observations from ground (scatterometer) and space
(satellite) are useful in assessing the variations of backscattering due to the growth of rice plant.
simultaneous observation is difficult.
Thus, studying and comparing the relationship
between rice plant and ground-based instrument vs rice plant and space borne sensor is important. Objectives of the study To study biomass from rice paddies through analyses of the multi-temporal simultaneous datasets and to invert biomass of rice plant from rice paddies in the study area.
In particular, the scatterometer data are used in the
development and validation of a backscatter model and NN inversion algorithm. Method.. Q sets of rice growth parameters , , , , , , ,
Growth parameters plus needed
model inputs values are derived.
Q sets of backscattering coefficients , , are modeled using backscattering model.
We train the NN with { ,
, ), } Until the RMSE of Biomass Rice biomass retrieval using neural network (NN), the from NN is <0.5 kg/2 procedure consists of a. simulation and training b. inversion and validation c. biomass mapping formulas The validation consists of two aspects Validation on the site:- the derived biomass data are assessed with those measured in the field.
The modeled backscattering
coefficients are applied to the NN to output the biomass that is compared with the biomass from the growth model.
SAR image classification was done
using SVM classifiers. Rice backscattering, coefficients are extracted from rice map and input to NN to create rice biomass map.
Rice biomass retrieval using neural network (NN), the
procedure consists of a. simulation and training b. inversion and validation c. biomass mapping Results
The trend of simulated biomass values
was similar with the observed one, however, the values fluctuated because each parameter of the growth model was generated based on an individual random distribution function that reflects the variability of the actual sampling data.
Observed and simulated biomass values after
transplanting. Eight open circles representing measurements made at eight stages of the growth period in 2012. Conti.. Monte Carlo simulations At HH, the modeled and observed backscattering coefficients were close to each other until ripening stage. After that stage, the model overestimated the backscattering.
At VV, the modeled and observed
backscattering coefficients were also close to each other except for the tile ring stage, at which the measured value was larger than the modeled one.
Measured and modeled backscattering coefficients
versus rice age. Comparisons of multitemporal scatterometer and radarsat-2 datasets
The average errors of four
simultaneous observations were -0.41,-0.64, and 0.87 for HH, VV, and HV, respectively, and the correlation coefficients were 0.93, 0.87, and 0.99.
Therefore, the backscattering
coefficients agreed with each polarization in the rice-growing season, although the spatial scale of observation is different. Comparison of retrieved biomass by the NN and simulated biomass by the growth model. Conti..
Multitemporal rice biomass data measured and NN retrieved
Conti..
RMSE = 0.582 kg/m & r= 0.983
and the Overall, results show that a good retrieval performance was obtained using the NN inversion model.
Multitemporal RADARSAT-2 images can be used to monitor the growth of rice very well.
Comparison of inverted and measured biomass
of rice plants at four dates, using RADARSAT-2 images as inputs to the NN. Conclusion
The inversion result was satisfactory because RMSE =
0.582 kg/2 & R= 0.983 was achieved by comparing NN-retrieved and measured biomass data.
In summary, using the trained NN, successful retrieval
of rice plant biomass with the acquired multitemporal ground-based scatterometer data and space borne RADARSAT-2 images has been achieved. Thank you for your attention!!!!