Day 4

Detailed paper information

Back to list

Paper title Crop biomass estimate using Sentinel-1 and COSMO-SkyMed data in Italy
Authors
  1. SIMONETTA PALOSCIA PLSSNT56B57D612S Speaker
  2. Emanuele Santi National Research Council - Institute of Applied Physics (CNR - IFAC)
  3. Simone Pettinato CNR-IFAC
  4. Alessandro Lapini CNR-IFAC
  5. Giacomo Fontanelli CNR-IFAC
  6. Simone Pilia CNR-IFAC
  7. Fabrizio Baroni CNR-IFAC
  8. Leonardo Santurri CNR-IFAC
  9. Giuliano Ramat CNR-IFAC
Form of presentation Poster
Topics
  • A3. Biosphere
    • A3.06 Biomass monitoring
Abstract text Modern agriculture should combine the needs of productivity with those of environmental, economic and social sustainability, in an uncertain climate context due to the effects of climate change. Information useful for implementing advanced and integrated monitoring and forecasting systems to promptly identify the risks and the impacts of calamities and crop practices on agricultural environments are essential. Satellite Earth observation data revealed to be optimal for the aforementioned tasks because they can cover wide areas with different spatial resolutions and frequent revisit time, allowing the collection of historical series for long-term analysis, and they can be punctual thanks to the continuous acquisition of Copernicus constellations. Finally, from an economic point of view they are becoming more convenient thanks to the provision of free satellite data and dedicated software for their processing and display.
Agricultural ecosystems are characterized by strong variations within relatively short time intervals. Depending on the observation period the agricultural scenario can present itself in a totally different way, due to the difference in biomass and phenological cycle, that can be driven by cultivar and agricultural working, as well as weather conditions. These dynamics are challenging for crop monitoring and the knowledge of vegetation status can deliver crucial information that can be used to improve the classifiers performance.
In order to consider these aforementioned changes in agricultural vegetation and soil status, a multitemporal approach based on the study of time series of SAR indices can be successful. Time series of satellite images offer the opportunity to retrieve dynamic properties of target surfaces by investigating their spectral properties combined with temporal information on their changes.
This research work was carried out using SAR images from Sentinel-1 (at VV and VH polarizations) and COSMO-SkyMed (at HH polarization for Himage and VV+HH polarizations for PingPong) satellite sensors, which have been collected for a few years over an agricultural area in central Italy. The sensitivity of backscattering and related polarization indices at both C and X bands was investigated and assessed in several experiments. Both frequencies revealed to be sensitive to crop growth although with different behaviors according to crop type, the backscatter being influenced by the two phenomena of absorption and scattering caused by the dimensions of leaves and stems. In particular, crops characterized by large leaves and thick stems cause an increasing of backscattering as the plants grow and the biomass increases; whereas crops characterized by narrow leaves and thin and dense stems cause a decreasing trend of backscattering during the growth phase. Typical representatives of these two types of crops are wheat for the first case and sunflower for the second one.
First of all, an accurate crop classification was performed in order to identify the various crop types responsible for the different backscatter behaviors. The backscattering trends have been simulated by using simple electromagnetic models based on radiative transfer theory. Subsequently, algorithms based on machine-learning approaches and in particular Neural Network methods have been implemented for estimating the crop biomass by using multi-frequency and multi-polarization SAR data at C and X band.
To this scope, an «experimental + model driven» approach was adopted. In detail, the ANN training was based on subsets of experimental data combined with model simulations, while testing and validation have been carried out using the remaining part of experimental data. This strategy preserved the statistical independence between training and validation sets, by also overcoming the site dependency of the data driven approaches based on experimental data only, thus ensuring some generalization capabilities of the proposed algorithms.
Although still preliminary, the results obtained are encouraging, confirming the peculiar sensitivity of each frequency to different vegetation features, and enabling the mapping of vegetation biomass in the test area with satisfactory accuracy.