Close
Research Title
Next Generation Crop Monitoring and Forecasting system
- More about Feng
- 2019 – Present, University College London: PhD student in Remote Sensing
- 2015 – 2016, University College London: MSc in Remote Sensing (Distinction)
- 2011 – 2015, China University of Geosciences, Beijing: BEng in Resource exploration
- Publications
Yin, F., Lewis, P. E., Gomez-Dans, J., & Wu, Q. (2019, February 21). A sensor-invariant atmospheric correction method: application to Sentinel-2/MSI and Landsat 8/OLI. https://doi.org/10.31223/osf.io/ps957
- Research Interests
Food security remains one of the key global challenges in this century. The increasing food consumption driven by the growing global population, negative effects from climate changes and ever sever resource scarcity, along with other adverse factors, will consequently worsen the situation in the coming decades. In the meantime, sustainable intensification has been proposed to replace the unsustainable crop practices used in the green revolution over the last half-century. Under the increasing pressures over world food supplies, timely, comprehensive, transparent, and accurate information on crop production is critical. Since the 1970s remote sensing technology has been used to monitor crop growth and estimate crop yields because it can supply measurements related to instantaneous values of various canopy state variables. Unfortunately, inconsistent methods and data are used in current crop monitoring systems and yield gap analysis. Meanwhile, a decision support system, which has local relevance to individual farmland, is yet to exist to determine the potential risks and suggest effective procedures for all levels of users in the crop system. Lastly, equal accessibility for farmers across countries and regions should be one indispensable characteristic for any solution in securing future food status, as the benefits of the Green Revolution have been unequally shared and show large inter-regional disparities.
We propose a practical while effective crop production monitoring and forecasting system over a regional to national scale with field-level details. Data input to the system is Copernicus Sentinel 2 and Sentinel 1 time-series images, which have 10-60 meters spatial resolution to allow information derived from them to be relevant to individual farmland even smallholdings with hundreds of m^2 area. State-of-art radiative transfer models are used to reduce the atmospheric effects on optical measurements and biophysical parameter retrievals for both optical and microwave data, which guarantees the compatibility and interoperability of data and information used in our system. The crop simulation model, WOFOST, is used to bridge time series of biophysical parameters to the final crop yield, which is also mechanistic and has a high level of explainability and interoperability than the empirical methods. Machine learning methods are used to emulate radiative transfer models to allow for fast mapping from satellite measurements to crop state parameters. Since this system involves a large volume of data inputs and extensive computation that can be hardly handled by most of the users, the whole system has been deployed on Google Earth Engine, which provides all the required data as well as planetary-scale analysis capabilities for free to the public. Another very important characteristic of this system is that it is fully compatible with the Open-data policy, which serves the basics of equal accessibility to this system for all users. Currently, this system has been tested over North China Plain with an area of more than 400,000 square kilometres, and we are able to deliver winter wheat yield maps with 10 meters spatial resolution for the last three years. The application of this system will be further expanded to Ghana to help relieve the food security in that country.
