A simulation study on the comparison of Diagnosis and Recommendation Integrated System (DRIS), Modified-DRIS (M-DRIS), and Compositional Nutrient Diagnosis (CND) for pineapple nutrient diagnosis

Foliar diagnostic helps assess plant nutritional status and drives appropriate fertilizer recommendations to enhance quality and productivity of plants. Several foliar diagnostic methods are used but the literature is not sufficiently documented regarding the comparison of these methods using a varied range of comparison criteria. This study compared DRIS (Diagnosis and Recommendation Integrated System), M-DRIS (Modified-DRIS), and CND (Compositional Nutrient Diagnosis) in diagnosing pineapple leaf nutrient levels with varying sample sizes. Empirical data from a subtractive experiment was used to simulate and constitute a new database considering that nutrient contents were normally distributed. For each sample size, data were generated per treatment and replicated 3,000 times. DRIS, M-DRIS, and CND indices were computed from the simulated data for each nutrient. The methods were subsequently evaluated based on four criteria: (i) the Diagnosis Concordance Frequency, which assesses the consistency of diagnoses across different methods for determining nutritional indices; (ii) the sensitivity, or True Positive Rate, which gauges a model's ability to accurately identify a specific nutritional status when it is present; (iii) the precision, or Positive Predictive Value, which indicates the proportion of correctly identified diagnoses for a particular nutritional status relative to the total number of diagnoses made for that status; and (iv) the accuracy, which measures the closeness of the model's results to the true value. As results, we found that N, P, and K nutrient indices differed significantly between DRIS, M-DRIS, and CND models and with sample size. The nutritional diagnosis methods were also discordant, except DRIS versus M-DRIS (mean agreement = 66%). Compared to DRIS, and M-DRIS models, CND appeared to be the most sensitive and accurate model (average accuracy of 27.86%) for nutrient deficiency and excess diagnosis. The models’ accuracy varies with the sample size, but it becomes almost unchangeable from a sample size of 330. For all sample sizes, the CND model was more accurate and efficient for N, P, and K nutrient status diagnosis, compared to DRIS and M-DRIS models.