The stress index can also be computed based on a combination of thermal and reflectance data. Combining remotely sensed visible and thermal infrared data can provide more information for estimating SM than the single one. It is important to determine how to combine these methods reasonably to obtain highly accurate SM. Among such indexes, the Surface Energy Balance Index (SEBI, Menenti and Choudhury, 1993), Water Deficit Index (WDI, (Moran, 2004; Moran et al., 1994) are different expressions of the stress Factor, and have been derived from a surface energy balance model, based on the same theory as the CWSI. The WDI can be used in different surface conditions, like covered area and sparsely covered one. Relying on the surface energy balance principle, the soil adjusted vegetation index (SAVI, Huete, 1988) and the temperature difference form a trapezoidal space, and the index can be directly calculated from remotely sensed data without any on leaf and air temperature measurements.
Some index like Temperature Vegetation Index (TVI, Prihodko and Goward, 1997), Temperature Vegetation Dryness Index (TVDI, Sandholt et al., 2002) and Vegetation Temperature Condition Index (VTCI, Wan et al., 2004; Wang et al., 2004), do not rely on any parameterization of the energy balance and can thus be computed directly from remote-sensing data. The above stress index can be estimated depending on the triangle/trapezoid method. These methods are usually based on the trapezoidal or triangle shape formed between VTCI is defined as a ratio of the dry to actual LST difference to the dry to wet LST temperature difference, with wet/dry LST being estimated as the minimum/maximum LST that the surface can reach for a given meteorological forcing. VTCI = LSTNDVI.max−LST (I.7).
Where LSTNDVI.max and LSTNDVI.min are the maximum and the minimum LST of pixels which have same NDVI value in the studied area, respectively. The both extremes temperature are estimated from the dry and the wet edge in the LST-NDVI space. The most important in the image-based method is that, the number of pixels should be sufficient to cover all conditions. In the LST-VI (vegetation index) space the conservation of energy is a key elements, the surface energy maintains balance. In the LST-VI space the wet edges represents an adequate SM and higher ET and the dry edge represents that the vegetation is subjected to water stress in which evapotranspiration reaches the minimum and the SM is minimal.
The above thermal- and shortwave-derived SM index has a wide coverage and a fine spatial resolutions due to the existence of space-borne satellites. The thermal-based index uses the responses of the soil energy balance to soil moisture to determine SM. By contrast, the shortwave-based indexes use characteristic changes in the soil reflectance or vegetation physiology to estimate SM. One of the drawback of these methods is the temporal resolution. The thermal/optical remote sensing SM indices can be inferred only on clear sky days. In addition, using Landsat’s TIR data we will have the data at best every 16 days.
Alternatively, a microwave-based SM index could be used to estimate SM. Microwave data are not affected by clouds, which allows SM to be monitored at high frequency. However, microwave-based SM indexes are easily perturbed by vegetation and surface roughness. Considerable efforts have been made for the characterization of the spatial and temporal variability of SM over vegetated areas. Passive and active microwave-based indices are developed for monitoring SM from space. Among the active microwave-based indices, SM can be retrieved using multi-date SAR imagery. These indexes are based on change detection techniques for multi-temporal SAR data. Shoshany et al. (2000) presented the normalized radar backscatter soil moisture index (NBMI), which is obtained from the backscatter measurements at two different times (t1 and t2) over the same area.
Nowadays, there is a limitation in the existing satellite thermal sensors, since those with high revisit cycles (e.g, MODIS) do not offer high spatial resolution (HR), and those offering HR (e.g, Landsat-8) generally have low temporal resolution (Agam et al., 2007b). In contrast, the visible and near infrared (VNIR) reflectance data are available at a resolution finer than that of most thermal sensors (Ha et al., 2013). To bridge the gap, the finer-resolution VNIR data have hence been extensively used as ancillary data to disaggregate low-resolution (LR) LST at HR.
Recently, various efforts have been devoted to disaggregate LST to a finer –typically 100 m-resolution. Techniques are generally based on a relationship between LST and ancillary (vegetation cover indexes, emissivity and/or albedo) data, the obtained relationship being assumed to be scale invariant (and thus applied at both HR and LR). The statistical downscaling methods in particular, developed by Kustas et al. (2003) over a homogenous vegetated area, has been widely used. This method is based on a linear regression relationship between LST and NDVI (Normalized Difference Vegetation Index) calibrated at LR. The relation between LST and NDVI is also used in Bindhu et al. (2013), with the aim of developing a nonlinear method to estimate LST at HR. Agam et al. (2007a) used the fraction of green vegetation cover instead of NDVI. This method showed its capability and good performance over areas with relatively uniform soil and vegetation hydric status, where the temperature of bare soil is set to the average between the dry and wet soil over the studied area. Other studies reported that NDVI (or fgv) shows some limitations and cannot explain all the variations in LST over agricultural areas (Agam et al., 2007b, 2007a; Inamdar and French, 2009; Merlin et al., 2010b; Olivera-Guerra et al., 2017). Especially, Agam et al. (2007b) and Merlin et al. (2010) observed a shortcoming when using the LST-NDVI or LST-fgv relationship over areas with high moisture content, or with various photosynthetic activity vegetation types. Merlin et al. (2010) adapted this method to heterogeneous vegetation status, by adding the fraction of senescent vegetation cover to include the photosynthesis activity of vegetation, and to distinguish between areas of bare soil and dry vegetation cover. Dominguez et al. (2011) integrated the surface albedo to estimate HR LST by fitting the relationship between LST, NDVI and surface albedo. Following the same idea of adding other information that affect the spatial distribution of LST, Merlin et al. (2012) used the projection technique theoretically developed in Merlin et al. (2005) that aims to strengthen the correlation between two variables (LST and NDVI) by representing the dependence of these variables on other additional variables, based on a radiative transfer equation. Moreover, other studies were further presented involving additional factors that reflect the vegetation type (Merlin et al., 2010b; Sandholt et al., 2009; Zhan et al., 2011). Sandholt et al. (2002) summarized the variables that affect LST variability, and they mentioned that, NSSM mainly controls ET and the energy balance components of the surface, which affect LST. Therefore, optimal LST disaggregation approaches should include the variability of SM in addition to NDVI (or fgv), in order to represent the variability of the bare soil temperature bounded by its wet and dry endmembers. Advanced regression tools using spectral bands, have been successfully used in different studies to produce better disaggregation results than simple polynomial functions (Ghosh and Joshi, 2014) Recently, some studies have attempted to represent the SM effect. Liu and Zhu (2012) used a NMDI for monitoring soil and vegetation moisture, based on the absorption properties of the vegetation water in the NIR and the sensitive characteristics of water absorption differences between soil and vegetation in the SWIR. However, NMDI has inconsistent relationships with vegetation and SM changes (i.e. positive correlation with vegetation water content and negative correlation with SM changes). Therefore, it poorly performed over mixed pixels of vegetation and soil. Chen et al. (2010) took into account SM variations using a soil wetness index (SWI) estimated based on the interpretation of the triangular LST–NDVI space. However, the errors were found to be larger with low fractional vegetation cover. In the same manner, Yang et al. (2010) discussed the impact of SM variations using the LST-NDVI space and assumed uniform SM conditions in a coarse pixel. Therefore this technique is only appropriate in regions where SM varies at large scale and in pixels with high fgv. In general, the previously proposed proxies or indexes that aim to incorporate the SM effect on LST poorly performed over the areas with low vegetation cover.
Factors conditioning ET
ET depends on two parameters: i) energy provided by solar radiation and the amount water available in the soil. Indeed, one of the main factors conditioning ET is the energy available at the surface (provided by net radiation (Rn)). Aerodynamic resistance (ra) to sensible (H) and latent heat (LE) transfer, and water stress (related to the amount of water in the soil) are also important factors that affect ET, because when water is scarce in the soil, the stomata close again and transpiration slows down.
According to Schultz and Engman (2000), remote sensing observations combined with auxiliary meteorological data were used to estimate ET over a range of spatio-temporal scales. Much progress has been made recently in the measurement of remote sensing parameters, including: i) Solar radiation; ii) Surface albedo; iii) Vegetation cover; iv) LST; v) NSSM.
Recently satellite data have been used to estimate real ET at different scales (eg, Bastiaanssen et al., 1998; Granger, 1997). The above important parameters to estimate ET are obtained by measuring, by remote sensing of an electromagnetic radiation of a given wavelength, emitted or reflected from the surface. Incident solar radiation, albedo, and LST can be estimated using the same satellite measurements and SM can be estimated by measuring the microwaves emitted or received by the soil (emission and reflection, or backscattering from the ground). However, there are uncertainties in these estimates due to other factors such as surface roughness and vegetation cover.
Remote sensing-based modelling approaches
As an alternative to observational methods of ET, numerous modelling methods have been proposed such as Simple Soil Plant Atmosphere (SiSPAT) (Braud et al., 1995), Interaction Soil-Biosphere-Atmosphere (ISBA) (Noilhan and Mahfouf, 1996) and simple SVAT (Soil Vegetation Atmosphere Transfer) (Boulet et al., 2000), Interactive Canopy Radiation Exchange (ICARE) (Gentine et al., 2007). Other models like Crop Environment REsource Synthesis (CERES) (Ritchie, 1986), Simulateur multidisciplinaire pour les Cultures Standard (STICS) (Brisson et al., 1998) and the crop-water productivity model (Aquacrop) (Raes et al., 2009) have combined the water balance with the crop growth, development and yield components. These modelling methods, whether complex or simple, are generally not easy to implement in an operational context as they require several parameters (e.g. soil and vegetation hydrodynamic properties) and forcing variables (e.g. climate and irrigation) that are often unavailable at the desired space and time scale. As a matter of fact, simpler models based on a few input data have been developed (Merlin, 2013; Merlin et al., 2014). Among them, the surface energy balance model (SEBS) estimates the turbulent fluxes and surface evaporative fraction (Su, 2002) by using remote sensing data (albedo, NDVI, emissivity and LST) in conjunction with meteorological forcing (solar radiation, air temperature, wind speed, air humidity) and surface parameters (e.g. roughness and stability correction functions for momentum and sensible heat transfer). Remote sensing energy balance using satellite imagery have been developed to estimate evaporation and ET from large areas (Allen et al., 2007; Bastiaanssen et al., 1998a, 1998c; Irmak et al., 2011; Kustas et al., 2003; Kustas and Norman, 1999). Those approaches are promising for application over a wide range of vegetation types and water availability and over large areas. Remotely sensed energy balance techniques are useful for identifying areas experiencing water stress and corresponding reductions in ET and to populate hydrologic models (Irmak and Kamble, 2009; Kamble and Irmak, 2009). There are some satellite-based energy balance models such as SEBAL (Bastiaanssen et al., 2005, 1998b, 1998c) and METRIC (Allen et al., 2007). Theses model are using inverse modeling at extreme conditions for the calibration. ET is estimated at the two conditions based on knowledge of available energy and surface conditions, usually with ties to ground-based weather data. Other more regional-scale models such as the ALEXI model (Anderson et al., 2005) use inversion based on radiosonde profilings of temperature and specific humidity over time to estimate large-scale heat flux and evaporation.
Table of contents :
Chapter I Bibliographic synthesis
I.2 Soil moisture
I.2.1 In situ measurements
I.2.2 Remotely sensed approaches
I.2.2.1 Soil moisture indices
I.220.127.116.11 Shortwave-based index
I.18.104.22.168 Thermal-based Index
I.22.214.171.124 Thermal/shortwave-based index
I.126.96.36.199 Microwave-based index
I.2.2.2 Soil moisture retrieval
I.2.2.3 Soil moisture missions
I.3 Land surface temperature
I.3.1 Remote sensing approaches
I.3.2 Spatio-temporal representativeness
I.4.1 Direct measurements of ET
I.4.2 Factors conditioning ET
I.4.3 Remote sensing-based modelling approaches
I.4.4 Surface evaporative efficiency
Chapter II Data & study sites description
II.2 Sites and in situ data description
II.2.1 Study areas
II.2.2 Meteorological data
II.2.3 In situ soil moisture data
II.2.4 In situ LST data
II.2.5 Flux data
II.3 Preprocessing satellite data
II.3.1 Satellite data characteristics
II.3.2 Data preprocessing
II.3.2.1 Thermal infrared (TIR) data
II.3.2.2 Radar imagery
II.188.8.131.52 Thermal noise removal
II.184.108.40.206 Radiometric calibration
II.220.127.116.11 Terrain correction
II.18.104.22.168 Filtering speckle effects
II.3.2.3 High resolution reflectances
Chapter III Models & methods
III.2 Soil moisture indices (SMP)
III.3 Endmembers temperatures estimation
III.3.1 Modelling extreme temperatures: physically based energy balance model
III.3.2 Image based extreme temperature: contextual method
III.4 Integrating the SM indices to improve the water need estimates
III.4.1 Enhance Penman-Monteith method to estimate ET: thermal-based SMP
III.4.2 Calibration of the radar data to retrieve SM: radar/thermal based SMP
III.4.2.1 Benchmark approach: based only on radar data
III.4.2.2 New approach: combined radar/thermal data
III.4.3 Improve the spatio-temporal resolution of MODIS LST data: radar-based SMP
III.4.3.1 MLR technique
III.4.3.2 RTM technique
III.22.214.171.124 Model description
III.126.96.36.199 LST endmembers
III.188.8.131.52 Backscatter endmembers
III.5 Models evaluation
Chapter IV Results and discussions
IV.2 Consistency between image- and EBsoil-based extreme soil temperatures
IV.3 Wheat evapotranspiration using thermal/optical-based approach
IV.3.1 Relationship between surface resistance and stress index
IV.3.2 Evapotranspiration estimation at parcel scale
IV.3.3 Evapotranspiration mapping at perimeter scale
IV.3.3.1 Wheat stress index mapping at 100 m resolution
IV.3.3.2 Wheat evapotranspiration mapping at 100 m resolution
IV.3.3.3 Validation over flood and drip irrigation parcels
IV.4 Improving the LST spatio-temporal resolution
IV.4.1 Application to aggregated Landsat-7/8 data: R3 and Sidi Rahal sites
IV.4.2 Application to MODIS data: R3 area
IV.5 Surface soil moisture at parcel scale
IV.5.1 Sensitivity of VV- and VH-polarized data to soil moisture
IV.5.2 Relationship between thermal-derived SMPTs and radar signal
IV.5.3 SM estimation at high spatio-temporal resolution
IV.5.3.1 SM retrieval
IV.5.3.2 Sensitivity to temperature endmembers
IV.5.3.3 SM validation: Improvement of soil evaporation estimation
IV.6 Summary and conclusion
Conclusions and perspectives