992

31.5 27 4.5 973

34.7 23 2.2 973

33.2 29 1.5 961

32.3 58 1.4 882

27 63 3.5 963

18 43 1.8 916

426

M. Bisquert et al. / Remote Sensing of Environment 187 (2016) 423–433

high resolution estimated LSTs to the MODIS TIR spatial resolution, and a residual was obtained as the difference between the estimated and the original MODIS LSTs at 1 km. Then, the adjusted equation was applied to the high resolution NDVI images, and the residuals were added to the previous result. Note that the residuals are obtained at coarse resolution, thus the same residual is applied to all the high resolution pixels belonging to the same coarse resolution pixel. Since these coarse residuals produced a boxy effect, they were smoothed using a Gaussian filter as suggested by Anderson et al. (2004) and as it was done in Bisquert et al. (2016). The disaggregation procedure was ideally applied to pairs of concurrent Spot and MODIS images; however, when this was not possible (for example, because the MODIS images acquired at the Spot overpass date had a very large viewing angle (N40°) or because there were some clouds at the MODIS overpass time), close in time images were used under the assumption of minimum changes in NDVI. Moreover, the previously applied normalization procedure reduced possible differences at this point. 2.4. Evapotranspiration estimation In this work we used the Simplified Two-Source Energy Balance (STSEB) model (Sánchez et al., 2008a) to estimate instantaneous ET as a residual from the SEB (Eq. (2)). Rn ¼ λET þ H þ G

ð2Þ

where Rn is the net radiation flux (W m− 2), G is the soil heat flux (W m−2), H is the sensible heat flux (W m− 2) and λET is the latent heat flux that represents the energy required for ET (W m−2). The instantaneous net radiation (Rn) can be obtained from the balance between incoming and outgoing radiation: Rn ¼ ð1−αÞS þ εLsky −εσLST4

ð3Þ

where S is the solar global radiation (W m−2), LST is the radiometric land surface temperature, α is the surface albedo, ε is the surface emissivity, σ is the Stefan-Boltzmann constant and Lsky is the incident longwave radiation (W m−2). The instantaneous soil heat flux (G) is obtained as a fraction of the net radiation (Choudhury et al., 1987): G ¼ CG ð1−Pv ÞRn

ð4Þ

where Pv stands for the fractional vegetation cover. In this work a value of CG = 0.275 was used, according to Sánchez et al. (2008b). Calculation of sensible heat flux (H) is done by adding the soil and canopy components, weighted by the fractional vegetation cover: H ¼ Pv Hc þ ð1−Pv ÞHs

ð5Þ

where Hc and Hs are obtained as follows:

  NDVI 1− NDVIs    Pv ¼  NDVI NDVI −K 1− 1− NDVIs NDVIc

ð7Þ

where the coefficient K is obtained by: K¼

RNIRc −RREDc RNIRs −RREDs

ð8Þ

where RNIR is the near-infrared reflectivity, and RRED is the red visible reflectivity. The subscripts c and s correspond to canopy (completely vegetated) and bare soil (unvegetated) areas, respectively, selected by establishing statistical thresholds in the NDVI images. In this work, NDVI values N 0.8 and b 0.15 were initially set to select canopy and soil areas, respectively. In case the number of pixels was not representative enough, the corresponding threshold was reduced or increased by 0.05 until no b 1% of the total pixels in the scene were available for each category. These selected areas were also used to estimate Tc and Ts, required in Eqs. (6a) and (6b), respectively, from the LST maps (Sánchez et al., 2008b). Mean values of the representative pixels distributed around the whole scene were used for the soil and canopy parameters to account for the possible spatial variability within the area. Daily surface energy fluxes were also considered in this paper. Daily values were obtained using the methodology described in Sánchez et al. (2008b), based on Seguin and Itier (1983) relationship between diurnal and instantaneous (subscripts d and i, respectively) H and Rn. Hd H ¼ i Rnd Rni

ð9Þ

Considering that daily G can be neglected in Eq. (2), the daily λET (λETd) can be estimated as follows in cloud free days: λETd ¼

Rnd ðR −Hi Þ Rni ni

ð10Þ

Rnd / Rni values in Eq. (10) were calculated for each date from the ground-collected data. Since the ratio Rnd / Rni has been shown to vary with the time, date, or the site latitude, but not with the vegetation type (Sobrino et al., 2005; Sánchez et al., 2007), a constant value was used for each image. The meteorological variables required as inputs in the STSEB model were registered by an in situ weather station. 2.5. Performance assessment

Tc −Ta Hc ¼ ρCp rha

ð6aÞ

Ts −Ta raa þ rsa

ð6bÞ

Hs ¼ ρCp

speed, u (m s−1), plays an important role in the estimation of the aerodynamic resistances. For a comprehensive description of the aerodynamic resistance equations the reader is referred to Sánchez et al. (2008a). The fractional vegetation cover was obtained through the expression (Valor and Caselles, 1996)

where ρCp is the volumetric heat capacity of air (J K−1 m−3), Ta is the air temperature at a reference height (K), Tc and Ts are the canopy and soil radiometric temperatures, respectively, rha is the aerodynamic resistance to heat transfer between the canopy and the reference height (s m−1), raa is the aerodynamic resistance to heat transfer between the point z0 + d (z0: roughness length, d: displacement height) and the reference height (s m− 1), rsa is the aerodynamic resistance to heat flow in the boundary layer immediately above the soil surface (s m−1). The wind

The performance assessment was conducted in terms of disaggregated LST and ET, at two different scales. The comparison between estimated and reference data was performed in terms of several statistical metrics, including classical statistical measures, such as: mean bias (Bias) and its standard deviation (SD), mean absolute deviation (MAD), mean absolute deviation in percentage (MADP), root mean square deviation (RMSD) and coefficient of determination (r2). Following Niclos et al. (2016) and Schneider et al. (2012) other statistics considered more robust and less influenced by outliers were also included. These are the median bias (Me), robust standard deviation (RSD) and robust RMSD (R-RMSD). The skewness and kurtosis were also included, which quantitatively describe the distribution of the differences between the estimated and observed values. The median bias and the

M. Bisquert et al. / Remote Sensing of Environment 187 (2016) 423–433

robust statistics (RSD and R-RMSD) minimize the effect of outliers and are considered as more consistent statistics than the classical bias, SD, and RMSD (Niclos et al., 2016). 2.5.1. Field scale assessment The field scale assessment was performed using the ground data as reference. Average values of the disaggregated LSTs were compared to the ground-measured LSTs in the five different plots selected (Fig. 1). Previously, the radiometric temperatures were corrected for atmospheric and emissivity effects using the radiative transfer equation adapted to ground measurements (Sánchez et al., 2008a). Emissivity values from Rubio et al. (2003) were used in this work. We considered only the temperatures measured within 5 min around the sensor overpass time for comparison with the satellite LSTs. These data were averaged for each transect/radiometer and the standard deviation was calculated. The ET assessment was performed only in the two plots (vineyard and grass) where ground ET measurements were available. In addition, comparison between MODIS LST and ground data was conducted, in order to show the improvements when using disaggregated LST. MODIS LSTs were finally used as inputs in the STSEB model to evaluate these improvements also in terms of evapotranspiration and net radiation. 2.5.2. Full scene assessment Assessment of the entire disaggregated images was performed using Landsat images concurrent with the MODIS overpasses (note that the disaggregated LST corresponds to the MODIS overpass time). Previously, Landsat spectral bands were normalized to those corresponding to MODIS, TIR band included, in order to minimize the difference in the spectral characteristics and the overpass time (10–15 min difference) between both sensors. The disaggregated LSTs (Spot-MODIS) and surface energy fluxes (obtained at 10 m) were aggregated to the equivalent 60 m Landsat pixels for the comparison by averaging the values of all the high resolution pixels belonging to the same Landsat pixel, in the case of LST the aggregation was based on the Stefan-Boltzmann law as suggested by Gao et al. (2012). The normalized Landsat LST was used to validate the disaggregated LST at scene scale. Also, the Landsat LST was used to obtain the surface energy fluxes later used to validate the fluxes obtained from the disaggregated Spot images. The comparison between disaggregated and Landsat images was performed using the same statistical metrics as in the comparison between the disaggregated results and ground data. In addition, comparison between MODIS and Landsat was conducted also. 3. Results and discussion 3.1. Field scale assessment 3.1.1. Disaggregated LST The field scale LST assessment was performed in the five different plots using the ground LST values measured with the IRTs in 10 dates, as described in Section 2.5.1. Fig. 2a shows the agreement between MODIS and observed LST values, whereas Fig. 2b shows the agreement between disaggregated and observed LST values. Horizontal error bars correspond to the standard deviation of the ground LST measurements. Vertical error bands in Fig. 2a are not available since there is only one MODIS pixel covering each field. Vertical error bands in Fig. 2b represent the standard deviation of the disaggregated LST pixels corresponding to each field. The number of pixels used for averaging each field in the disaggregated images is: 61, 116, 277, 1960 and 2024 for poppy, grass, vineyard, barley and maize respectively. A better agreement is observed between disaggregated and ground data than between MODIS and ground LST. Even so, several disaggregated LST values overestimate ground measurements as much as 10 °C. These large discrepancies are linked to irrigation events. After a water supply, the surface cools down and the ground-measured LST is significantly lower than that

427

predicted by the disaggregation technique based on the non-affected NDVI. This deviation should be mitigated by the residual correction, but the large heterogeneity present in a coarse resolution pixel in this study site and the big ratio between the MODIS pixel size and the field area make this task very difficult. When these irrigated points are discarded from the analysis, significantly better assessment between disaggregated and ground LSTs is observed (Fig. 2c and d). A bias of 0.2 °C and a RMSD of ±2.4 °C are obtained. Table 3 lists several statistics for the differences between disaggregated and measured in situ LST, and between MODIS and ground LST for the dates and plots where fluxes data were also available. A bias of 0.6 °C and a RMSD of ±3 °C were obtained with the disaggregated images focusing on this dataset. The differences observed in Table 3 between the SD and the RSD, and between RMSD and R-RMSD, are due to the presence of outliers. Moreover, the kurtosis values (~−1) indicates a behavior close to the normal distribution, while the very minor skewness observed indicates a difference distribution closely centered at 0. Errors are significantly larger for the comparison between MODIS and ground data, with a bias of ±3.4 °C and a RMSD of ±7 °C.

3.1.2. Surface energy fluxes The assessment of the surface energy fluxes focused on the two plots, grass and vineyard, where flux instrumentation was available. The same previous set of 10 dates, for the period May–August 2015, was used in this study. Data from the lysimeters in each field were used for the ET assessment. The scatter plot in the top row of Fig. 3a shows the comparison between observed and modelled instantaneous fluxes from the disaggregated images. Ground data were sampled at the MODIS overpass time. The low row shows the results of the comparison when using the MODIS LST as input in the STSEB model. The vertical error bars correspond to the standard deviation of the pixels selected for each field. Note a delay in the installation of the net radiometer in the vineyard limited Rn dataset to 7 dates. Overall, modelled Rn and ET agree well with ground measurements when using the disaggregated images, whereas the plots obtained when using the MODIS LST images show greater scatter. Table 3 summarizes the quantitative analysis of the energy flux comparison both between disaggregated and ground data and the results obtained from the MODIS LSTs versus ground data. Relative errors of 3% and 22% were obtained at instantaneous scale for Rn and ET, respectively, when using the disaggregated images, whereas errors of 5% and 31% for Rn and ET, respectively, were obtained using the MODIS LST. Comparison between observed and disaggregated daily Rn and ET is shown in Fig. 3b and c, respectively. Statistics of this comparison are also included in Table 3, as well as the comparison between the use of MODIS LST and the ground data. At daily timescale, relative errors in Rn were similar to that at instantaneous scale, while errors in ET were reduced to 19% when using the disaggregated LST and to 28% when using the MODIS LST. Estimation errors for ETd estimated using the disaggregated LST are in agreement with those obtained by other authors. Cammalleri et al. (2014) showed relative errors of 20–27% using fusion techniques with 11 images acquired in 2003 and 2008 in two different sites, a shrubland with small fields of ~5 ha and a mead site with fields of ~50 ha. Semmens et al. (2015) obtained relative errors of 19–23% in the estimation of ETd in two large vineyards (21–34 ha) using a spatial resolution of 30 m with a total of 22 images acquired through 2013. Note that, in the present paper, the vineyard plot is only 3 ha and ET was modelled at 10-m spatial resolution, which is an additional challenge compared to previous works.

3.2. Image scale assessment The full scene validation was performed for six dates with concurrent Landsat 7 and MODIS images (DOYs: 141, 173, 198, 205, 214 and 237).

428

M. Bisquert et al. / Remote Sensing of Environment 187 (2016) 423–433

50

60

LSTMODIS= 0.3 LST + 30 N=48

Disaggregated LST (°C)

MODIS LST (°°C)

60

40 30 20 10

40 30 20 10

10

60

20

30

40

50

60

20

30

40

50

Ground LST (°C)

(a)

(b) 60

LSTMODIS= 0.2 LST + 31 N=29

50

10

Ground LST (°C)

Disaggregated LST (°C)

MODIS LST (°C)

50

LSTdis= 0.7 LST + 14.3 N=48

40 30 20 10

50

60

LSTdis= 0.9 LST + 4.6 N=29

40 30 20 10

10

20

30

40

50

60

10

20

Ground LST (°C)

(c) Grass

30

40

50

60

Ground LST (°C)

(d) Poppy

Vineyard

Barley

Maize

Fig. 2. MODIS (a) and disaggregated (b) LST versus ground measured LST including the full dataset, and excluding those points after an irrigation event (c) (d).

3.2.1. Disaggregated LST The comparison between disaggregated and Landsat LST was carried out at the original Landsat 7 TIR spatial resolution (60 m) by aggregating the high resolution LST images based on the Stefan-Boltzmann equation as suggested by Gao et al. (2012). Table 4 shows the quantitative analysis of the differences between disaggregated and Landsat LST including the data from all the dates. The comparison between Landsat and MODIS LST is also included. In general, both mean (bias) and median deviations were lower than 1 °C. Kurtosis values higher than 1.5 indicate that a dataset differs from

the normal distribution. In these cases the classical statistics are nonmeaningful (Vancutsem et al., 2010) and robust statistics which are less sensitive to outliers should be preferred. The R-RMSD obtained with the disaggregated images was ± 1.4 °C, an analysis per dates showed that the R-RMSD were equal or lower to ± 2 °C in all dates. Comparing Landsat to MODIS LST led to an R-RMSD of ±1.6 °C which is not very different from that obtained with the disaggregated images. This is due to the fact that the Landsat scene contains large areas highly homogeneous, so the overall difference between Landsat and MODIS is not so evident. The differences between using MODIS or disaggregated

Table 3 Quantitative analysis of the differences between the use of disaggregated LST or MODIS LST, and ground-measured variables in the vineyard and grass fields. Surface energy fluxes are assessed at both, instantaneous (Rn, ET) and daily (Rnd, ETd) scales. The statistics include: mean bias (Bias), standard deviation (SD), mean absolute deviation (MAD), Mean Absolute Deviation in Percentage (MADP) obtained as the MAD divided by the mean observed value, root mean square deviation (RMSD), coefficient of determination (r2), median bias (Me), robust standard deviation (RSD, RSD=Me(|xi −Me(xi)|)*1.4826, where x = LSTdisag − LSTground), robust RMSD (R−RMSD=(Me2 +RSD2)1/2), skewness and kurtosis. Variable

N Bias SD MAD MADP RMSD r2 Me RSD R-RMSD Skewness Kurtosis

Rn (W m−2)

LST (°C)

ET (W m−2)

Rnd (W m−2)

ETd (mm day−1)

Disag.

MODIS

Disag.

MODIS

Disag.

MODIS

Disag.

MODIS

Disag.

MODIS

16 0.6 3 3 7 (%) 3 0.84 0.91 3 3 −0.15 −1.3

16 3.4 6 6 17 (%) 7 0.32 6 4 7 −0.7 −0.9

16 6 30 20 3 (%) 27 0.73 10 8 12 −0.9 2.0

16 −15 40 30 5 (%) 40 0.28 −14 26 29 −0.8 0.6

16 15 90 80 22(%) 90 0.67 10 60 60 −0.3 −0.08

16 −70 120 110 31 (%) 130 0.59 −60 70 90 −0.4 0.9

16 2.1 7 6 3 (%) 7 0.92 3.2 5 6 −0.7 1.4

16 −3.5 10 8 5 (%) 11 0.70 −3.1 6 7 −0.5 −0.1

16 −0.4 1.1 0.9 19 (%) 1.2 0.60 −0.4 0.8 0.9 −0.8 1.7

16 −1.0 1.4 1.3 28 (%) 1.7 0.41 −0.9 0.9 1.2 −0.4 −0.2

M. Bisquert et al. / Remote Sensing of Environment 187 (2016) 423–433

429

Fig. 3. Disaggregated (top row) and MODIS (low row) versus ground measured net radiation (Rn) and evapotranspiration (ET) of: a) Instantaneous Rn and ET, b) daily Rn, c) daily ET.

LST are stressed at small field scale as seen above. Fig. 4 shows the comparison of the errors in LST obtained from both disaggregated and MODIS images, versus the CV calculated for the MODIS pixels. Similar LST errors are observed focusing on the most homogeneous pixels (CV b 0.1). However, as the CV and the heterogeneity increases, the errors in MODIS LST become more and more significant compared to the errors from the disaggregated images. Fig. 5 shows an example of the LST at three different resolutions: 10 m (disaggregated images), 60 m (Landsat and disaggregated) and 1000 m (MODIS). The enhancement of both disaggregated and Landsat vs the MODIS is more than evident since the MODIS images do not capture the heterogeneity within the area. For example, note the pivots in dark red showing low temperature values in both, disaggregated and Landsat images. These low LST values are not captured though in the MODIS images. For a better understanding of the LST deviation, images of LST differences were computed, using Landsat as a reference. Fig. 5e shows the good performance of the disaggregated LST with dominant differences ranging ±5 K. On the contrary, large LST deviations are observed in the MODIS-Landsat difference image (Fig. 5f), with LST errors up to ±10 K.

3.2.2. Surface energy fluxes Comparison between disaggregated and Landsat, in terms of the surface energy fluxes, was performed at 60 m spatial resolution too. In this case, the surface energy fluxes from the MODIS-Spot disaggregated images were previously calculated at 10 m spatial resolution, and aggregated afterwards to the Landsat 60-m spatial resolution. This aggregation was done by averaging the values within a 6 × 6 10-m pixel window matching the 60-m Landsat pixels. Table 4 shows the statistics for the quantitative analysis of the differences between disaggregated and Landsat-based surface energy fluxes for the six dates available. Average relative errors result 3, 14, 11 and 8% for Rn, ET, H and G, respectively, when using the disaggregated images and 5, 27, 22 and 9% when using the MODIS LST. The kurtosis values varied with dates and variables, while the skewness values kept low in all cases, only in few cases it was N 1. The average R-RMSD values for the six dates are 16, 26, 23 and 7 W m−2 for Rn, ET, H, and G, respectively, when using the disaggregated images, and 20, 30, 30 and 10 when using the MODIS LST. Again, the improvements in the estimation of the surface energy fluxes from disaggregated LST, compared to MODIS LST, are stressed at

Table 4 Statistics of the differences between the surface energy fluxes obtained from the disaggregated and the Landsat 7 images including all dates (a total of 2,350,000 data). For description of the variables see caption of Table 3. Variable

Mean SD RMSD MAD MADP r2 Me RSD R-RMSD Skewness Kurtosis

Rn (W m−2)

LST (°C)

ET (W m−2)

H (W m−2)

G (W m−2)

Disag.

MODIS

Disag.

MODIS

Disag.

MODIS

Disag.

MODIS

Disag.

MODIS

−0.17 2.6 2.6 1.9 5 (%) 0.87 −0.3 1.4 1.4 0.5 3

−0.2 3.1 3.1 2.2 5 (%) 0.78 −0.4 1.6 1.6 0.9 3.9

0.6 26 26 20 3 (%) 0.84 0.9 15 16 −0.3 1.9

−8 40 40 29 5 (%) 0.64 −2 20 20 −0.8 1.3

−21 50 50 30 14 (%) 0.88 −15 22 26 −1.2 5.6

−16 90 90 60 27 (%) 0.52 −3 30 30 −0.8 3.2

23 29 40 27 11 (%) 0.87 19 14 23 1.1 5.8

15 80 90 50 22 (%) 0.20 19 23 30 −0.8 3.7

−1.2 13 13 9 8 (%) 0.87 −3 6 7 1.4 4

−7 11 13 11 9 (%) 0.87 −8 6 10 0.5 3.8

430

M. Bisquert et al. / Remote Sensing of Environment 187 (2016) 423–433

4.5

40 LST disaggregated LST MODIS ET disaggregated ET MODIS

3.5

35 30

3 2.5

25

2

20

MAPD ET (%)

R-RMSD LST (º C)

4

1.5 15

1 0.5

10 0

0.2

0.4

0.6

CV Fig. 4. Evolution of the errors in LST and ET obtained from both disaggregated and MODIS images, with the CV calculated for the MODIS pixels.

small field scale and at those areas where MODIS pixels are highly heterogeneous. Fig. 4 includes the errors in ET obtained from both disaggregated and MODIS images, versus the CV calculated for the MODIS pixels. The same behavior described above for the errors in LST applies now for ET. This proves the enhancement of the results when using the

disaggregated images in areas where MODIS pixels are heterogeneous. Moreover, we observe that errors in both, LST and ET from MODIS images double those observed for the disaggregated in pixels with high CV. The relationship between the errors in surface energy fluxes (Rn and ET) and the errors in LST, including all the data, is shown in Fig. 6. The higher errors in the surface energy fluxes globally from MODIS are mainly consequence of the large heterogeneity of this area combined with its low spatial resolution. This increases the uncertainty in the retrieval of the soil and canopy components, required as inputs in the STSEB approach. As an example, a zoom to our study area is included in Fig. 7 showing daily ET for DOY 237 obtained from the disaggregated images at 10 and 60 m, the reference Landsat at 60 m and the MODIS LST at 1 km. Note that using MODIS LST as input leads to very homogenous values of ET through the entire area, whereas the ET resulting from disaggregated LST images represents better the heterogeneity of the area. This enhancement is especially remarkable in the pivots where higher ET values are observed. Note that ET values in the large pivots in the scene are very similar between the disaggregated and Landsat images. However focusing for example on the central upper pivot, planted with a variety of different crops, the corresponding different ET values observed in the disaggregated image can be barely discerned in the Landsat image. For a better understanding of the ET deviations, images of the ET differences, using Landsat as the reference, were computed. Fig. 7e shows the good performance of the disaggregated ET with dominant differences ranging between ±0.5 mm day−1 with some minor areas showing an underestimation around 1 mm day− 1, most likely consequence of border effects. On the contrary, large ET deviations are observed in the MODIS-Landsat difference image (Fig. 7f) mainly

Fig. 5. Example of LST images corresponding to DOY 237 from disaggregated at 10 m spatial resolution (a), disaggregated LST at 60 m spatial resolution (b), original Landsat (c) and MODIS (d). Absolute error images in LST, using Landsat as a reference, obtained from disaggregated LST (e), and MODIS LST (f).

M. Bisquert et al. / Remote Sensing of Environment 187 (2016) 423–433

431

Fig. 6. Density plots between the fluxes (Rn, left; ET, right) and the LST errors for the disaggregated images.

focused on the pivots, where ET values are underestimated as much as 2.5 mm day− 1. For the rest of the scene, ET uncertainty values are again within the ±0.5 mm day−1 range. 4. Summary and conclusions In this work we have tested and assessed the use of disaggregated LST images for estimating ET with high spatial and temporal resolution. We have applied disaggregation techniques to the combination of images from a low-spatial/high-temporal resolution (including TIR

information) sensor (MODIS) and a high-spatial resolution sensor provided with no TIR band (Spot-5). The STSEB approach was used for the retrieval of the surface energy fluxes. The result assessment was performed at two different levels: field scale and scene scale. At field scale, data extracted from 10 m spatial resolution images were compared to ground-measured data. At the scene scale, the disaggregated LST and fluxes images at 60 m spatial resolution were validated using respective Landsat 7 ETM + images. Good results were obtained in both cases. LST assessment showed R-RMSD of ± 1.9 °C and ± 1.4 °C at field and scene scales, respectively. Relative errors (MADP) of 22%

Fig. 7. Maps of daily ET (mm day−1) corresponding to DOY 237 obtained from: (a) disaggregated 10 m resolution image, (b) disaggregated image re-aggregated at 60 m resolution, (c) Landsat 60 m resolution image, and (d) MODIS LST image at 60 m resolution. Absolute error images in daily ET, using Landsat as a reference, obtained from disaggregated LST (e), and MODIS LST (f).

432

M. Bisquert et al. / Remote Sensing of Environment 187 (2016) 423–433

and 14% in instantaneous ET estimation were obtained at field and scene scales, respectively. Comparison between disaggregated daily ET and ground data led to a MADP of 19%. The good performance observed at the field scale assessment shows the potential of using disaggregated LSTs at 10 m spatial resolution as inputs in the STSEB model to calculate instantaneous and daily ET. Moreover, better results of surface energy fluxes were obtained when using disaggregated LSTs as a comparison to MODIS LST, with an average enhancement of 13% at image scale. This improvement is highlighted at a field scale, in which the effects of the surface heterogeneity become essential. Note the use of disaggregated LSTs for ET estimation is not limited to the STSEB, and could be extended to other models requiring TIR data. Also, the disaggregation methods could be applied to other pairs of sensors, for example combining Sentinel-2 and Sentinel-3. The use of disaggregated images for the estimation of ET is especially interesting for high heterogeneous areas. It will be very useful in regions worldwide where the agriculture parcels are prone to be very small in size and are then difficult to discern given the poor spatial resolution of the TIR sensors orbiting nowadays. However, the analysis at local scale showed that there is still a limitation when applying the disaggregation in recently irrigated fields, since disaggregated LSTs may result artificially higher than actual LSTs, as a direct consequence of the cooling produced by a recent water supply. These cases were discarded from the local scale analysis. The proposed use of disaggregated images for the estimation of ET could be further combined with fusion methods to obtain daily time series of surface energy fluxes at the high spatial resolution of sensors provided with no TIR bands, such as Sentinel-2. Acknowledgements This work was supported by the Spanish Ministry of Economy and Competitiveness (projects CGL2013-46862-C2-1/2-P and AGL201454201-C4-4-R, co-financed with European Union FEDER funds), and the Generalitat Valenciana (project PROMETEOII/2014/086). Spot data was provided by CNES and ESA under the Spot 5 Take 5 project. The authors would like to thank the logistic support in the instrumentation maintenance of Laura Martinez and Hector Picazo. References Agam, N., Kustas, W.P., Anderson, M.C., et al., 2007. A vegetation index based technique for spatial sharpening of thermal imagery. Remote Sens. Environ. 107:545–558. http://dx.doi.org/10.1016/j.rse.2006.10.006. Allen, R., Tasumi, M., Trezza, R., 2007. Satellite-based energy balance for mapping evapotranspiration with internalized calibration (METRIC)—model. J. Irrig. Drain. Eng. 133: 380–394. http://dx.doi.org/10.1061/(ASCE)0733-9437(2007)133:4(380). Amorós-López, J., Gómez-Chova, L., Alonso, L., et al., 2013. Multitemporal fusion of Landsat/TM and ENVISAT/MERIS for crop monitoring. Int. J. Appl. Earth Obs. Geoinf. 23:132–141. http://dx.doi.org/10.1016/j.jag.2012.12.004. Anderson, M.C., Norman, J.M., Diak, G.R., et al., 1997. A two-source time-integrated model for estimating surface fluxes using thermal infrared remote sensing. Remote Sens. Environ. 60:195–216. http://dx.doi.org/10.1016/S0034-4257(96)00215-5. Anderson, M.C., Norman, J.M., Mecikalski, J.R., et al., 2004. A multiscale remote sensing model for disaggregating regional fluxes to micrometeorological scales. J. Hydrometeor. 5: 343–363. http://dx.doi.org/10.1175/1525-7541(2004)005b0343:AMRSMFN2.0.CO;2. Anderson, M.C., Norman, J.M., Mecikalski, J.R., et al., 2007. A climatological study of evapotranspiration and moisture stress across the continental United States based on thermal remote sensing: 2. Surface moisture climatology. J. Geophys. Res. http://dx.doi. org/10.1029/2006JD007507. Barsi, J.A., Barker, J.L., Schott, J.R., 2003. An atmospheric correction parameter calculator for a single thermal band earth-sensing instrument. Geoscience and Remote Sensing Symposium, 2003. IGARSS'03. Proceedings. 2003 IEEE International. IEEE, pp. 3014–3016. Barsi JA, Schott JR, Palluconi FD, Hook SJ (2005) Validation of a Web-based Atmospheric Correction Tool for Single Thermal Band Instruments. In: Butler JJ (ed). (p 58820E– 1–58820E–7). Bastiaanssen, W.G.M., Menenti, M., Feddes, R.A., Holtslag, A.A.M., 1998. A remote sensing surface energy balance algorithm for land (SEBAL). 1. Formulation. J. Hydrol. 212– 213:198–212. http://dx.doi.org/10.1016/S0022-1694(98)00253-4. Berger, M., Rast, M., Wursteisen, P., et al., 2001. The DAISEX campaigns in support of a future land-surface-processes mission. ESA Bull. 105, 101–111. Bindhu, V.M., Narasimhan, B., Sudheer, K.P., 2013. Development and verification of a nonlinear disaggregation method (NL-DisTrad) to downscale MODIS land surface

temperature to the spatial scale of Landsat thermal data to estimate evapotranspiration. Remote Sens. Environ. 135:118–129. http://dx.doi.org/10.1016/j.rse.2013.03. 023. Bisquert, M., Sánchez, J.M., Caselles, V., 2016. Evaluation of disaggregation methods for downscaling Modis land surface temperature to Landsat spatial resolution in Barrax test site. IEEE J. Sel. Top. Appl. Earth Observations Remote Sens. 9, 1430–1438. Cammalleri, C., Anderson, M.C., Gao, F., et al., 2013. A data fusion approach for mapping daily evapotranspiration at field scale: DATA fusion approach for mapping daily ET. Water Resour. Res. 49:4672–4686. http://dx.doi.org/10.1002/wrcr.20349. Cammalleri, C., Anderson, M.C., Gao, F., et al., 2014. Mapping daily evapotranspiration at field scales over rainfed and irrigated agricultural areas using remote sensing data fusion. Agric. For. Meteorol. 186:1–11. http://dx.doi.org/10.1016/j.agrformet.2013.11. 001. Choudhury, B.J., Idso, S.B., Reginato, R.J., 1987. Analysis of an empirical model for soil heat flux under a growing wheat crop for estimating evaporation by an infrared-temperature based energy balance equation. Agric. For. Meteorol. 39:283–297. http://dx.doi. org/10.1016/0168-1923(87)90021-9. Coll, C., Caselles, V., Galve, J.M., et al., 2005. Ground measurements for the validation of land surface temperatures derived from AATSR and MODIS data. Remote Sens. Environ. 97:288–300. http://dx.doi.org/10.1016/j.rse.2005.05.007. Coll, C., Wan, Z., Galve, J.M., 2009. Temperature-based and radiance-based validations of the V5 MODIS land surface temperature product. J. Geophys. Res. http://dx.doi.org/ 10.1029/2009JD012038. Corbari, C., Bissolati, M., Mancini, M., 2015. Multi-scales and multi-satellites estimates of evapotranspiration with a residual energy balance model in the Muzza agricultural district in Northern Italy. J. Hydrol. 524:243–254. http://dx.doi.org/10.1016/j. jhydrol.2015.02.041. Doña, C., Chang, N.-B., Caselles, V., et al., 2015. Integrated satellite data fusion and mining for monitoring lake water quality status of the Albufera de Valencia in Spain. J. Environ. Manag. 151:416–426. http://dx.doi.org/10.1016/j.jenvman.2014.12.003. Dorigo, W.A., Zurita-Milla, R., de Wit, A.J.W., et al., 2007. A review on reflective remote sensing and data assimilation techniques for enhanced agroecosystem modeling. Int. J. Appl. Earth Obs. Geoinf. 9:165–193. http://dx.doi.org/10.1016/j.jag.2006.05.003. French, A.N., Hunsaker, D.J., Thorp, K.R., 2015. Remote sensing of evapotranspiration over cotton using the TSEB and METRIC energy balance models. Remote Sens. Environ. 158:281–294. http://dx.doi.org/10.1016/j.rse.2014.11.003. Gao, F., Masek, J., Schwaller, M., Hall, F., 2006. On the blending of the Landsat and MODIS surface reflectance: predicting daily Landsat surface reflectance. IEEE Trans. Geosci. Remote Sens. 44:2207–2218. http://dx.doi.org/10.1109/TGRS.2006.872081. Gao, F., Kustas, W., Anderson, M., 2012. A data mining approach for sharpening thermal satellite imagery over land. Remote Sens. 4:3287–3319. http://dx.doi.org/10.3390/ rs4113287. Hagolle, O., Huc, M., Pascual, D., Dedieu, G., 2015a. A multi-temporal and multi-spectral method to estimate aerosol optical thickness over land, for the atmospheric correction of FormoSat-2, LandSat, VENS and Sentinel-2 images. Remote Sens. 7: 2668–2691. http://dx.doi.org/10.3390/rs70302668. Hagolle, O., Sylvander, S., Huc, M., et al., 2015b. Spot-4 (Take 5): simulation of Sentinel-2 time series on 45 large sites. Remote Sens. 7:12242–12264. http://dx.doi.org/10. 3390/rs70912242. Inglada, J., Arias, M., Tardy, B., et al., 2015. Assessment of an operational system for crop type map production using high temporal and spatial resolution satellite optical imagery. Remote Sens. 7:12356–12379. http://dx.doi.org/10.3390/rs70912356. Katul, G.G., Oren, R., Manzoni, S., et al., 2012. Evapotranspiration: A process driving mass transport and energy exchange in the soil-plant-atmosphere-climate system: evapotranspiration and climate. Rev. Geophys. http://dx.doi.org/10.1029/2011RG000366. Kustas, W.P., Norman, J.M., Anderson, M.C., French, A.N., 2003. Estimating subpixel surface temperatures and energy fluxes from the vegetation index–radiometric temperature relationship. Remote Sens. Environ. 85:429–440. http://dx.doi.org/10.1016/S00344257(03)00036-1. Latorre, C., Camacho, F., de la Cruz, F., et al., 2014. Seasonal monitoring of FAPAR over the Barrax cropland site in Spain, in support of the validation of PROBA-V products at 333 m. Recent Advances in Quantitative Remote Sensing. Torrent (Spain), pp. 431–435. López-Urrea, R., Martín de Santa Olalla, F., Fabeiro, C., Moratalla, A., 2006. Testing evapotranspiration equations using lysimeter observations in a semiarid climate. Agric. Water Manag. 85:15–26. http://dx.doi.org/10.1016/j.agwat.2006.03.014. López-Urrea, R., Montoro, A., González-Piqueras, J., et al., 2009. Water use of spring wheat to raise water productivity. Agric. Water Manag. 96:1305–1310. http://dx.doi.org/10. 1016/j.agwat.2009.04.015. López-Urrea, R., Montoro, A., Mañas, F., et al., 2012. Evapotranspiration and crop coefficients from lysimeter measurements of mature “Tempranillo” wine grapes. Agric. Water Manag. 112:13–20. http://dx.doi.org/10.1016/j.agwat.2012.05.009. Moran, M.S., 2000. New imaging sensor technologies suitable for agricultural management. Asp. Appl. Biol. 60, 1–10. Moran, M.S., Inoue, Y., Barnes, E.M., 1997. Opportunities and limitations for image-based remote sensing in precision crop management. Remote Sens. Environ. 61:319–346. http://dx.doi.org/10.1016/S0034-4257(97)00045-X. Moreno, J.F., Alonso, L., Fernàndez, G., et al., 2004. The SPECTRA Barrax campaign (SPARC): overview and first results from CHRIS data. Proceedings of the Second Workshop CHRIS/Proba. European Space Agency, (Special Publication) ESA SP, Frascati (Italy), pp. 30–39. Mu, Q., Heinsch, F.A., Zhao, M., Running, S.W., 2007. Development of a global evapotranspiration algorithm based on MODIS and global meteorology data. Remote Sens. Environ. 111:519–536. http://dx.doi.org/10.1016/j.rse.2007.04.015. Mu, Q., Zhao, M., Running, S.W., 2011. Improvements to a MODIS global terrestrial evapotranspiration algorithm. Remote Sens. Environ. 115:1781–1800. http://dx.doi.org/10. 1016/j.rse.2011.02.019.

M. Bisquert et al. / Remote Sensing of Environment 187 (2016) 423–433 Niclos, R., Rivas, R., Garcia-Santos, V., et al., 2016. SMOS level-2 soil moisture product evaluation in rain-fed croplands of the Pampean Region of Argentina. IEEE Trans. Geosci. Remote Sens. 54:499–512. http://dx.doi.org/10.1109/TGRS.2015.2460332. Norman, J.M., Kustas, W., Humes, K., 1995. A two-source approach for estimating soil and vegetation energy fluxes from observations of directional radiometric surface temperature. Agric. For. Meteorol. 77, 263–293. Roy, D.P., Ju, J., Lewis, P., et al., 2008. Multi-temporal MODIS–Landsat data fusion for relative radiometric normalization, gap filling, and prediction of Landsat data. Remote Sens. Environ. 112:3112–3130. http://dx.doi.org/10.1016/j.rse.2008.03.009. Rubio, E., Caselles, V., Coll, C., et al., 2003. Thermal–infrared emissivities of natural surfaces: improvements on the experimental set-up and new measurements. Int. J. Remote Sens. 24:5379–5390. http://dx.doi.org/10.1080/0143116031000102412. Sánchez, J.M., Caselles, V., Niclòs, R., et al., 2007. Evaluation of the B-method for determining actual evapotranspiration in a boreal forest from MODIS data. Int. J. Remote Sens. 28, 1231–1250. Sánchez, J.M., Kustas, W.P., Caselles, V., Anderson, M.C., 2008a. Modelling surface energy fluxes over maize using a two-source patch model and radiometric soil and canopy temperature observations. Remote Sens. Environ. 112:1130–1143. http://dx.doi.org/ 10.1016/j.rse.2007.07.018. Sánchez, J.M., Scavone, G., Caselles, V., et al., 2008b. Monitoring daily evapotranspiration at a regional scale from Landsat-TM and ETM+ data: Application to the Basilicata region. J. Hydrol. 351:58–70. http://dx.doi.org/10.1016/j.jhydrol.2007.11.041. Sánchez, J.M., López-Urrea, R., Rubio, E., Caselles, V., 2011. Determining water use of sorghum from two-source energy balance and radiometric temperatures. Hydrol. Earth Syst. Sci. 15:3061–3070. http://dx.doi.org/10.5194/hess-15-3061-2011. Sánchez, J.M., López-Urrea, R., Rubio, E., et al., 2014. Assessing crop coefficients of sunflower and canola using two-source energy balance and thermal radiometry. Agric. Water Manag. 137:23–29. http://dx.doi.org/10.1016/j.agwat.2014.02.002. Schmidt, M., Lucas, R., Bunting, P., et al., 2015. Multi-resolution time series imagery for forest disturbance and regrowth monitoring in Queensland, Australia. Remote Sens. Environ. 158:156–168. http://dx.doi.org/10.1016/j.rse.2014.11.015.

433

Schneider, P., Ghent, D., Corlett, G., et al., 2012. Land Surface Temperature Validation Protocol. Seguin, B., Itier, B., 1983. Using midday surface temperature to estimate daily evaporation from satellite thermal IR data. Int. J. Remote Sens. 4:371–383. http://dx.doi.org/10. 1080/01431168308948554. Semmens, K.A., Anderson, M.C., Kustas, W.P., et al., 2015. Monitoring daily evapotranspiration over two California vineyards using Landsat 8 in a multi-sensor data fusion approach. Remote Sens. Environ. http://dx.doi.org/10.1016/j.rse.2015.10.025. Senf, C., Leitão, P.J., Pflugmacher, D., et al., 2015. Mapping land cover in complex Mediterranean landscapes using Landsat: Improved classification accuracies from integrating multi-seasonal and synthetic imagery. Remote Sens. Environ. 156:527–536. http://dx.doi.org/10.1016/j.rse.2014.10.018. Sobrino, J.A., Gómez, M., Jiménez-Muñoz, J.C., et al., 2005. A simple algorithm to estimate evapotranspiration from DAIS data: Application to the DAISEX campaigns. J. Hydrol. 315:117–125. http://dx.doi.org/10.1016/j.jhydrol.2005.03.027. Sobrino, J.A., Jiménez-Muñoz, J.C., Sòria, G., et al., 2008. Thermal remote sensing in the framework of the SEN2FLEX project: field measurements, airborne data and applications. Int. J. Remote Sens. 29:4961–4991. http://dx.doi.org/10.1080/01431160802036516. Valor, E., Caselles, V., 1996. Mapping land surface emissivity from NDVI: application to European, African, and South American areas. Remote Sens. Environ. 57:167–184. http://dx.doi.org/10.1016/0034-4257(96)00039-9. Vancutsem, C., Ceccato, P., Dinku, T., Connor, S.J., 2010. Evaluation of MODIS land surface temperature data to estimate air temperature in different ecosystems over Africa. Remote Sens. Environ. 114:449–465. http://dx.doi.org/10.1016/j.rse.2009.10.002. Wang, K., Dickinson, R.E., 2012. A review of global terrestrial evapotranspiration: Observation, modeling, climatology, and climatic variability: GLOBAL TERRESTRIAL EVAPOTRANSPIRATION. Rev. Geophys. http://dx.doi.org/10.1029/2011RG000373. Zarco-Tejada, P.J., Ustin, S.L., Whiting, M.L., 2005. Temporal and spatial relationships between within-field yield variability in cotton and high-spatial hyperspectral remote sensing imagery. Agron. J. 97:641. http://dx.doi.org/10.2134/agronj2003.0257.