Speaker
Description
This study deals with the high-resolution mapping of soil moisture (SMC) and crop biomass (expressed as plant water content PWC) in agricultural fields by using SAR data at C and X bands.
For this purpose, timeseries of Sentinel-1 (S-1, at VV and VH polarizations) and COSMO-SkyMed (CSK, StripMap Himage at HH polarization) images have been collected for some years in an agricultural area located in Tuscany (Central Italy), mainly covered by wheat, corn, sorghum, sunflower, pastures, and vineyards. CSK images have been kindly provided by the Italian Space Agency (ASI) in the framework of scientific projects and Open Calls, whereas S-1 and S-2 data have been downloaded from Copernicus system. In situ measurements of the main soil and vegetation parameters, including PWC, SMC and soil roughness, have been collected in correspondence of each satellite overpass to provide the reference data.
The backscattering (°) dependence on SMC and PWC was firstly investigated at both frequencies. The sensitivity analysis confirmed the better sensitivity of X- than C- band to PWC of agricultural fields, by pointing out the well-known dependence on crop type. In detail, crops characterized by large leaves and thick stems (broad leaf crops) cause an increasing of backscattering as the plants grow and the biomass increases; whereas crops characterized by narrow leaves and thin and dense stems (narrow leaf crops) cause a decreasing trend of backscattering during the growth phase [1]. Conversely, the analysis of ° as a function of SMC confirmed that C- is more sensitive to SMC than X- band, being the latter more influenced by the vegetation effects.
Based on these results, a method for estimating SMC and PWC at high resolution based on S-1, CSK, and machine learning (ML) has been developed. The method articulated in three steps: as first, a fully supervised crop classification based on CSK data and deep learning has been implemented to separate broad- and narrow-leaf crops. The classifier has been trained with ground data gathered during the measurement campaigns [2].
In the second step, SMC maps at high resolution have been generated by using the Artificial Neural Networks (ANN) based approach proposed in [3], which exploits the S-1 images for disaggregating low resolution SMC product generated from AMSR2 radiometer by the IFAC’s HydroAlgo algorithm [4].
In the third step, the CSK images are used as input of the PWC retrieval algorithm together with the maps of classified crops and SMC. The PWC algorithm is based on ANN as well: to generate a dataset sufficient for training, the experimental measurements have been merged with data simulated by a forward electromagnetic model obtained by coupling the Water Cloud Model (WCM – [5]) with the Advanced Integral Equation Model (AIEM – [6]).
The obtained results were almost encouraging: the validation of SMC algorithm carried out against in-situ measurements resulted in correlation coefficient (R)= 0.74 and Root Mean Square Error (RMSE)=0.034 m3/m3 when using S-1 only and R= 0.89 and RMSE =0.025 m3/m3 when exploiting the S-1 and CSK synergy. The validation for PWC algorithm was carried out independently for each crop type: for example, it reached R=0.92 and RMSE =0.58 kg/m2 for the wheat crops. The prosecution of this research will aim at extending spatially and temporally the investigations, by including other crops and seasons and the intercomparison of other ML approaches, as random forest (RF) and Support Vector Regressor (SVR).
REFERENCES
[1] S. Paloscia, E. Santi, G. Fontanelli, F. Montomoli, M. Brogioni, G. Macelloni, P. Pampaloni, S. Pettinato, 2014, The Sensitivity of Cosmo-SkyMed Backscatter to Agricultural Crop Type and Vegetation Parameters, 2014, IEEE Journal of Selected Topics In Applied Earth Observations and Remote Sensing, 7, 7, July 2014, pp. 2856-2868
[2] Fontanelli, G., Lapini, A., Santurri, L., Pettinato, S., Santi, E., Ramat, G., . . . Cigna, F., Paloscia, S., 2022, Early-Season Crop Mapping on an Agricultural Area in Italy Using X-Band Dual-Polarization SAR Satellite Data and Convolutional Neural Networks, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2022, 15, pp. 6789–6803
[3] Santi E., F. Baroni, G. Fontanelli, A. Lapini, E. Palchetti, S. Paloscia, P. Pampaloni, S. Pettinato, S. Pilia, G. Ramat, L. Santurri, "High Resolution Mapping of Vegetation Biomass and Soil Moisture by Using AMSR2, Sentinel-1 and Machine Learning," IGARSS 2022 - 2022 IEEE International Geoscience and Remote Sensing Symposium, 2022, pp. 4943-4946, doi: 10.1109/IGARSS46834.2022.9884950.
[4] Santi E., S. Pettinato, S. Paloscia, P. Pampaloni, G. Macelloni, and M. Brogioni, 2012, “An algorithm for generating soil moisture and snow depth maps from microwave spaceborne radiometers: HydroAlgo”, Hydrol. Earth Syst. Sci., 16, pp. 3659-3676, doi:10.5194/hess-16-3659-2012.
[5] Attema, E. P. W. & Ulaby, F. T., “Vegetation modeled as a water cloud”. Radio Science, 13, 357-364 (1978).
[6] Wu, T.D. & Chen, K.S., “A Reappraisal of the Validity of the IEM Model for Backscattering From Rough Surfaces”. IEEE Transactions on Geoscience and Remote Sensing, 42(4), 743-753 (2004).
Summary
This study aims at the high-resolution monitoring of soil moisture (SMC) and crop biomass (expressed as plant water content PWC) in agricultural fields by using SAR data form Sentinel-1 C- band (S-1, at VV and VH polarizations) and COSMO-SkyMed X- band (CSK, StripMap Himage at HH polarization) and algorithms based on machine learning.
