Conference Agenda

In recent years, 10 to 30 meter resolution optical multispectral and hyperspectral satellites data and SAR data from Europe and China became available and is encouraging the Remote sensing community to explore the new technology to harness all advantages of all these satellites to achieve best crop monitoring at various scale. The capability of agricultural monitoring in general is being enhanced and improved with these diverse satellite data in term of the monitoring spatial extent and the quality of the retrieved crop growth information. However, the agricultural cultivation is diverse in China and the rest of the world. There are existing large fields with mono crop and small fields with multiple strips of various crop types. It has to take this situation into account when the crop monitoring is being conducted. What are the performance differences between multispectral time series or hyperspectral imagery from both side for crop type mapping, in particular between similar crop types? What is the transferability from one year to another for crop type classification models based on multispectral time series on one hand, and hyperspectral data on the other hand? In this project, 5 study sites, 4 from China and 1 from Belgium, are selected representing the major cropping systems, including winter wheat, maize, rice, sugarcane in China and winter cereals, potato, sprint cereals and maize in Europe. These sites also will be representing the agricultural systems in the flat area or in hilly area, irrigated or rainfed, in the North and the South. The field campaigns will be organized and the data collection will be following the JECAM guideline. The Sentinal1/2, EnMAP, from Europe and GF1/6, Obita and BNUE from China as well as other satellite data will be mainly investigated to support this study. The crop classification algorithm will be evaluated with various satellite data, like optical, SAR and hyperspectral data, either alone or combination to make best crop type maps, but in particular with a focus on hyperspectral data. Through this joint project and the heavy involvement of young scientists from Europe and China, the satellite data finely processing and information retrieval algorithm will be exchanged and the objective of this project will be fulfilled as the task team brings a step forward to support agricultural monitoring at fine scale. The collaboration will also make great technological contributions to GEO Global Agricultural Monitoring Imitative, GEOGLAM.

Agriculture is the largest water user in the world and irrigation water management is facing major challenges in the sustainable development of food production and water use. The objective of the project is to evaluate the role of satellite earth observation (EOs), and in particular of optical-thermal-microwave data, in support of water management at different spatial scales (field, regional, and global). Satellite data can play a major role in assessing water availability, as well as crop water needs and stress conditions, as either stand-alone sources of information or in combination with hydro-meteorological modelling tools. In this context, satellite data represent an optimal compromise between the need to monitor large areas in a cost-efficient manner, and the need of detailed local estimations akin to ground data.

A large suite of satellite-based products is available for water balance modelling at multiple spatial scales, which are relevant for climate, meteorological, basin scale water management models and field scale precision irrigation. Nevertheless, there exists a need to understand how the accuracy of such data products affects the reliability of hydrological model simulations when used for operational application at multiple spatial scales. These needs are exacerbated by the increasing expansion of areas devoted to water-intensive crops and simultaneous impact of climate change on water availability.

Example on Chinese and Italian / European case studies will be presented on several operative topics supporting the discussion on SAA4Water project.

Chinese case studies will be presented and discussed in relation to the following application:

• Development of a decade-long gap-free daily L-band soil moisture (SM) product from SMOS observations. Current satellite-derived SM products exhibit pronounced spatial and temporal discontinuities due to orbital coverage, radio frequency interference, and retrieval failures, constraining their utility to agricultural applications. We introduce a fully automated gap-filling method that combines Discrete Cosine Transformation with Partial Least Squares (DCT-PLS) to reconstruct missing data in the SMOS Multi-Temporal and Multi-Angular (MTMA) product without relying on any external ancillary data. A seamless-continuity product (MTMA-SC_SM) at 25 km and daily resolution was generated with high accuracy (ubRMSE = 0.057 m³/m³). The seamless product preserved mesoscale patterns and seasonal amplitudes across all climate zones, with no discernible boundary artifacts around reconstructed regions. The MTMA-SC_SM dataset constitutes the first decade-long, gap-free, global daily L-band SM record, providing a robust foundation for climate trend assessment and land surface modelling at the global scale.
• Precipitation retrieval based on geostationary satellite observation. Real-time precipitation retrieval is crucial for timely warnings of extreme weather events such as heavy rain or floods. A novel dynamic two-step real-time precipitation retrieval algorithm (D-PRA) was developed by integrating geostationary satellite observations and machine learning method to achieve real time precipitation retrieval. D-PRA was applied to Himwari-8 observations and was validated with the rain gauge observations in Chinese regions at the hourly scale. The results showed that D-PRA exhibits significant enhancements when comparing with Global Satellite Mapping of Precipitation (GSMaP)_NOW. Moreover, D-PRA is stable across seasons and time periods, demonstrating its good reliability and robustness.
• Estimation of actual evapotranspiration (ET) of soil-vegetation system by combining a physics-based model and machine learning algorithm. Incorporating machine learning (ML) into physics-based framework strikes a balance between process constraints and data-driven efficiency. A hybrid model was proposed by integrating ML-based canopy surface resistance estimation into the Shuttleworth-Wallace (SW) dual-source scheme under the ETMonitor framework, replacing traditional physics-based resistance parameterization, to improve ET estimation. The integrated hybrid model (ETM-SW-DNN) outperformed the others, e.g., original physics-based model, full ML models (e.g., RF, GBDT, DNN), other hybrid models (e.g., ETM-SW-RF). The ETM-SW-DNN hybrid model mitigates the under- or over-estimation biases commonly found in full ML models, produce fewer outliers and maintain reliable performance for PFTs with limited training data. The hybrid models show better performance when more explanatory variables are used, indicating the importance of using comprehensive input variables in ML training.
• Evaluating cropland water use efficiency (WUE) at high resolution. The WUE framework, developed in our previous study for global application at moderate resolution (1-km), was applied to high resolution (30-m) to support field scale WUE evaluation. We focused on two maize croplands located in different climatic regions: one in northern China and the other in northern Italy. The estimated ET and GPP, both at 30-m resolution, showed satisfactory performance when comparing with the eddy covariance tower measurements as reference data. The resulting WUE values were found to be highly similar across the two croplands. The results indicate that higher spatial resolution provides valuable agronomic and environmental insights into vegetation cropland water use. They also demonstrate that, when high-resolution inputs are adopted, the estimation framework performs reliably after site-specific recalibration, which appears not only to the climatic characteristics, but also to the crop types.

European / Italian case studies will be presented and discussed regarding operative application:

• Precise irrigation supported by satellite information will be analysed providing a discussion on the satellite spatial resolution and temporal frequency required: the FEST-EWB model has been integrated with satellite land surface temperature (Sentinel3, MODIS) and soil moisture (Sentinel1) for improving ET estimates and irrigation water needs quantification in the Po Valley. Comparison with ground data allows quantifying the estimates accuracy.
• The study uses the Integrated Multi-satellitE Retrievals for GPM (IMERG) product, produced under the Global Precipitation Measurement (GPM) mission, to analyze precipitation across the Po valley study region. IMERG combines passive microwave and infrared sensor data from a satellite constellation to generate half-hourly precipitation fields at 0.1-degree spatial resolution. The Final run, which incorporates monthly gauge adjustments from the Global Precipitation Climatology Centre (GPCC), is selected here because it carries the lowest systematic bias of the three available IMERG products.
• Large scale estimation on ET using thermal data from meteorological satellites. These data represent a unique source of information, given their high temporal frequency (sub-hourly) and extensive spatial coverage. ET estimates based on these datasets rely on a combination of satellite and reanalysis data, suffering from uncertainty driven by potential inconsistencies between the two data sources. Applications over Europe highlight the major sources of errors in ET estimates, in both their temporal and spatial patterns.