Conference Agenda

Estimating evapotranspiration at a regional level is crucial for managing water irrigation practices in many parts of the world experiencing droughts. It is essential that irrigation is effective and optimized towards sustainable productivity. Nevertheless, maximizing yield has to follow sustainable management practices and preserve healthy soils. The study aims to present an approach for water productivity mapping in winter wheat at the municipality level based on satellite data combining evapotranspiration and yield modelling. The study area is the Parvomay Municipality, Bulgaria, located in the Upper Thracian Lowland. The agricultural landscape in this area is diverse, dominated by crops such as winter wheat, maize, and sunflower. However, rice and vegetables are also grown in the region. The production of these crops is heavily reliant on the irrigation infrastructure, which unfortunately is poorly maintained. The area is also characterized by medium to high baseline water stress, which adversely affects farming. We generated daily evapotranspiration rasters with the SenET SNAP plugin for all cloud-free images during the agricultural season 2021. The Sen-ET plugin uses data sources from Copernicus, including optical data from the Sentinel-2 MSI sensor, thermal data from the Sentinel-3 SLSTR sensor, and meteorological data from the ERA-5 dataset. This plugin enhances the spatial resolution of thermal images from Sentinel 3 SLTR (from 1 km to 20 m) by regression models that incorporate Sentinel-2 data. We then interpolated the data to obtain gap-free daily evapotranspiration time series. The overall evapotranspiration for the growing season 2021 was than calculated. A winter wheat yield model was developed based on in-situ data from the study area and Sentinel-2 vegetation indices. The model was used to predict and map yield at the municipality level for 2021. Finally, a water productivity map was generated presenting the evapotranspiration per winter wheat yield unit (t/ha).

The green area index (GAI), i.e., half of the green leaf and stem area per unit of horizontal ground surface area, is a key variable for assessing the development, health, and productivity of crops. Currently, most large-scale and cost-effective methods for its estimation exploit optical remote sensing data. Frequent cloud cover can, however, hinder their reliability by blocking the view of the sensors on which they rest. As a result in many parts of the world, its timely monitoring cannot be ensured using optical systems alone. Synthetic aperture radars (SARs), however, thanks to their cloud-penetrating ability, are capable of producing dense time series that can be used to improve the spatial and temporal coverage of their optical counterparts.

In this study, SAR-to-optical GAI regression has been performed using a transformer encoder with past and current values of SAR backscatter and interferometric coherence, as well as past values of LAI when available. Sentinel-1 and -2 images acquired from 2018 to 2021 over the Hesbaye region of Belgium have been used for cross-validation. The model has been trained on three growing seasons and tested on the fourth for each fold.

The results show that the model can successfully predict Sentinel-2-derived GAI with a cross-validation average R2=0.88 and RMSE=0.74, outperforming methods relying on radiative transfer model (e.g., the Water Cloud model) inversion. The model is also particularly effective compared to non-recurrent regression models, such as Random Forest and Multi-layer Perceptron, over long temporal gaps in the GAI time series, i.e., 30 to 60 days (15 to 30% of the growing season), a common occurrence in Belgium and many other parts of the world.

These promising results pave the way for the generation of accurate, dense GAI time series throughout the growing season, allowing for timely crop monitoring in cloud-prone regions.

Real-time and accurate monitoring of soil moisture content is the basis of agricultural water management. It is of great significance to explore the optimal model of soil water inversion in wheat fields for improving agricultural water use efficiency and sustainable development in wheat fields. [Method] This study takes the soil moisture content of the winter wheat planting area in Xunxian County, Hebi City, Henan Province as the research object, uses the UAV multi-spectral and thermal infrared data, combined with Sentinel-1A SAR data and field measured data, and uses three methods of temperature vegetation drought index model, water cloud model and improved water cloud model respectively to carry out the inversion comparison and validation analysis of soil moisture content.[Result] The results show that the inversion accuracy of the three methods at 10 cm depth is higher than 20 cm, and R2 is greater than 0.4.[Conclusion] Among them, the improved water cloud model method is better than the water cloud model and the temperature vegetation drought index in the inversion of soil moisture in wheat field, which is suitable and has high inversion accuracy. The R² is 0.6795 at the depth of 10 cm, the RMSE is 0.0141, the R² is 0.5742 at the depth of 20 cm, and the RMSE is 0.0133.

Abstract: Rice is one of the three major food crops. More than half of the world's population depends on rice as their staple food. China is the world's largest rice producer and consumer, so China's rice production is a strong support for ensuring world food security. Agriculture is the foundation of national economic development, and increasing rice yield per unit area has always been a key research topics in the world. Timely and accurate acquisition of rice yield information can not only assist policy regulation and reasonable planning of agriculture production at the large scale, but also be used for guiding agricultural production activities. Heilongjiang Province is an important rice production base in China and represents the highest rice production at national level to a certain extent. Therefore, accurate estimation of rice yield per unit area in Heilongjiang Province will help understand the status of rice production in China. Traditional rice yield estimation methods mainly include the statistical sampling survey and the crop growth model estimation, but the former is labor and resources costing, and the latter is limited in practical application because many relevant parameters are difficult to obtain. At present, the statistical yield estimation method based on meteorological data is mainly used in the meteorological departments and relatively mature. In addition, remote sensing technology has the advantages of non-destructive, high efficiency and high accuracy. As the access to satellite remote sensing data has become easier in recent years, rice yield estimation based on satellite remote sensing data has been well promoted and applied. This study took the rice yield per unit area in the Jiansanjiang Farming Area of Heilongjiang Province as the study topic. First, five sample points were selected and field surveys in the study area were conducted, and finally the yield data of the sample points were obtained. The FY-3D MERSI satellite data was used to calculate the normalized vegetation index for all sample points, obtain the time series of vegetation index for all sample points, and extract the curve characteristic parameters, and then the correlation between each characteristic parameter and the yield was analyzed. Secondly, the meteorological yield was separated from the rice yield from 2000 to 2019, and then the multiple stepwise linear regression was performed with various meteorological factors during the rice growth period from May to October in Jiansanjiang farming area to build a rice yield estimation model. Finally, the accuracy, correctness, and correlation coefficient between the estimated yield and the actual yield of the two methods were calculated, and a fitting test was performed to compare the characteristics and advantages and disadvantages of the two methods. The research results show that in this farming the rice yield estimation model based on remote sensing data is better than the statistical model based on meteorological factors, and has good potential for estimating rice yield. The study may provide scientific basis for the estimation of rice yield and model optimization in Jiansanjiang farming area.

Keywords: rice yield; remote sensing; meteorological factors; model

Maize is one of the most widely grown crops in the world, and plays a major role in farming operations. Monitoring the water and dry matter content of maize crops reaching their final stage of development is essential not only to ensure optimum crop yields, but also to improve harvest logistics by facilitating the sharing of mechanical resources between multiple small landowners. SAR remote sensing, thanks to its sensitivity to vegetation water content, can provide a cost-effective approach to its large-scale monitoring.

The present study aims to evaluate the feasability of a plant water content retrieval method in senescent maize crops from multi-frequency (C- and X-band) SAR backscatter data. To accomplish the stated objective, a 5-week field campaign was conducted in the Shanxi province, China, at the end of the summer of 2023. Wet biomass, plant water content, sowing density, BBCH, and plant height were measured in five maize parcels. Fields were visited four to five times, around one-week apart, resulting in 22 observations.

A linear regression analysis was used to assess the sensitivity of SAR backscatter to the plant water content (in cobs, plants, and both). It pointed to the fact that TerraSAR-X and Sentinel-1 might have too little capacity for monitoring maize at the end of the growing season, in the conditions of the campaign. The highest correlations were observed for C-band data in cross-polarization VH (R=-0.55 for cobs and R=-0.53 for plants). A random forest regressor was implemented to capture the non-linear relationship between SAR and in situ data, but poor leave-one-out cross-validation results (R2=0.24) demonstrated the ineffectiveness of the method for plant water content retrieval.

Several factors may have contributed to this outcome. Primarily, the small size of the parcels in the region posed a significant challenge due to the speckle effect limiting the representativity of the radar measurements. Furthermore, the reduced penetration abilty of C- and X-band SAR in high-biomass crops, such as maize, might have led to signal saturation during the entire observation period, in spite of decreasing plant water content.

Accurate estimation of evapotranspiration (ET), which plays a crucial role in the land surface energy balance and terrestrial water cycle, has long been a challenging issue in hydrological research. An emerging research topic is to combine process-based model and data-oriented approach for earth surface system modeling. In this study, a hybrid model was developed by integrating the machine learning (ML)-based canopy resistance estimation into the ETMonitor model to simulate actual ET of the soil-vegetation system. Three ML algorithms, random forest, gradient boosting regression tree and deep neural network were used for canopy resistance estimation, so that the performance of multiple versions can be assessed. In-situ observations of 179 eddy covariance flux tower sites across the globe were used for model training, and the daily ET estimates were validated at 45 sites independently. The hybrid model with deep neural network showed the best accuracy, with KGE increased from 0.7 to 0.84, the R² increased from 0.66 to 0.72, and the RMSE decreased from 0.85 (mm d^-1) to 0.65 (mm d^-1) when comparing with the process-based model with default parameters. The better behavior of the deep neural network than other shallow ML algorithms highlights the benefits of deep learning algorithms for intricate land surface process modeling and model parameter retrieval. Compared with pure ML models, the hybrid models tend to produce less outliers and achieve more concentrated data distribution at different validation sites, indicating the model stability when applied in different situations. The interpretability analysis shows that leaf area index, vapor pressure deficit and soil moisture are the main regulators of ET estimation in the hybrid model, which provides deeper insight into the accuracy of ET prediction in the absence of fully measured data. The findings of this study demonstrate the enhanced capability and potential of the hybrid model to achieve accurate terrestrial ET estimation.

Quantifying net irrigation is crucial for optimizing agricultural water management. However, traditional survey-based approaches often fail to capture the spatial and temporal dynamics of net irrigation. The widely-adopted methods, which rely on hydrological models to retrieve irrigation, have high uncertainties due to oversimplified model structures and data scarcity. This study proposes a new approach for estimating the net irrigation with a high spatial resolution of 1 km at annual scale, using the Budyko hypothesis and multiple satellite observations. Green water ET(GET), representing ET from precipitation stored in the unsaturated soil layer for terrestrial ecosystems, was estimated using the Budyko equation. Net irrigation was derived by dividing blue water ET (BET) (i.e. ET from irrigation water), obtained from satellite-retrieved actual ET minus GET, by the irrigation efficiency. The net irrigation estimates showed a percentage error of 6.67% and a root mean square error of 80.79mm when compared to field observations. The superior of the proposed method was also demonstrated by comparing with the accuracy of other existing methods. The average annual net irrigation volume in the Heihe River Basin from 2001 to 2018 was 2.58ⅹ10¹⁰m³. Although the net irrigated area per unit area decreased, the net irrigation volume showed a significant increasing trend over the years. In 2018, the net irrigation volume was 36.39% higher than in 2001. This trend is confirmed by the statistical bulletin. In the southeastern part of the middle reaches, the croplands exhibited significantly lower net irrigation compared to the central riverine region, as well as the western and downstream areas. The proposed approach has the potential to be transferred to other regions and assist decision makers in promoting sustainable water management.

Spatio-temporally continuous daily soil moisture data plays a pivotal role in the study of ecological and hydrological processes. However, there are significant gaps in the spatial and temporal coverage of current soil moisture products acquired by spaceborne microwave radiometers. Such gaps result from an array of factors, including the vacancy in satellite orbit coverage, radio frequency interference, and errors from retrieval algorithms. In this study, we applied a fully automatic Discrete Cosine Transformation-Partial Least Square (DCT-PLS) method for filling missing data in the soil moisture product (SMOS-MTMA), which is derived by using the Multi-Temporal and Multi-Angular (MTMA) method based on the Soil Moisture Ocean Salinity (SMOS) observations. The reliability of the DCT-PLS method is first performed through a filling experiment that simulated the loss of values in both in-situ soil moisture time series and satellite-observed soil moisture maps. The experiments prove the method’s high performance, with correlation coefficients (R) between reconstructed and the original (true) soil moisture surpassing 0.9. Moreover, the root mean squared error (RMSE) and mean absolute error (MAE) for both experiments remain below 0.04 m³/m³. After applying the DCT-PLS method to the SMOS-MTMA product, it is found that the reconstructed soil moisture exhibits continuity between the reconstructed data regions and their spatially adjacent original data regions, and also exhibits a strong spatio-temporally consistency with original data products. The reconstructed soil moisture product was finally validated using in-situ measurements from 22 soil moisture networks. The reconstructed soil moisture product could reveal the annual periodic variations of soil moisture and was in good agreement with the in-situ measurements (overall R > 0.7 and overall ubRMSE = 0.042 m³/m³), and achieved similar performance with the original SMOS-MTMA soil moisture product. It is expected the reconstructed SMOS soil moisture product over a decade (since 2010) may provide a stronger support for research in Earth science.

Remote sensing of Surface Soil Moisture (SSM) has historically displayed larger uncertainties and lower spatial resolutions than the land surface temperature (LST) or vegetation indices counterpart. In this work, an assessment of the reliability of existing satellite products and an application test for a heterogeneous agricultural area in Italy will be presented.

In the first part, the focus is on the great variability among satellite products of SSM, in terms of spatial, temporal, and radiometric resolutions and retrieval techniques. Data of precipitation and irrigation over an irrigation consortium in southern Italy are contrasted with different SSM products: Soil Moisture Ocean Salinity (SMOS), Soil Moisture Active Passive (SMAP), European Space Agency Climate Change Initiative (ESA-CCI) products, Copernicus SSM1km, and Advanced Microwave Scanning Radiometer 2 (AMSR2). The hydrological consistency of these products is analysed, as a measure of the coherency between SSM and precipitation/irrigation. Any SSM retrieval is classified as either a positive or negative consistency, depending on the presence/absence of water inputs recorded since the previous satellite overpass. Generally, no SSM dataset stands out with a positive performance, as negative consistencies are recorded for roughly 50% of the retrievals.

In the second part, the observed, high-resolution (1km) Copernicus SSM dataset is employed in a data assimilation scheme over the whole consortium aimed at retrieving irrigation volumes. A distributed hydrological model (FEST-EWB) is run, uninformed of any irrigation activity, simulating seamlessly in time the energy and water balances without the need for LST in input. Whenever an SSM retrieval is available, an update of the corresponding model state is triggered, and the related irrigation volumes are recorded. The final results, as opposed to official irrigation volumes from the consortium, are good, although influenced by the spatial gap between satellite observations (1 km²) and average size of the local plots (on average, 7 ha, 14 times lower).

Remote sensing-based estimates of Evapotranspiration (ET) play a crucial role in detecting agricultural water stress. Operational ET products derived from the Spinning Enhanced Visible and Infrared Imager (SEVIRI) sensor aboard the Meteosat Second Generation (MSG) satellites offer high temporal resolution and moderate spatial coverage, making them well-suited for water stress monitoring. Nonetheless, there remains a need for rigorous evaluation of SEVIRI observations and the development of streamlined workflows, ideally deployable on cloud-based platforms, to effectively utilize this data for large-scale water stress monitoring. In this study, we first conducted a comprehensive assessment of European actual and reference ET (SEVIRI-ET_a and SEVIRI-ET₀) observations against ground-based measurements collected at 54 Eddy Covariance (EC) sites along seven key dimensions, i.e., diurnal cycle, daily, intra-annual, inter-annual, ecosystem, climate zone, and products intercomparison from 2004 to 2018. The evaluated SEVIRI-ET (both ET_a and ET₀) products were then employed for two primary objectives: i) facilitating a water stress detection workflow based on monthly evaporative stress index (ESI) anomalies, implemented on the cloud-based Virtual Earth Laboratory (VLab) platform, to monitor spatio-temporal variations in water stress across Europe over a decade (2011 to 2020), and ii) examining the overall response of terrestrial ecosystems to water stress. For daily SEVIRI-ET_a, the KGE (RMSE [mm day^-1]) error metrics varied between -0.88 (0.43) to 0.93 (1.79), with a median value of 0.6 (0.77) and for daily SEVIRI-ET₀ the KGE (RMSE [mm day^-1]) varied between 0.51 (0.40) to 0.94 (1.50), with a median value of 0.77 (0.57). For daily SEVIRI-‎ET_a, intra-annual accuracy was low from January to March, increased in the mid-year, and then began to decline from November to December. Although accuracy remained relatively stable during the middle of the year, it varied considerably in the winter period. In the inter-annual dimension, the mid-year positive KGE values and distributions changed over time from 2004 to 2018. In spatial dimensions, the highest accuracy was in peat and grassland ecosystems, and the lowest in cropland ecosystems, with similar patterns observed in the boreal snow fully humid warm summer and warm temperate fully humid hot summer climate zones. Regarding SEVIRI-ET₀ results, similar to SEVIRI-‎ET_a, intra-annual accuracy was low in the first quarter of the year and the last one but high in the mid-year. In the inter-annual dimension, unlike SEVIRI-ET_a, almost an identical pattern was observed for the mid-year positive KGE values, demonstrating only a slight change in SEVIRI-ET₀ accuracy during 2004 – 2018. However, the highest accuracy was found in crop ecosystem, while the lowest was in forests, reflecting similar trends in the warm temperate fully humid hot summer and warm temperate summer dry hot summer climate zones. The observed range of median RMSE changed between 0.4 to 1.5 mm day^-1, also suggests a reasonable accuracy for SEVIRI-ET estimates in all spatial domains. The SEVIRI-ESI-based monthly water stress workflow deployed on the VLab platform offers valuable insights into spatio-temporal variations of water stress across Europe over the last decade, facilitating scientific research and practical applications. Analysis of ecosystem responses to water stress indicates observable effects on vegetated ecosystems reflected in SEVIRI-ESI-based water stress values and anomalies. Overall, our findings underscore the utility and potential of SEVIRI-ET products and the VLab platform for agricultural water stress detection at regional to continental scales.

In China, the demand for accurate and up to date information on wheat Grain Protein Content (GPC) has gained increased urgency, driven by the rising demands in the food consumption market and intensifying international market competition. However, due to the lack of extensive, prolonged high-resolution benchmark data, previous GPC studies have primarily focused on experimental fields, small geographic units, and limited temporal scopes. Additionally, the diversified geographical landscape in China introduces spatiotemporal heterogeneity and intricacy to the influence of wheat GPC, further amplifying the challenges of large-scale GPC estimation. To address this challenge and the data gap, the first 500-meter spatial resolution, long-term winter wheat dataset covering major planting regions in China (CNWheatGPC-500) was created by integrating multi-source data from ERA5 and MODIS. Firstly, a semi-mechanistic GPC estimation model based on Hierarchical Linear Model (HLM) was established by deciphering the spatial hierarchical relationships among temporal phenological, meteorological, and remote sensing data. Secondly, the performance of model was comprehensively evaluated through the integration of multi-model comparisons, cross-validation, and sensitivity analysis. Subsequently, a nationwide GPC map dataset spanning from 2008 to 2019 was generated for China. Finally, the GPC pattern was explored considering two dimensions, spatial and temporal variations. The results demonstrate that the GPC estimation model based on HLM significantly outperforms other conventional models. The calibration dataset achieved an R² of 0.57 and an RMSE of 0.89%, while the validation dataset exhibited an R² of 0.45 and an RMSE of 0.96%. In cross-validation, the RMSE values ranged from 0.90% in Gansu to 1.32% in Anhui, showing a relatively consistent pattern across provinces. For leave-one-year-out cross-validation, the RMSE values ranged from 0.77% to 1.11%, indicating consistent and satisfactory performance across different years. The spatial distribution of wheat GPC in China exhibited a clear pattern, with higher levels in the northern regions and lower levels in the southern regions. Notably, the lower latitude regions in the South showed a greater GPC increment over the 12-year period compared to the higher latitude regions to the North. This spatiotemporal pattern was influenced by the cumulative impact of intricate and regionally variable environmental factors. Overall, CNWheatGPC-500 provides valuable insights for improving wheat production, enhancing quality control, and supporting decision-making in the agricultural sector.