Regional crop monitoring and assessment with quantitative remote sensing and data assimilation

China has only 10% of the world arable land and water resources, but has to feed 20% of the world population. Moreover, the population continues to increase, while the amount of arable land is shrinking due to pollution, urban sprawl, groundwater depletion, and other stresses. With future climate change expected to only worsen these pressures, the accurate monitoring of agricultural productivity is essential to China's future food security, in addition to the economic development of low-income rural regions. In no part of the country is this more essential than China's north plain. This has historically been the breadbasket of China. Today, however, it faces an exceptionally challenging combination of very high population densities and ecological stresses, and low levels of household income.

Traditionally, researchers have used two general methods for monitoring agricultural productivity. The first, which has long been used by Chinese government agencies, is to combine field surveys of crop growth, with mathematical models of crop growth processes, to construct estimates of changing harvest yields over time. The second, which has risen to prominence more recently, is to use satellite imagery to continuously assess agricultural productivity. Each of these techniques has its own notable strengths and weaknesses. Survey-calibrated models of crop growth are able to produce highly accurate estimates of yields in the limited areas where survey data has been collected; however, their accuracy drops off significantly outside of these areas. On the other hand, satellite remote sensing data offers universal geographic coverage; however, the resolution of this data is extremely coarse over either time or space. MODIS data, for example, provides near-daily data that can be used to assess the productivity of every single farm in China. However, the spatial resolution of pixels is only 500-1000 meters, an area which will invariably be contaminated, in densely populated China, by a mixture of roads, villages, and other non-agricultural land uses in addition to the farmland actually being studied. Other satellites (e.g. LandSat TM and forthcoming Sentinel) provide finer scale pixel resolution than MODIS; however, they do not cover the same sites as often, making it harder to smoothly track agricultural production over time.

Reflecting the wider explosion of the field of "big data" analysis, rapid strides have been made recent years in the development of so-called "data assimilation" techniques. These can be broadly described as statistical methodologies that allow for otherwise incompatible datasets to be combined together, in order to produce hybrid datasets that are superior to any of their predecessors. The basic objective of the proposed project is to apply advanced data assimilation techniques to multiple types of crop data-from both survey-calibrated crop growth models and satellite imagery-to produce superior estimates of Chinese agricultural productivity than would be possible using any of these data sources by itself. In addition to making use of more advanced statistical methods than previous studies, this analysis will be among the first to make use of data from the forthcoming Sentinel and the Chinese GF satellites. Taken together, we expect that the result will be the most accurate portrait created to date of changing agricultural production in the North China Plain. Moreover, having created this data, we will be able to apply it predictively in conjunction with modelled scenarios of future climate change, in order to map and assess the likely geographies of agricultural stress that this will create. Ultimately, the findings of this project will directly inform work by academic researchers, national and regional Chinese governmental authorities, agritech companies in both China and the UK, and extension workers directly advising farmers in China.

The outputs of the proposed research will feed directly into agricultural production planning in China. With ever increasing demand for staple foods, decreasing availability of agricultural land, and a requirement to control the use of fertilizers, pesticides and water for irrigation, accurate and timely management information is essential. Our research will deliver improved simulations and predictions of crop performance that will assist in strategic planning, optimizing regional production and managing national food supply and security. The lead partner in China, CAAS-IARRP runs the China Agriculture Remote Sensing Monitoring System (CHARMS) which monitors crop acreage change, yield, production, drought and other agriculture-related information for 7 main crops in China. CHARMS provides information directly to the Ministry of Agriculture (MoA) and related agriculture management sectors in the form of detailed reports to MoA through the growing season. The optimizations being undertaken in the project will allow the use of advanced DA techniques on regional/national scale which had previously been impossible due to the high computational demand of the algorithms. This will feed directly into improving the quality and usefulness of CHARMS outputs. Understanding seasonal crop production and yield outcomes for each growing season helps with national scale planning of food production and gives advanced warning of any significant imbalance food balance between production and demand. Our research to improve the information content and quality of crop predictions will therefore feed directly into the key Chinese agricultural management stakeholders.

In addition to the primary seasonal production applications, the research will also contribute to improve long term planning. Physically based crop yield simulation using EO data will allow crop responses to different climate scenarios to be modelled more accurately, giving simulations that are more robust to changes in climate than the currently widely used NDVI curve matching approaches. The results of the project will also be disseminated via the Group on Earth Observations GEOGLAM and GECAM initiatives to which the CHARMS contributes. At the National scale our research will support analysis of yield performance at local/regional scale and will contribute to development of adaptation strategies to optimize agricultural production under different climate scenarios.

The impacts of our research will initially be focused on regional and national scale monitoring, also have a longer term application in farm scale and precision agriculture applications Increasing the resolution at which simulations can be run, and improving the information content of products through the use of crop models will allow more regionally specific information on crop performance to be derived. The optimization techniques being developed in this project will allow simulations and forecasts to be run at higher spatial resolutions, taking advantage of new satellite data such as that from Sentinel-2. The use of crop growth models will provide more precise analytics of crop development and responses to different stresses which will allow more suitable management practices to be deployed. We will work proactively with the Newton Agritech Network+ which is currently being established, to promote the results of the project amongst commercial precision farming operators in China.

In addition to the specific applications and users described here the techniques and tools developed in the project will be disseminated within the science community to promote wider uptake. Both BNU and PKU operate summer schools in advanced remote sensing techniques and these schools will be used as method for dissemination to graduate students and early-career researchers. In the UK UCL is already making extensive use of these notebooks for training purposes and will promote these tools through the wider NCEO to maximize uptake.