While ensuring food security worldwide, irrigation is altering the water cycle and generating numerous environmental side effects, such as groundwater depletion and soil salinization. As detailed knowledge about location, timing and amounts of water used for irrigation over large areas is still lacking, remotely sensed soil moisture has proved to be a convenient means to fill this gap. However, the spatial resolution and/or the revisit time of satellite products represent a major limitation to accurately estimating irrigation.
In this work, we systematically and quantitatively assess the impact of the spatio-temporal resolution of soil moisture observations on the reliability of the retrieved irrigation information, i.e., timing and water amounts. Through a synthetic experiment based on soil moisture timeseries simulated by a hydrological model, we evaluate first the individual and then the combined impact of varying spatial and temporal resolution on both the detection and quantification accuracy. Furthermore, we investigate the effect of instrument noise typical of current satellite sensors, i.e., retrieval error, and irrigation rate, i.e., irrigation system and/or farmer’s decision on how much to irrigate.
Satisfactory results, both in terms of detection (F-score > 0.8) and quantification (Pearson’s R > 0.8), are found with soil moisture temporal samplings up to 3 days, or irrigated fractions as low as 30%, i.e., at least one-third of the pixel covers the irrigated field(s). Although lower spatial and temporal resolutions lead to a decrease in detection and quantification accuracy, the presence of random noise in the soil moisture timeseries has a more significant negative impact. As expected, better performances are found when higher volumes of irrigation water reach the soil. Finally, we show that current high-resolution satellite soil moisture products (e.g., from Sentinel-1) agree significantly better with model simulations forced with irrigation compared to rainfed simulations. On the other hand, coarse-scale products achieve higher correlations with soil moisture simulated without irrigation. Hence, our analysis highlights the potential for employing Sentinel-1 derived soil moisture for field-scale irrigation monitoring.

Sustainable water use in agriculture, while maintaining or increasing high yield levels, is becoming increasingly important to tackle challenges imposed by climate change and population growth. With increasing frequency and severity of drought events, competition for water resources is bound to intensify ever more. Knowing the water status of crops in the field allows to optimize water consumption by adapting water management practices to the actual water demand in the field, and make best use of limited water resources.

Crop canopy temperature (Tc) measured by thermal infrared (TIR) is an excellent indicator of crop water stress due to its close relation to relative transpiration rate and correspondingly transpiration cooling. Satellites equipped with TIR sensors can provide a cost-efficient global solution for irrigation management based on crop water stress monitoring. However, canopy temperature must be recorded with high spatial and temporal frequency, ideally daily, to accurately track crop water supply within a specific field. Current space borne TIR data products are available at high spatial or high temporal frequencies, but not both. Hydrosat is building a 16+ satellite constellation to provide high-resolution global TIR data products every day, multiple times per day. Hydrosat’s data will be a game changer in agricultural monitoring and management, making detailed sub-field-level irrigation management practical without any groundwork required for sensor installations or maintenance.

It is crucial for large-scale remote irrigation management that the stress indices used to quantify water stress produce accurate results independent of local weather conditions and without easy access to reliable ground data. Various stress metrics based on canopy temperature, e.g., canopy air temperature difference (CATD), crop water stress index (CWSI), and temperature vegetation dryness index (TVDI) have been proposed to account for varying environmental conditions and were shown to be successful in quantifying crop water stress under suitable experimental conditions. Diurnal variation in plant transpiration and mixing of vegetation and soil signals are further challenges for satellite data with discrete acquisition time and ground resolution much coarser than individual plants.

Field trials were carried out near Nelspruit in South Africa, where different crops (including maize, soybeans, potatoes, and dry beans) were studied on predominantly sandy soil in a humid but hot climate. Crop water stress indices obtained from ground TIR radiometers, handheld TIR camera images, and frequent unmanned aerial vehicle (UAV) flights were compared to extensive ground measurements including volumetric soil moisture, soil matric potential, and leaf water potential at pre-dawn and noon.

Different irrigation treatments resulted in significant yield differences, with higher and more homogeneous yield values obtained on well-irrigated plots. The spatial and temporal pattern of crop water stress under reduced irrigation was clearly resolved by thermal stress indices based on canopy temperature, with TVDI exhibiting highest sensitivity to water stress. However, while TVDI is sensitive to moderate and severe levels of water stress, well-watered crops show no significant difference from crops experiencing mild water stress. Using the Penman-Monteith formalism to calculate reference evapotranspiration (ET0), we estimated crop water demand following the FAO double crop coefficient approach using TVDI as scaling factor for crop transpiration and soil evaporation. A soil water balance calculation based on a single-layer water bucket model was able to reproduce the experimental water content curve very well (R2 > 0.9), with largest discrepancies occurring after heavy rainfall due to an underestimation of runoff. Results from a currently ongoing field trial using these insights for real-time irrigation management will be presented as well.

Consistent with previous studies, TVDI was most sensitive to water stress during sunny hours around noon. Under these conditions corresponding to high incident net radiation, the dry and hot soil surface layer significantly affects the effective TIR emission of mixed soil/vegetation pixels in spatially resolved TIR data. In principle, TVDI was defined to account for variation in fractional vegetation cover (FVC), but a simple trapezoidal model was insufficient in this experiment to separate soil and vegetation contributions during early growth stages with incomplete canopy closure. Furthermore, we found that the FVC-Tc parameter space, and therefore TVDI, is strongly affected by ground sampling distance and spatial misalignment between the spectral bands used to derive FVC and Tc. Consequently, care must be taken when upscaling crop water stress indices validated on field-scale to global satellite observation.

Applying the same crop water stress and soil water balance models to Landsat-8 scenes acquired over selected areas in the central United States of America, we were able to adequately quantify actual evapotranspiration and soil water balance using only normalized difference vegetation index (NDVI) and land surface temperature (LST), as well as weather data for soil water balance calculations.

Among the human activities altering the natural circulation of water on the Earth’s surface, irrigation is the most impactful one. In the near future, water exploitation aimed at improving food production through irrigation practices is expected to further increase to face population growth and rising living standards under a climate change scenario. Nevertheless, a detailed knowledge of irrigation dynamics (i.e., extents, timing, and amounts) is generally lacking worldwide. This open problem is the main driver of the European Space Agency’s (ESA) Irrigation+ project, whose main goals are: (i) the development of methods and algorithms to detect, map, and quantify irrigation at different spatial scales, (ii) the production of satellite-derived irrigation products, and (iii) the assessment of the impacts of irrigation on society and science.
This study presents a comparison between two different methodologies, developed within the Irrigation+ project, aimed at estimating irrigation water amounts over different test sites: a satellite-based approach, only relying on satellite data, and a data assimilation approach, developed within the NASA Land Information System (LIS) framework, which integrates land surface modelling and remote sensing (hereafter model-based approach). The satellite-based method, namely the Soil-Moisture-based (henceforth SM-based) inversion approach, relies on the backwards estimation of irrigation rates by exploiting satellite soil moisture and evapotranspiration data as an input. The method has been implemented with two Sentinel-1-derived soil moisture products: RT1 Sentinel-1, whose spatial resolution is 1 km, and S2MP, produced at a plot scale by merging Sentinel-1 Synthetic Aperture Radar (SAR) data with Sentinel-2 observations. On the other hand, the model-based approach investigates possible benefits, in terms of irrigation quantification, provided by the assimilation of 1-km spatial resolution Sentinel-1 backscatter into the Noah-MP land surface model coupled with an irrigation scheme. The assimilation updates soil moisture and vegetation via a calibrated backscatter forward operator which is represented by a Water Cloud Model.
Results shed a light over the irrigation information contained in the high-resolution Sentinel-1 soil moisture products. In addition, the comparison of different methodologies is useful to understand the limits and potential of each, thus highlighting the existing gaps to be filled in order to obtain reliable irrigation estimates.

Irrigation in olive groves reduces water stress and allows trees to absorb more nutrients, which increases olive productivity. To improve yield through water volume rationing, farmers often apply a Regular Deficit Irrigation (RDI) throughout the growing season. RDI is used to apply a percentage (typically between 70% and 90%) of a crops’ evapotranspiration during all or a part of the irrigation season. A tool to accurately estimate evapotranspiration in olive groves can drive RDI strategies, optimize effective irrigation management, and improve yield in relation to the applied water volume. Benchmark systems for monitoring evapotranspiration routinely, such as the eddy covariance approach, are not feasible due to their expense and limited spatial extent. Obtaining spatial information from satellites is an ideal solution to overcome in-situ sensor limitations, as they can provide imagery at both high spatial and temporal resolution. Optical remote sensing has been widely used to estimate the evapotranspiration via the use of vegetation indices. However, optical sensors are not usable in the presence of clouds, and the spatial resolution of non-commercial optical sensors is generally insufficient for resolving within field variations, individual trees or even hedgerows in orchards. Such limitations can be overcome through using high-resolution Synthetic Aperture Radar (SAR) data. Here, we evaluate the potential of SAR data to provide information suitable for irrigation management in olive groves. The analysis focused on an irrigated olive plantation located in Saudi Arabia. For each olive plot, Sentinel-1 SAR backscatter (C-band) at a given acquisition date was computed as the average of the pixel values located in that plot. For this analysis, the SAR backscatter was compared to simultaneous evapotranspiration measurements. Results demonstrate a correlation between SAR backscatter and evapotranspiration with a coefficient of determination of 0.80. Overall, the study demonstrates that SAR data can track variation in locally measured evapotranspiration and illustrate their potential to be used for irrigation management.

Sensitivity Analysis of the C and L bands SAR Data for Detecting Irrigation Events

Hassan Bazzi 1, Nicolas Baghdadi 1, Mehrez Zribi 2 and François Charron 3

1 INRAE, UMR TETIS, University of Montpellier, AgroParisTech, 500 rue François Breton, CEDEX 5, 34093 Montpellier, France; nicolas.baghdadi@teledetection.fr (N.B.)
2 CESBIO (CNRS/UPS/IRD/CNES/INRAE), 18 av. Edouard Belin, bpi 2801, CEDEX 9, 31401 Toulouse, France; mehrez.zribi@ird.fr (M.Z.)
3 G-EAU Unit, University of Montpellier, AgroParisTech, CIRAD, INRAE, Institut Agro, IRD, Domaine du Merle, 13300 Salon de Provence, France; francois.charron@supagro.fr

For better management of water resources and water consumption, the detection of the irrigation timing at the agricultural plot scale is of great importance. With the availability of the Sentinel-1 (S1) SAR data in free and open access mode at 6 days revisit time, several studies have demonstrated the potential of the S1 data for detecting irrigation events (Bazzi et al., 2020a; Le Page et al., 2020). Current irrigation detection models built using S1 SAR data are based on the monitoring of the increase of the SAR backscattering signal between two consecutive SAR acquisitions (increase in soil moisture). However, the detection of irrigation events based on the soil moisture-SAR correlation using the C-band S1 data (wavelength ~ 6cm) is sometimes limited to the penetration capability of the C-band over some developed vegetation cover. For certain high vegetation cover, especially for some cereal crops such as wheat and barley, several studies have demonstrated that the sensitivity of the C-band SAR signal to soil moisture values becomes negligible when the vegetation is well-developed (El Hajj et al., 2018; Joseph et al., 2010; Nasrallah et al., 2019). To overcome the penetration limitation of the C-band, SAR data with a higher wavelength could be required such as the L-band (wavelength ~24 cm) SAR data.

The objective of this study is to compare the sensitivity of the S1 C-band and the ALOS-2 L-band images for irrigation detection over well-developed vegetation cover. A sensitivity analysis of both C and L bands for irrigation detection was performed over 45 references irrigated grassland plots located in the Crau plain of southeast France during two growth cycles. The first growth cycle is rich in grasses (coarse hay) and resembles wheat crops and the second grass cycle is richer in legumes with less percentage of coarse hay. The Normalized Difference Vegetation Index (NDVI) is used in the analysis to describe the vegetation cover.
In order to understand the capability of detecting irrigation events using the C-band SAR data, we studied first the response of the S1 backscattering signal (σ_C^0) following irrigation events by examining the temporal evolution of the σ_C^0 according to rainfall and irrigation events.
Two L-band images each acquired in each growth cycle were used to compare between the C and L bands sensitivity for irrigation detection. We analyze the relationship between the σ_C^0 and L-band backscattering coefficients (σ_L^0) as a function of the time difference between the SAR acquisition date and the irrigation date (∆t). The parameter ∆t is considered to be a proxy measure of the soil dryness-wetness.

The results showed that when the vegetation cover develops in the first growth cycle (coarse hay), the response of the σ_C^0 to the water supplements (either irrigation or rainfall) becomes negligible and the irrigation events could be hardly detected. The σ_C^0 in the first growth cycle shows no correlation with the ∆t value which means that the σ_C^0 are not sensitive to the dryness-wetness of the soil. The behavior of the σ_L^0 as a function of the time difference between the image acquisition date and the irrigation date in the first growth cycle indicates that in the presence of either low or high developed vegetation cover, the σ_L^0 in HH polarization could be still sensitive to the soil water content. Regardless of the NDVI values, wet soil conditions due to irrigation on the same day of the L-band acquisition induce high σ_L^0 (around -11 dB). In contrast, dry soil due to the absence of irrigation 5 to 6 days before the ALOS-2 acquisition shows low σ_L^0 values (around -15 dB).

In the second growth cycle of grass (rich in legumes), both C and L bands are sensitive to the soil moisture values in the presence of either high or low vegetation cover. In L-band, the σ_L^0 values decrease from -13 dB when the irrigation is on the same day of the ALOS-2 acquisition to less than -17 dB when the irrigation is 15 days before the ALOS-2 acquisition. Similar behavior with less tendency was observed for the C-band. Results showed that L-band is more sensitive than the C-band and the HH polarization in L-band is more sensitive than the HV polarization for detecting irrigation events.

References

Bazzi, H., Baghdadi, N., Fayad, I., Zribi, M., Belhouchette, H., and Demarez, V. (2020a). Near Real-Time Irrigation Detection at Plot Scale Using Sentinel-1 Data. Remote Sensing 12, 1456.
Bazzi, H., Baghdadi, N., Fayad, I., Charron, F., Zribi, M., and Belhouchette, H. (2020b). Irrigation Events Detection over Intensively Irrigated Grassland Plots Using Sentinel-1 Data. Remote Sensing 12, 4058.
El Hajj, M., Baghdadi, N., Bazzi, H., and Zribi, M. (2018). Penetration Analysis of SAR Signals in the C and L Bands for Wheat, Maize, and Grasslands. Remote Sensing 11, 31.
Joseph, A.T., van der Velde, R., O’Neill, P.E., Lang, R., and Gish, T. (2010). Effects of corn on C- and L-band radar backscatter: A correction method for soil moisture retrieval. Remote Sensing of Environment 114, 2417–2430.
Le Page, M., Jarlan, L., El Hajj, M.M., Zribi, M., Baghdadi, N., and Boone, A. (2020). Potential for the Detection of Irrigation Events on Maize Plots Using Sentinel-1 Soil Moisture Products. Remote Sensing 12, 1621.
Nasrallah, A., Baghdadi, N., El Hajj, M., Darwish, T., Belhouchette, H., Faour, G., Darwich, S., and Mhawej, M. (2019). Sentinel-1 Data for Winter Wheat Phenology Monitoring and Mapping. Remote Sensing 11, 2228.

Satellite sensors have been promoted widely as a technology to optimize water and crop productivity in agriculture. Different remote sensing technologies are able to detect crop stress and water shortages, with a special emphasis on water stress. Improving water and crop productivity goes hand in hand. Moreover, crop stress resulting from waterlogging leads to suboptimal crop productivity, however, has so far received little attention in literature and, consequently, technological development. This is surprising because approximately twenty percent of the global agricultural land suffers from the consequences of waterlogging and secondary soil salinization. While irrigation is expected to increase productivity, excess water can hamper the crop growth and decrease water use efficiency.

In this study we focus on an irrigated sugarcane plantation in southern Mozambique burdened by waterlogging. We show how Sentinel-1 backscatter and Planet NDVI can be used to monitor sugarcane development. Our results demonstrate Sentinel-1 backscatter is influenced by sucrose accumulation and can be used to predict sucrose yield early in the season. In addition, we demonstrate how poor sucrose development is linked to waterlogging and when additional irrigation application is counterproductive. To test the usefulness of these findings the next step will be to integrate the methodology into the decision making framework of the plantation and continue validation of the work done so far.