@article{fdi:010095684,
  title = {{E}valuation of nonparametric machine-learning algorithms for an optimal crop classification using big data reduction strategy},
  author = {{A}l-{A}war, {B}. and {A}wad, {M}.{M}. and {J}arlan, {L}ionel and {C}ourault, {D}.},
  editor = {},
  language = {{ENG}},
  abstract = {{A}ccurate crop classification can support analyses of food security, environmental, and climate changes. {M}ost of the current research studies have focused on applying available algorithms to classify dominant crops on the landscape using one source of remotely sensed data due to geoprocessing constraints (e.g., big data access, availability, and processing power). {I}n this research, we compared four classification algorithms, including the support vector machine ({SVM}), random forest ({RF}), regression tree ({CART}), and backpropagation network ({BPN}), to select a robust and efficient classification algorithm able to classify accurately many crop types. {W}e used multiple sources of satellite images such as {S}entinel-1 ({S}1) and {S}entinel-2 ({S}2) and developed a new cropping classification method for a study site in the {B}ekaa valley, {L}ebanon, fully implemented on {G}oogle {E}arth {E}ngine {P}latform, which minimized those geoprocessing constraints. {T}he algorithm selection was based on their popularity, availability, simplicity, similarity, and diversity. {I}n addition, we adopted different strategies that included changing the number of crops. {T}he first strategy is to reduce the number of collected {S}2 images thereafter {S}1; the second strategy is to use {S}2 images separately and then combining {S}2 and {S}1. {T}his study results proved that the {RF} is the most robust algorithm for crop classification, showing the highest overall accuracy ({OA}) (95.4%) and a kappa index of 0.94, followed by {BPN}, {SVM}, and {CART}, respectively. {T}he performance of these algorithms based on major crop types such as wheat or potato showed that {CART} is the highest with {OA} (98%) followed by {RF}, {SVM}, and {BPN}, respectively. {N}evertheless, {CART} fails to classify other minor crop types. {W}e concluded that {RF} is the best algorithm for classifying different crop types in the study area, using multiple remote sensing data sources.},
  keywords = {{LIBAN} ; {BEKAA} {VALLEE}},
  booktitle = {},
  journal = {{R}emote {S}ensing in {E}arth {S}ystems {S}ciences},
  volume = {5},
  numero = {},
  pages = {141--153},
  ISSN = {2520-8195},
  year = {2022},
  DOI = {10.1007/s41976-022-00072-7},
  URL = {https://www.documentation.ird.fr/hor/fdi:010095684},
}