Iranian Remote Sensing and GIS Society / Shahid Beheshti University Iranian Journal of Remote Sensing and GIS 2008-5966 15 1 2023 03 21 Mapping Sugarcane Leaf Area Index by Inverting PRISMA Hyperspectral Images Mapping Sugarcane Leaf Area Index by Inverting PRISMA Hyperspectral Images 85 108 102643 10.52547/gisj.15.1.85 FA Mohammad Hajeb P.hd. Student, Dep. of Remote Sensing and GIS, Faculty of Geography, University of Tehran, Tehran Saeid Hamzeh Associate Prof., Dep. of Remote Sensing and GIS, Faculty of Geography, University of Tehran, Tehran 0000-0001-9233-8460 Seyed Kazem Alavipanah Prof. of Dep. of Remote Sensing and GIS, Faculty of Geography, University of Tehran, Tehran 0000-0002-3554-111X Jochem Verrelst Image Processing Laboratory (IPL), Parc Científic, Universitat de Val`encia, Val`encia, Spain Journal Article 2022 06 28 Leaf Area Index (LAI) plays a critical role in the mass and energy exchanges between the earth and the atmosphere. Like of other plants, LAI of sugarcane is a good indicator of the health status and growth of this crop which is of great economic importance due to its role in the food and energy industries. Launched in 2019, the PRISMA satellite provides one of the most recent hyperspectral data sources which are applicable especially for mapping plant variables. In this study, a new kind of Artificial Neural Networks (ANN) so-called Bayesian Regularized Artificial Neural Networkk (BRANN) which applies Bayes' theorem to overcome the overfitting problem of neural networks is used. The model was implemented on a data set consisting of spectrum obtained by PRISMA satellite as an independent variable and sugarcane LAI measurements as a dependent variable. The ground measurements of sugarcane LAI were carried out in 118 elementary sampling units on the fields of Amir Kabir sugarcane cultivation and industry in Khuzestan province and on seven different dates during a sugarcane growth period in 2020. Comparing the performance of BRANN in retrieving sugarcane LAI from PRISMA spectra with that of a conventional ANN trained with the Levenberg-Marquardt algorithm (LMANN) indicates that the retrieval RMSE is reduced from 2.26 m²/m² applying LMANN to 0.67 m²/m² applying the BRANN method. In this study, the principle component analysis was also used dimensionality reduction. Retrieving LAI from the first 20  principle components, RMSE was also reduced from 1.41 m²/m² applying LMANN to 0.71 m²/m² applying BRANN. Exploiting principal components significantly reduced computational time. By implementing the calibrated BRANN model over the PRISMA image pixel by pixel, the sugarcane LAI map was generated. Evaluating this map showed that this map represents the spatial variations of sugarcane LAI well. The results of this study indicate the high performance of the BRANN method and high potential of PRISMA images to retrieve sugarcane LAI. Leaf Area Index (LAI) plays a critical role in the mass and energy exchanges between the earth and the atmosphere. Like of other plants, LAI of sugarcane is a good indicator of the health status and growth of this crop which is of great economic importance due to its role in the food and energy industries. Launched in 2019, the PRISMA satellite provides one of the most recent hyperspectral data sources which are applicable especially for mapping plant variables. In this study, a new kind of Artificial Neural Networks (ANN) so-called Bayesian Regularized Artificial Neural Networkk (BRANN) which applies Bayes' theorem to overcome the overfitting problem of neural networks is used. The model was implemented on a data set consisting of spectrum obtained by PRISMA satellite as an independent variable and sugarcane LAI measurements as a dependent variable. The ground measurements of sugarcane LAI were carried out in 118 elementary sampling units on the fields of Amir Kabir sugarcane cultivation and industry in Khuzestan province and on seven different dates during a sugarcane growth period in 2020. Comparing the performance of BRANN in retrieving sugarcane LAI from PRISMA spectra with that of a conventional ANN trained with the Levenberg-Marquardt algorithm (LMANN) indicates that the retrieval RMSE is reduced from 2.26 m²/m² applying LMANN to 0.67 m²/m² applying the BRANN method. In this study, the principle component analysis was also used dimensionality reduction. Retrieving LAI from the first 20  principle components, RMSE was also reduced from 1.41 m²/m² applying LMANN to 0.71 m²/m² applying BRANN. Exploiting principal components significantly reduced computational time. By implementing the calibrated BRANN model over the PRISMA image pixel by pixel, the sugarcane LAI map was generated. Evaluating this map showed that this map represents the spatial variations of sugarcane LAI well. The results of this study indicate the high performance of the BRANN method and high potential of PRISMA images to retrieve sugarcane LAI. https://gisj.sbu.ac.ir/article_102643_c29c51343e260079d4f63ae29f23cf19.pdf