Effective monitoring of our agricultural lands is a key component in the fight for global food security, and a shared global challenge that can only be addressed through international collaboration across countries, organizations and sectors, and through innovation in science, technology and more open sharing of data, methods and expertise. At the international level, GEOGLAM is addressing this challenge by providing coordination across national and international programs to enhance the use and uptake of Earth Observations for agricultural application and for articulating future observation requirements for agricultural monitoring. It is implemented through a broad set of well-coordinated international initiatives and programs. One of these key programs is NASA Harvest, under which activities including the GEOGLAM Crop Monitor are coordinated. It was launched in November of 2017 with the shared GEOGLAM vision to advance the awareness, use, and operational uptake of satellite-based Earth observations to guide decisions that support food security, stable markets, economic progress, and sustainable, resilient crop production. Harvest is a NASA Applied Sciences program on agriculture and food security that is led by University of Maryland in partnership with the NASA Goddard Space Flight Center Office of Food Security and reflects NASA’s contribution to GEOGLAM. It is made up of a diverse consortium of over 40 partners from government, private sector, academia, UN and non-profit organizations across the agricultural sector building on the strong international GEOGLAM Community of Practice and associated initiatives. It is a stakeholder driven program, motivated by the fact that more timely and accurate agricultural information, enabled by EO data and technologies, could significantly enhance key agricultural decisions which are constantly being made, whether by humanitarian organizations, governments, insurance companies or farmers. Its program of activities is focused on addressing some of the longstanding challenges for full utilization and uptake of remote sensing technologies including, developing more robust and operationally viable crop monitoring capabilities across scales and cropping systems, openly available ground data, sustainable technology transfer, and communications between communities. Today, we are in a new era of satellite data availability, in large part owing to the Sentinels, with major advances in capability with respect to just 3 years ago. This, coupled with rapid developments in cloud and high performance-compute capabilities, big data analytics, GPS technologies and citizen science, is revolutionizing both the RS field and our ability to realize the goal shared by Harvest, GEOGLAM and agricultural monitoring community writ large, of providing accurate, timely information reliably across cropping systems at scale. This presentation will provide an overview of Harvest, and will discuss its program of activities with examples of its impact within the GEOGLAM framework.

Identification of agricultural crops through Earth Observation (EO) data-driven models applied over Europe is of main interest to crop modellers, scientists and decision makers in the fields of policy and food security. Many crop monitoring institutions working at large scale, such as (inter)governmental organizations or NGO’s, identified maps of agricultural crop types as critical data gaps in their crop monitoring or forecast systems.
We propose a semi-automatic way to identify pure pixels clusters of different crop groups using satellite remote sensing based time series. Here, we describe an almost parameter-free method, developed to identify pure pixels covering broad crop categories or crop groups (e.g. summer crops) within the 28 member countries of the European Union, exploiting satellite data from the MODIS sensors AQUA and TERRA at 250 m spatial resolution. Pre-processed and smoothed daily MODIS NDVI time series ranging from 2001 to 2017 are used to first extract phenological variables for a defined growing season period. As phenological indicator, the Day of Year (DOY) when the NDVI seasonal peak occurs (DOY_VImax), turned out to be the most suited variable. In order to account for regional characteristics (varying climate, agro-management, crop varieties, etc.), the phenological data are clustered region-wise and year-wise using a Gaussian Mixture Model, based also on regional agricultural acreage statistics from EUROSTAT to determine the number of clusters. After pixel labelling, following DOY_VImax chronology, clusters are filtered based on agronomic knowledge, phenological information extracted from the same time series and statistical indicators, to achieve an archive of yearly labelled pure pixels. Ways to find pure pixel clusters as early as possible within a season are explored and assessed at various dates during the growing season. Validation based on different high resolution remote sensing data products shows good to high agreement depending on region and validation type, and explores as well the effective purity degree of the clusters.

The ESA-Sen2-Agri system opens unprecedented capabilities for operational crop monitoring and crop type identification. It was built to automatically identify the major crops grown in a given region using time series of Sentinel-2 and Landsat-8 satellite images over a growing season of interest. The system has been designed to be as generic as possible in order to be applicable for any cropping system around the globe. Users may have a wide range of specific requirements in terms of products generated with the system. Indeed, some might be interested in the area estimate for just one of the major crops, others would like to be able to identify crops as early in the season as possible or might be interested in getting accurate area statistics for all major crops in a given region at the end of the season.

The quality and quantity of the in-situ data used for training of the machine learning algorithm for crop classification is one of the main drivers of the output map quality. However, collecting good quality data can be challenging and time consuming, and therefore extremely costly. Recommendations for in-situ data collections should therefore be tuned to be as cost efficient as possible.

Leveraging the generic capabilities offered by the ESA-Sen2-Agri system, this study aims to analyze and determine specific data collection guidelines in the context of operational crop monitoring. The benchmarking site was an irrigated region in the Yaqui Valley, northwestern Mexico. We studied two winter seasons: 2016-17 and 2017-18. In that region, farmers are required to report to the irrigation authorities the crop types and the area they are sowing. The Yaqui Valley has an elaborate scheme in which each crop field is identified by a block and lot number, though some of the field boundaries have changed over the years. Making use of this massive dataset, we were able to allocate the crop type to each field in an area measuring 900 km2. This resulted in a data set consisting of roughly 7000 fields for each of the 2 seasons. Wheat was the predominant crop, covering about 70 % the fields, followed by chick pea, maize and kidney beans, each representing about 5 %. Alfalfa, potato, safflower, asparagus and chile covered about 2 %. All in all, farmers reported 27 different crops for the region we analyzed.

In order to address the requirements for in-situ data collection, we will report our findings on
• optimal size and composition of the training data set
• effect of sampling method on mapping accuracy (wind shield survey vs random sampling)
• evolution of the classification accuracy throughout the growing season
• how well we can predict the crops in the second year based on the training from the first year

Operational products for global agricultural monitoring will play an important role in addressing some challenges that the world will face in the coming years. The capabilities of the Sentinel satellite missions with short repeat frequency and large coverage offer new opportunities to study the whole diversity of the agricultural landscapes. These perspectives are important from multiple perspectives such as researchers', farmers', regional and national authorities' or policy makers'.

The recently launched Sentinels Synergy for Agriculture (SensAgri) project aims to exploit the unprecedented capacity of the radar and optical payload of the Sentinel-1 and Sentinel-2 satellites to develop an innovative portfolio of prototype agricultural monitoring services. This project is financed by the European Commission under a H2020 program with the objective to propose and complement Copernicus land services. In the framework of this project, more robust, accurate, frequently updated and comprehensive crop maps are expected from the seldom exploited synergy of both types of measurements. Specifically, one major goal of the project is to develop a prototype service capable of mapping crops several times across the agriculture season.

In the past, few studies integrated both radar and optical dense image time series due to low data availability. However, the complementarity of these two types of information for crop classification has been widely proved in the literature. The most valuable advantage of radar data is that they are not affected by cloud coverage, thus allowing continuous multiple measurements along the short dynamic growing season of crops.

The results assessing the capabilities of combining Sentinel missions for crop monitoring will be the main focus of this presentation. The results obtained on different european test sites will be presented.

Accurate and timely crop yield forecasts are critical for making informed agricultural policies and investments, as well as increasing market efficiency and stability. Earth observation data from space can contribute to agricultural monitoring, including crop yield assessment and forecasting. This work presents a new EO-based empirical winter wheat yield model. It is based on the un-mixing of the wheat signal from a coarse resolution EO data at 1 km by using yearly crop type masks to a pure wheat signal (100% of wheat within the pixel). This wheat signal is used to calibrate the model using as inputs the seasonal amplitude and length of the Difference Vegetation Index (DVI) peak extracted from Moderate Resolution Imaging Spectroradiometer (MODIS) data and the average of the evaporative fraction (EF) 30 days after the peak. The three regressors proposed are focused on the reproduction stage of the wheat (DVI amplitude and length) and the grain filling process (EF 30-days average). The model is calibrated at subnational level using historical statistics from 2001 to 2016. In each administrative unit, a different calibration coefficient (based on all possible combination of the three regressors in a linear model) is selected depending on the statistical significance of each variable. The model was applied to estimate the national and subnational winter wheat yield in the United States, Ukraine, Russia and France from 2001 to 2017. At the subnational level the model shows very good performance with a coefficient of determination higher than 0.7 and Root Mean Square Error (RMSE) lower than 15%. At the national level the model provides a strong coefficient of determination higher than 0.8 and a RMSE lower than 8%, which demonstrates good performance of the models at this scale. Additionally, it was also able to capture low winter wheat yields during years with extreme weather events, for example 2002 in US and 2003 in Ukraine.

On a global scale, rice cultivated area accounted for 160 Million ha. However, rice is a very sensitive cereal crop which is significantly affected by the adverse impacts of climate change. In the last decade, rice has become increasingly threatened by the effects of drought. Drought stress greatly influences the growth duration, but also leads to damages during reproductive stages of the rice crop. More important, to cope with the increasing rice demand due to population growth, and because of the limitation in the land available for agriculture, rice production will have to increase through cropping density and increasing yield. The latter can be achieved through cultivation practices. Yet, increasing rice growth and increasing cropping density can stimulate methane (CH4) emissions, exacerbating global climate change, as rice cultivation is a major source of this powerful greenhouse gas. Furthermore, increasing irrigated rice will require more water during the dry season, which impacts on the water availability for people consumption.
The challenge of maintaining food security while reducing water used and greenhouse gas emissions is an important tradeoff issue for scientists and policy makers. An evaluation of tradeoffs requires understanding of the complex interactions between rice growth, water and GHG emission processes across spatial scales and over time. In many research works, in situ experiments have been conducted to capture key processes in crop biomass formation and in methane emission. Subsequently, experimental data have been used to parameterize models on crop growth and development, and on soil denitrification-decomposition processes.
These process-based models can simulate yield or methane emission, but also the water demand by capturing biophysical growth drivers (microclimate, water, and nutrient). However, for regional upscaling to obtain the regional crop productivity, and the regional methane emission, most studies rely on the concept of ‘homogeneous’ cropping zones for which the same processes and the same estimates of yields, water, or methane emissions are assumed.
In this study, we assess the use remote sensing for monitoring the seasonal and inter-annual variation of rice crop acreage, and for regional upscaling of the process-based models.
Specifically, we assess the variation of the rice grown maps produced using of Sentinel-1 data in the GeoRice project in the Mekong Delta from 2015 to present. The rice harvested areas during 3 years with 3 rice seasons per year showed important inter-annual changes, for example an increase of 16.7% has been observed between 2015 and 2016, caused by the drought and saline water intrusion effect due to El Niño. This decrease in area was followed by an increase in 2017 to compensate for the loss of rice stock.
Second, we derive from the GeoRice results the spatial and temporal information related to agro-practices (e.g., crop establishment dates) and seasonal crop development (i.e., phenology) and vegetation status (e.g., height, biomass, etc.), in order to apply the process-based models on a pixel basis for regional application. For this purpose, we use the same input parameters derived from Sentinel-1 in the two major models for rice crop, ORYZA 2000 for yield prediction, and the DNDC model for methane emission and water use.
The models are first tested against experimental data using all the input parameters from field survey. In a second step, the input parameters describing the cultural practices and crop development in the models are replaced by the information retrieved from Sentinel-1. Good agreement has been obtained between the simulations using survey data and using Sentinel-1 derived data, of rice yield, CH4 emission and water demand for each of the 3 rice seasons in 2016. The results for the fields under two major irrigation schemes (continuous flooding and Alternate Wetting and Drying AWD) indicated that AWD reduces in most cases water and CH4, without decreasing the yield. It was also shown that during the dry season 2016, the water demand was the highest and the yield was the lowest.
This preliminary work, which will be completed with more datasets, shows a concept that could be extended regionally to estimate the regional rice production, the water demand and the CH4 emission.