Achieving food security while using water resources in a sustainable manner is a major challenge. Global strategies to increase food production should focus efforts on increasing production per unit resources, i.e. the combined increase of production per unit land surface (yield expressed in kg/ha) and the increase of production per unit water consumed (water productivity expressed in kg/m³). Closing land and water productivity gaps is a complex task which requires: (i) monitoring of current levels of productivity in various crop production systems; (ii) assessment of observed productivity relative to potential; (iii) identification and analysis of the underlying causes of the productivity gaps; and (iv) evaluation of options and identification of viable solutions to close the productivity gaps in the local context. To support these processes, FAO and partners maintain a publicly accessible near real time database that applies analysis of high-resolution satellite images in conjunction with specific algorithms to determine spatial and temporal variability of agricultural water and land productivity. Through the project activities, a validated remote sensing based methodological framework is being created to assess and monitor land and, more specifically, water productivity. The provision of near real time information through an open access data portal (WaPOR https://wapor.apps.fao.org) enables a range of service-providers to assist farmers attain more reliable yields and to improve their livelihoods. Irrigation operators have access to new information to assess the performance of systems and to identify where to focus investments to modernize the irrigation schemes, and government agencies will be able to use the information to monitor and promote the efficient use of natural resources.
However, translating this information into practical management advices at field level is constrained by the inherent complexity of agricultural landscapes and their monitoring through remote sensing. Worldwide, 72 percent of all farms are less than one hectare in size, and the number of small farms has grown over the past few decades, with average farm sizes decreasing since 1960 in most low- and middle-income countries, where the majority of the world’s farms are located . Small farm size, combined with cloud cover, call for remote sensing data at higher spatial and temporal resolution and further integration of Sentinel data is clearly the next step in WaPOR development.

Most of Northern Europe was affected by drought during the summer of 2018. In Denmark, the period from May to July was among the driest in the country in 99 years, with significant lower agricultural yields and reduced income by farmers as a consequence. This accentuates the need for irrigation and improved tools for irrigation management.
Conventional crop water requirement models are driven by standard meteorological observations and information about soil and crop type. Typically, model calculations use precipitation, temperature and global radiation for estimation of a reference evapotranspiration based on semi-empirical formulas and subsequently this is combined with a crop specific model. This calculation is not accurate in all weather conditions and typically, standard figures are used for the individual soil types. At the same time, there is uncertainty on how much water deficit may be permitted in various situations without yield and/or quality loss. Thus, there is a huge potential for improving this assessment by the inclusion of satellite observations directly related to crop status. In particular, early indication of crop water stress is a critical variable in optimal irrigation management.
Thermal-based remote sensing methods to estimate crop stress and ET rely on the increase of leaf temperature when stomata close under slight stress, and hence it provides a more rapid and earlier detection or quantification of stress and thus a larger potential for irrigation management. There have been numerous thermal-based energy balance models developed for estimating surface fluxes and ET as well as for detecting plant stress. However, few have the capability to compute robust fluxes over a variety of surface conditions.
In this study, as part of the MOIST project, funded by the Danish Innovation Fund, we have applied a simple approach to monitor crop water stress during the year 2018 for Denmark. We have applied TVDI – Temperature Vegetation Dryness Index estimated from Land Surface Temperature (LST) and Vegetation Index (NDVI) available from the Sentinel 3 satellite sensors to monitor crop water stress. The study illustrates the usefulness of thermal infrared remote sensing for agricultural applications in early detection of water stress on a national scale. At the same time, the study points to the need for and large potential of application of higher resolution land surface temperature data when moving to field scale applications.
The crop water stress information will be implemented and tested by the Danish Farmers association as part of their dissemination systems to farmers in Denmark in 2019, for further consolidation for the approach.

Field-scale estimates of actual evapotranspiration (ET) are of crucial importance for agricultural water accounting, improving the efficiency of irrigation practices and general integrated water resources management. Accurate modelling of ET with satellite observations is possible when both thermal infrared (TIR) and optical (visible, near- and shortwave-infrared) data are used. The current constellation of Sentinel satellites is lacking a thermal sensor with fine enough resolution to enable evapotranspiration modelling at the appropriate scale, despite offering valuable observations for field-scale monitoring of other agricultural practices. While optical observations from Sentinel-2 (S2) satellites are acquired at a resolution of 10 m to 20 m, TIR observations from Sentinel-3 (S3) satellites have a native resolution of around 1 km. As part of the Sentinels for Evapotranspiration project we aim to fill this data gap by using advanced machine-learning methods and S2 images to sharpen land surface temperature (LST) derived from S3 TIR observations. The resulting LST has a pixel spacing of 20 m and is used in land surface energy balance models to estimate actual ET (and other fluxes) at this fine resolution.

We have evaluated two sharpening methods and three land surface energy balance models at 11 flux tower sites located in Europe, Africa and North America. Both sharpening methods match a S3 scene with a S2 scene acquired on the same day or at most four days previously and build a specific self-calibrating model for each match. They also both employ the same techniques for training pixel selection, bias minimalization and prevention of over-fitting. They differ in the implementation of the model with one method being based on decision trees and the other on neural networks. The three land surface energy balance models represent three different modelling approaches. The first one is a Two-Source Energy Balance (TSEB) model that splits the LST into the soil and canopy components and estimates the fluxes from those two parts separately. The second model is METRIC and it is a one-source model that scales fluxes between hot and cold end-members. The third approach is an End-Member-Base Two-Source model (ESVEP) which is similar to TSEB but contains pixel-level scaling of the fluxes between wet and dry canopy and soil end-members. The 11 validation sites include 6 towers located in agricultural areas, 2 in savanna, 1 in grassland, 1 in conifer forest and 1 in broadleaved forest. Seven of the sites are in a temperate climatic zone and four in semi-arid zone. The validation was performed using all available images from year 2017.

Preliminary results indicate that the sharpened LST can be used to estimate accurate ET at field scale. When comparing the modelled fluxes against tower measurements, the fine-scale model outputs often produce more accurate results then the coarse-scale outputs. This demonstrates that by using sharpened LST we can minimize the scale mismatch between the flux tower footprint (on the order of tens of meters) and the modelled fluxes and that during the sharpening process the information content of the LST image is enhanced. Once the validation is complete, we will implement the best performing combination of sharpening method and land surface energy balance model as an open-source tool integrated in the SNAP software.

The main objective of MOSES (Managing crOp water Saving with Enterprise Services) H2020 project was to put in place and demonstrate at the real scale of application an information platform to provide information services to reclamation consortia, irrigation districts and other water management agencies.
The Beta version of the MOSES platform was delivered during the irrigation season 2017 and included two (remote sensing based) modules for the monitoring and predicting of crop and irrigation water requirements. The Crop Water Demand module delivers daily monitoring and forecast of crop water demand as well as the monitoring of bio-physical variables. Irrigation water requirements are estimated at the same temporal resolution taking into account precipitation. Forecast of crop and irrigation water requirements is done by combining daily weather forecasts for a seven day moving window, with the crop coefficient Kc on the first day or by extrapolating the Kc time series. The In Season Crop Mapping module performs frequent (bi-weekly) mapping of specific crops during the irrigation season. As documented by user requirements, the detailed in-season monitoring of crop water requirements would be useful in order to regularly update, fine tune and adjust allocation plans and management of the water resources to end users, both water managers and farmers. The in season crop mapping module delivers information useful to verify the cropped area determined on the basis of farmers’ pre-season intentions as stated to local water management authorities. Another MOSES module determines net irrigation water requirements by modeling the soil water balance, which applies crop – specific parameters and thus requires crop mapping.
The crop water demand module combined multispectral image data acquired at unprecedented high spatial (10-20 m) and temporal resolution (3-5 days) by the Sentinel 2 / MSI and Landsat 8 / OLI imaging radiometers to capture the inter- and intra-field variability of crop conditions over the growing cycle, while coping with cloudiness. The module has been conceived to be completely automated providing daily maps of current Normalized Difference Vegetation Index (NDVI), Leaf Area Index (LAI), Crop water demand (CWD) and Gross Irrigation Water Requirements (GIWR), as well as 7 days forecast of CWD and GIWR. Crop Water demand is calculated by means of two methods: a) semi – empirical estimate of Kc combined with meteorological data to detemineET0 and b) the energy balance based Penman – Monteith equation.
After a preliminary alpha version tested in Italy in summer 2016, during years 2017 and 2018, the preoperational (beta) version of the DSS system was activated in the four MOSES demonstration areas (DAs) located in Italy, Romania, Spain and Morocco. The results for 2017 and 2018 irrigation seasons were evaluated as regards accuracy and compliance with user requirements.
The presentation will focus on the assessment of the adequacy of the water applied to meet the crop water demand in the irrigation seasons 2017-2018 for all four MOSES demonstration areas (DA-s).
Actual and detailed measurements of water applied were available for the Spain DA. In this case the MOSES CWD data were very close to or slightly higher than the amount of water applied by farmers. The MOSES GIWR values were generally lower than water applied, thus suggesting a potential reduction in applied water by adjusting the irrigation water management to reduce the irrigation by 10-20%. This opportunity could be further captured by exploiting the CWD and GIWR forecasts.
In the Italy DA actual measurements of applied irrigation water were available for few farms. In the a Kiwi farm all the estimated crop water requirements were much lower than the water applied, i.e. the potential for water management adjustments towards water saving is high. In Peach farms the CWD values were about twice the water applied. At these locations farmers apply deficit irrigation during specific phenological phases of the crop.
Two stakeholders in Spain provided qualitative evidence on adoption of the MOSES services. The use – cases documented the interest raised in the managers of the two demonstration areas when they could see in a map the performance indicator “relative irrigation supply”. This indicator is computed by dividing the measured water supply by the EO-estimated irrigation requirements. The identification of farmers who consumed extra amount of water is a new means to improve on-farm water management through warnings from the Water User Associations or, in extreme cases, by intervention of the water jury.

Data from the European Sentinel sensors, part of the Copernicus program, carry a large potential for application for agriculture sector and water resource management. Now, for the first time, operational data from optical, thermal IR and radar sensors are freely available to users. To fully exploit the potential and to ensure high quality information to the end users is provided, a multi-disciplinary approach is needed.
In MOIST a joint effort by agronomists, remote sensing scientists and IT architects will provide new, scientifically based satellite products to farmers for intelligent irrigation management. MOIST addresses all steps in the value chain, from remote sensing research leading to new, improved satellite products, development of more precise irrigation strategies to the testing and dissemination of the products. A new approach to exploit data from different sensors data will ensure consist between products, timeliness and high-quality output, and at the same time, overcome some of the limitations of current systems.
MOIST exploits and combines optical, thermal infrared and SAR data from Sentinel-1, Sentinel-2 and Sentinel-3, each with distinct information content, which enables us to map and update crop water needs in due time before the crop has been damaged by water stress. From satellite data, spatially distributed information about variation within an agricultural field with frequent updates is provided giving timely information to the farmer. The project focuses on algorithms for deriving satellite products crucial for irrigation management: vegetation, water stress, evapotranspiration and soil moisture, based on the fusion of optical, thermal infrared and radar data, and field experiments over selected crops in different climate zones in Denmark, Spain and Italy.
The first year of data collection of the project, 2018 was particularly interesting due to the unusually low precipitation amounts in Northern Europe, and the late onset of growing season in Southern Europe. The presentation will highlight the first results from the 2018 field campaign and compare with the first satellite products developed for the project with focus on water stress, soil moisture and evapotranspiration.

Integrative environmental management by decision makers aims at making the best possible economic and societal use of the land, while at the same time maintaining its biotic, structural and chemical quality in a sustainable way. Dedicated authorities, such as the ministries of environment, are in charge of implementing and enforcing environmental legislation and thus act based on environmental data and on their perception of regional needs. Often these data, classically gathered from censuses, surveys or terrestrial sensor networks, inadequately represent spatial heterogeneities, are rapidly outdated and are likely to fail at observing impacts of short-term events, such as of droughts or floods. Nevertheless, future integrative environmental management strongly relies on spatially and temporally highly resolved, up-to-date and homogeneous information on most diverse natural processes in large areas to achieve the goal of sustainable development of the natural resources.

With the ESA COPERNICUS program, a fleet of Earth observation instruments is available, which largely contributes a new source of environmental data, both in terms of quality and quantity, to fill these gaps in administrative environmental data needs. The Sentinels cover key impacts on the environment, such as hydrological and agricultural activity with unprecedented spatial and temporal resolution, continuous quality and reliability. Within the research project VieWBay (Virtueller Wasserraum Bayern = Bavarian Virtual Water Space) we combine Sentinel data with sophisticated environmental models to create a virtual environment. Formerly unobservable, coupled environmental processes like soil water flows and temperatures, vegetation dynamics etc. are simulated in large spatial and temporal detail using information derived from the continuous Sentinel data flows and are visualized to prepare decision alternatives for the Bavarian State Ministry of the Environment and Consumer Protection and its subordinate administrative units.

VieWBay integrates the spatial information contained in the regular SENTINEL-2(a/b) observations by assimilating the 5-day observation series into a hydrological/agroecological modelling framework centered around the PROMET land surface model (Mauser and Bach, 2009). As a result, a full and detailed picture of the water balance, plant growth and carbon assimilation, runoff, water shortage and water temperature dynamics is created for all of Bavaria at a spatial resolution of 10-20 m. SuperMUC-2, the high performance computing facility of the Leibnitz Supercomputing Center of the Bavarian Academy of Science, is used for processing the Sentinel data stream and assimilating it into the land surface processes simulations. All available Sentinel data is preprocessed from TOA reflectance to level-3 biophysical variables in an automated processing chain developed by VISTA (Munich, Germany) based on the SLC canopy reflectance model (Verhoef and Bach, 2007) for the entire surface area of Bavaria (approx. 70 550 km²). The biophysical variables derived from the Sentinel data (crop specific leaf area, soil color, chlorophyll content, leaf angle distribution) are assimilated into the land surface model and thereby nudge PROMET towards reproducing land surface process which are spatially and temporally consistent with remote sensing observations (Hank et al., 2015). Due to the high environmental impact of agricultural management, agricultural areas are specially emphasized and are observed at the native geometric resolution of the instruments, i.e. 10m, whereas all other land use categories are simulated at resolutions of 20-300 m. To account for the high diurnal dynamics of the land surface, the model time-step is one hour.

By co-designing and co-developing the Bavarian Virtual Water Space in close collaboration with the stakeholders, i.e. the Ministry and the administrative authorities, and by using up-to-date technologies for data visualization, such as virtual and augmented reality, we expect the integrated modelling system to serve as a tool to both prepare environmental decisions and to communicate them with civil society. The Bavarian Virtual Water Space is designed to provide the Bavarian environmental authorities with easy access to spatially and temporally seamless data on key environmental variables. At the same time, the Bavarian Virtual Water Space provides a playground for assessing the impacts of different management scenarios on the Bavarian environment, e.g. the impact of increased irrigation in agriculture or the efficiency and effectiveness of establishing polders for flood protection and thus will help implementing sustainability in environmental decisions through improved spatial information based on space technology.

VieWBay (http://viewbay.geographie muenchen.de/) will be fully established until 2021 and will gradually take over its operational work until then. No final simulation results for the full area will therefore be presented at the Living Planet Symposium 2019. Nonetheless, first results from two preselected sub-catchments (Wern: approx. 600 km², Windach: approx. 130 km²), which both are part the Danube main catchment, will be presented. The presentation will allow for tracing the principles of data preprocessing and data assimilation and will provide previews of the results, which at a later stage of the project will fully cover all of Bavaria. We think that applications preparing the operational use of COPERNICUS data for environmental administration, such as the VieWBay initiative, are of large interest to the audience at the LPS 2019.

HANK, T., BACH, H. & MAUSER, W. 2015. Using a Remote Sensing-Supported Hydro-Agroecological Model for Field-Scale Simulation of Heterogeneous Crop Growth and Yield: Application for Wheat in Central Europe. Remote Sensing, 7, 3934-3965.

MAUSER, W. & BACH, H. 2009. PROMET - Large scale distributed hydrological modelling to study the impact of climate change on the water flows of mountain watersheds. Journal of Hydrology, 376, 362-377.

VERHOEF, W. & BACH, H. 2007. Coupled soil–leaf-canopy and atmosphere radiative transfer modeling to simulate hyperspectral multi-angular surface reflectance and TOA radiance data. Remote Sensing of Environment, 109, 166-182.