Publication Date:
2020-02-21
Description:
Agriculture monitoring and yield estimation are important for food security, both globally and locally. Understanding year-to-year variability in crop yield and it’s relationship to meteorological conditions is especially important for countries and regions where the yield losses due to environmental factors are already observed and further predicted under the impact of climate change. One of such countries is India, where wheat is a staple food for a large part of the population and where the wheat production is the second largest in the world. Wheat yields in India have been steadily increasing since 1960s and 1970s due to benefits of the Green Revolution. However, in recent years the wheat yield has been unstable as major crop yield losses were attributed to unfavorable meteorological conditions. Including various environmental variables in the crop yield modeling can be challenging, as their impact on the final yield is complex and varies depending on the crop growth stage. For example, moderate rainfall is beneficial for crops, but extensive and untimely rainfalls can lead to huge yield losses, as in 2015, which points out the importance of accounting for such non-linear processes. In this work, we exploit the relationship of meteorological conditions during the whole growing season simultaneously with the satellite data on vegetation, and the wheat yield in the Wheat Belt in India. We use GLDAS 2.1 data as the meteorological input and MODIS data for the vegetation remote sensing input. We show that adding satellite information on crop is crucial for yield estimation, as it carries information on both crop phenology, as well as on the crop response to the meteorological conditions. We apply the convolutional neural networks (CNNs) that can model non-linear processes and can extract important features on their own, without prior knowledge or human effort in feature design. This also means that we do not force assumptions on which time is the most important for the final crop yield. By doing so, we can include in the analysis the whole time series of multiple input variables at a high temporal resolution (4 days), and model nonlinear interactions among them. This is especially important for the studied region, as the timing of wheat planting and ripening vary across the Indian counties and inter-annually due to farmer practices and various meteorological conditions. Furthermore, we analyze the CNNs in terms of important features and important time windows for yield estimation, which shows that these vary across space and time. By combining meteorological and satellite vegetation data with CNNs this work may help to disentangle the complex interactions between the features in the time series of the input data and the wheat yield.
Type:
info:eu-repo/semantics/conferenceObject
Permalink