Identification of Soil Properties Associated with the Incidence of Banana Wilt Using Supervised Methods

Abstract

Over the last few decades, a growing incidence of Banana Wilt (BW) has been detected in the banana-producing areas of the central zone of Venezuela. This disease is thought to be caused by a fungal−bacterial complex, coupled with the influence of specific soil properties. However, until now, there was no consensus on the soil characteristics associated with a high incidence of BW. The objective of this study was to identify the soil properties potentially associated with BW incidence, using supervised methods. The soil samples associated with banana plant lots in Venezuela, showing low (n = 29) and high (n = 49) incidence of BW, were collected during two consecutive years (2016 and 2017). On those soils, sixteen soil variables, including the percentage of sand, silt and clay, pH, electrical conductivity, organic matter, available contents of K, Na, Mg, Ca, Mn, Fe, Zn, Cu, S and P, were determined. The Wilcoxon test identified the occurrence of significant differences in the soil variables between the two groups of BW incidence. In addition, Orthogonal Least Squares Discriminant Analysis (OPLS-DA) and the Random Forest (RF) algorithm was applied to find soil variables capable of distinguishing banana lots showing high or low BW incidence. The OPLS-DA model showed a proper fitting of the data (R2Y: 0.61, p value < 0.01), and exhibited good predictive power (Q2: 0.50, p value < 0.01). The analysis of the Receiver Operating Characteristics (ROC) curves by RF revealed that the combination of Zn, Fe, Ca, K, Mn and Clay was able to accurately differentiate 84.1% of the banana lots with a sensitivity of 89.80% and a specificity of 72.40%. So far, this is the first study that identifies these six soil variables as possible new indicators associated with BW incidence in soils of lacustrine origin in Venezuela.

Keywords: calcium; clay; iron; machine learning; random forest; zinc.

Figure 1

Symptoms of Banana Wilt disease in study area. (a) The yellow margins on the leaves and the necrotic stripes surrounded by the yellow margins on the lower or older leaves; (b) Set of dead leaves hanging from the pseudostem of a plant affected with Banana Wilt disease.

Figure 2

Cumulative incidence (%) of Banana Wilt in the study area during 2016 (a) and 2017 (b) (n = 78; mean = 2.17 ± 1.40%; min = 0.11%; max = 8.47%; asymmetry = 1.78; kurtosis = 4.46; P₅₀ = 1.90%).

Figure 3

Heatmap generated from soil data of the banana lots with low (green) or high (purple) incidence of BW evaluated in year 2016 (s6) and year 2017 (s7), which represents increasing concentration values of the soil variables (blue to red color) for the study periods.

Figure 4

Box plots of levels of soil variables showing significant differences between low and high BW incidence based on the Wilcoxon’s test.

Figure 5

(a) OPLS-DA score plot of all soil variables, based separation of the incidence (low incidence of BW, n = 29; high incidence of BW, n = 49); (b) Loading plot weights of each variable selected from OPLS-DA; The color indicates the class in which the variable has the maximum level of expression; (c) internal validation of the corresponding OPLS-DA model by permutation analysis (n = 100); fraction of the variance of descriptor class response (Y) (R²Y) = 0.61 (blue bars), p value < 0.01; fraction of the variance predicted (cross-validated) (Q²) = 0.50 (red bars), p value < 0.01.

Figure 6

Classification of bananas lots according to the incidence of banana wilt (BW). (a) Receiver operating characteristic (ROC) curve after obtained by Random Forest as classification method. The values generated for the area under the curve (AUC) (0.91) along with the 95% confidence intervals (CI) (0.80–0.99) are given within the graph and accuracy: 84.10%; (b) Predicted class probabilities for each banana lot, allowing display of misclassified bananas lots (lots of high BW incidence are shown as black dots; lots of low BW are shown as white dots). Since a balanced subsampling approach is used for model training, the classification limit is always in the center (x = 0.5, the dotted line); (c) Confusion matrix showing the number of true positives (44/49), true negatives (21/29), false positives (8/29) and false negatives (5/49). Sensitivity and specificity are given in the regions highlighted in purple, being 89.80% and 72.40%, respectively.

Figure 7

Geographical location of the study area with banana lots (marked with yellow color boundaries).

Figure 8

General scheme of the data analysis procedures (sample size, n = 78; variable size, n = 16) using non supervised and supervised analysis.