Specific Project Objectives & Deliverables


1         Crop water budget monitoring

Remote sensing has long been used for computing evapotranspiration estimates, which is an input for crop water balance monitoring. Up to now, only medium and low resolution data (e.g. MODIS) are available on regular basis to monitor cultivated areas. However, the increasing availability of high resolution high repetitivity VIS-NIR remote sensing, like the Sentinel-2, offers unprecedented opportunity to improve this monitoring.

Methods for computing evapotranspiration (ET) using remote sensing belong basically to two broad families, either using thermal remote sensing used to solve the energy budget of the surface, or using SVAT modeling forced by remotely sensed information of vegetation properties (e.g. fraction cover, leaf area index, crop coefficients…). The first group i.e. Surface Energy Balance (SEB) methods use RS data to estimate heat exchange between land surface and atmosphere compute the sensible heat flux first and then obtain the latent heat flux as the residual of the energy balance equation at the time of satellite overpass.The latter group includes the coupling of the dual crop coefficient method described in FAO paper 56 (Allen, 1998) with NDVI time series providing spatialized estimates of the fraction cover (fc) and the basal crop coefficient (Kcb). We developed in previous works the SAMIR tool (Satellite of Monitoring Irrigation) implementing this method using high resolution image times series (SPOT, Landsat, FORMOSAT, Sentinel-2) and this year for validation, sensible heat flux measurements were obtained using a large aperture scintillometer (XLAS) .

1.1       Spatialized estimates of evapotranspiration.
1.1.1       SAMIR model / validation

The objective of the work was to assess the operationality of SAMIR and the accuracy of the modelled evapotranspiration (ET) at the scale of irrigated perimeters, in a context of high land cover complexity (i.e. trees, winter cereals, summer vegetables) and limited data available for parameterization.

Using the spatialized computation of the crop water budget presented in last year’s JECAM report (published in Remote sensing, Saadi et al. 2015), we achieved a validation of the ET using XLAS scintillometer measurements. The model was calibrated on the basis of local ET measurements from flux towers (eddy-correlation devices) installed on irrigated wheat and barley plots. For other crops for which no calibration data was available, parameters were taken from bibliography.

For validation, half hourly sensible heat flux measurements were obtained using a large aperture scintillometer (XLAS) over the study area along a path length of 4 Km. The daily sensible heat flux(H) were used to compute daily latent heat flux (LE) using the energy budget conservation (Rn + G = H + LE). The daily net radiation (Rn) was computed using MODIS daily data at the time of satellite overpass (i.e. providing half hourly Rn estimates), and scaled at daily scale using ground meteorological station. The soil flux (G) is supposed to be null at daily scale. For the Rn scaling, we used the ration of radiation measure at the station and daily value, for both global radiation (Rg) or Rn computed using the FAO method. Both methods gave similar results. The comparison between modelled daily and measured ET are shown in figure 4

1.1.2        SPARSE model

Moreover, spatially distributed estimates of  ET were computed using the layer approach of the Soil Plant Atmosphere and Remote Sensing Evapotraspiration (SPARSE) model (Boulet et al., 2015) fed by low resolution remote sensing data (Terra and Aqua MODIS). The objective of the work was to assess the SPARSE model operationality and the accuracy of the modeled i) instantaneous  H and ii) daily evapotranspiration as well as its components (soil evaporation  and transpiration) over a semi-arid land surface, in a context of high land.

The SPARSE’s layer approach was run to compute instantaneous estimates of sensible heat flux H at the time of satellite overpass. The comparison between H estimates and large aperture scintillometer (LAS)’s H measurements over the study area along a pathlength of 4 Km showed that the SPARSE model presents satisfactory accuracy (figure 3).


Figure 3: Modeled vs. Observed sensible heat fluxes at Terra and Aqua time overpass (Saadi et al., 2017, in preparation)

In a subsequent step, daily modeled latent heat flux LE (i.e. ET)  computed on the basis of modeled half hourly LE were compared to observed LE returned as a residual term of the surface energy budget using the daily scintillometer’s H measurements and net radiation (computed using instantaneous MODIS data and then extrapolated to daily time step). Results are shown in figure 4.


Fiure 4: Daily Modeled Vs. Observed latent heat fluxes (Saadi et al., 2017, in preparation)


1.2       Long term irrigation monitoring / WEAP

We propose a generic toolbox based on the FAO-56 method and the Crop Coefficient/NDVI approach used in Remote Sensing. A toolbox has been used in order to build a WEAP21 model of the Merguellil basin in Tunisia for the period of 2000-2014 (Le Page et al, 2016). The toolbox can be separated in three main areas: 1) preparation of different input datasets, 2) A collection of algorithms based on the analysis of NDVI time series (MODIS) is proposed: Separation of irrigated vs non-irrigated area, a simplified annual land cover classification, Crop Coefficient, Fraction Cover and Efficient Rainfall, 3) Synthesis against points or areas produces the output data at the desired spatial and temporal resolution for Integrated Water Modeling or data analysis and comparison. Finally, the comparison to monthly statistics of three irrigated commands was performed over 4 years. Punctual evapotranspiration was compared to actual measurements obtained by flux towers on wheat and barley showing good agreements on a daily basis (r2=0.77). This late comparison showed a bad agreement which led us to suppose two things: First, the simple approach of (Evapotranspiration minus Efficient Rainfall) to estimate Irrigation at the monthly time step is not pertinent because only a subset of the irrigated commands is actually irrigated. Hence, a higher spatial resolution of remote sensing imagery is needed. Second, in this particular area, farmers have a different rationale about rainfall and irrigation water needs. Those two aspects need to be further investigated.

The toolbox has proven to be an interesting tool to integrate different sources of data, efficiently process them and easily produce input data for the WEAP1 model on a long term range. Yet some new challenges have been raised.


2         Soil water content monitoring

Soil water monitoring using microwave data has been studied at various scales, from 1km to 2m. We obtain new results on the JECAM site.

We propose the MHYSAN model (Modèle de bilan HYdrique des Sols Agricoles Nus / Bare Soil HYdrological Balance Model, Gorrab et al, 2017) for simulating soil water balance of bare soils. This model was used to simulate surface evaporation fluxes and SM content at daily time scale over a semi-arid, bare agricultural site in Tunisia (North Africa). Two main approaches are considered in this study. Firstly, the MHYSAN model was successfully calibrated for seven sites using continuous thetaprobe measurements at two depths. Then the possibility to extrapolate local SM simulations on distant sites based only on soil texture similarity was tested. This extrapolation was validated using SAR estimates and manual thetaprobe measurements of SM made on these distant sites. Results show bias about 0.63 and 3.04 % and RMSE equal to 6.11 and 4.5 % for the SAR volumetric SM and manual thetaprobe measurements, respectively. In a second approach, the MHYSAN model was calibrated using seven very high-resolution SAR (TerraSAR-X) SM outputs ranging over only two months. The simulated SM were validated using continuous thetaprobe measurements during 15 months (Fig. 5).

Although only seven dates of SM were used for calibration, the satisfying results we obtained could be attributed to the large SM variation captured by these seven images, allowing a good calibration of the soil parameters These results highlight the potential of Sentinel-1 images for daily soil moisture monitoring using simple models.

Figure 5: Estimation of time series of water balance variables, using calibrated MHYSAN SAR data and validation results (Gorrab et al., 2017).



3          Yield estimates

No field data in 2016.

©2015 Joint Experiment for Crop Assessment and Monitoring