2 Aralık 2022 Konferans bildirisi Açık Erişim

Gopp, Natalya; Kaya, Fuat; Keshavarzi, Ali

JSON-LD (schema.org)

{
  "@context": "https://schema.org/", 
  "@id": 262957, 
  "@type": "ScholarlyArticle", 
  "creator": [
    {
      "@type": "Person", 
      "affiliation": "Siberian Branch of the Russian Academy of Sciences", 
      "name": "Gopp, Natalya"
    }, 
    {
      "@id": "https://orcid.org/0000-0003-0011-9020", 
      "@type": "Person", 
      "affiliation": "Isparta Uygulamal\u0131 Bilimler \u00dcniversitesi", 
      "name": "Kaya, Fuat"
    }, 
    {
      "@type": "Person", 
      "affiliation": "University of Tehran", 
      "name": "Keshavarzi, Ali"
    }
  ], 
  "datePublished": "2022-12-02", 
  "description": "<p>It is well documented that the yield of cultivated crops decreases when the amount of mobile zinc in the soil is insufficient. Digital mapping techniques are needed to identify areas with a shortage of plant nutrition elements. In the present research, data collected from the Novosibirsk region (Russia) (50 observations) were used to compare the accuracy of geostatistics (Ordinary kriging (OK)) and machine learning approaches (Lasso Regression (LR) and Random Forest (RF)) to map the concentration of mobile zinc in the upper horizon of the soils in order to determine which method generates maps more accurately. The effectiveness of vegetation indices and morphometric relief factors for digital mapping was assessed using machine learning methods. Fifteen vegetation-based indices were calculated by Landsat 8 OLI (resolution 30 m). Ten morphometric relief parameters were calculated using the digital elevation model SRTM v.3. In the determination of mapping performance of the machine learning and geostatistics techniques for soil mobile zinc, coefficient of determination (R2), root mean square error (RMSE), and normalized root mean square error (NRMSE) were used through the k-fold cross-validation (n:10, repeated:5). The results of the three models showed that the LR model with lower RMSE (0.43 mg kg-1) and NRMSE (17%) was the best for soil mobile zinc content prediction. The LR and RF models had the advantage of spreading the prediction results over a large area and can be used with fewer samples. The method of OK does not have such advantages, since a large number of samples are needed for its implementation, therefore is not economically profitable. The use of digital mapping methods in agricultural practice is justified since it allows for the management of plant production processes by detecting soil boundaries with a deficit of particular plant nutrition elements on the maps and considered to be key agronomic strategies.</p>", 
  "headline": "Field\u2010scale digital soil mapping of mobile zinc: Combining different digital covariates and comparing geostatistical and machine learning models", 
  "identifier": 262957, 
  "image": "https://aperta.ulakbim.gov.tr/static/img/logo/aperta_logo_with_icon.svg", 
  "inLanguage": {
    "@type": "Language", 
    "alternateName": "eng", 
    "name": "English"
  }, 
  "keywords": [
    "Digital soil mapping", 
    "Covariate selection", 
    "Lasso regression"
  ], 
  "license": "http://www.opendefinition.org/licenses/cc-by-sa", 
  "name": "Field\u2010scale digital soil mapping of mobile zinc: Combining different digital covariates and comparing geostatistical and machine learning models", 
  "url": "https://aperta.ulakbim.gov.tr/record/262957"
}