1. Macro-Nutrient Prediction of Paddy Field Soil Using Artificial Neural Network and NIR Spectroscopy.
- Author
-
Firdaus, Jonni, Ahmad, Usman, Budiastra, I Wayan, and Made Subrata, I Dewa
- Subjects
PADDY fields ,SOIL fertility ,ARTIFICIAL neural networks ,NEAR infrared spectroscopy ,TRACE element content of soils - Abstract
Understanding soil fertility, influenced by macronutrients such as nitrogen, phosphorus, and potassium, is essential for the implementation of adaptive agriculture based on various soil conditions. Near-infrared spectroscopy technology provides non-destructive, rapid soil property measurements without chemicals, and is applicable both in the field and in the laboratory. However, the wide NIR spectrum range and neural network complexities can hinder Artificial Neural Network (ANN) training and inference, leading to time and resource inefficiency, particularly without sophisticated computing devices. This study examined data reduction methods to enhance ANN performance in predicting soil macronutrients using NIR spectra. Multiple Linear Regression (MLR) and Principal Component Analysis (PCA) were applied to select wavelengths from the 1000–2500 nm for ANN input, comparing their performance. Approximately 237 NIR reflectance data points from paddy soils were transformed into absorbance data. MLR used forward selection to identify wavelengths with correlations higher than 0.9, whereas PCA selected wavelengths corresponding to the loading factor peaks for each principal component. These selected wavelengths served as inputs for the ANN model. The ANN performance was assessed using the correlation and determination coefficients, RMSE, RPD, and model consistency. For nitrogen, the PCA+ANN model with reflectance spectra performed better (RPD 2.4-4.8) than the MLR+ANN model (RPD 2.2-2.6) using fewer wavelengths (5-9 for PCA+ANN vs. 9-12 for MLR+ANN). For phosphorus estimation, the PCA+ANN model also excelled (RPD 2.3-7.0 vs. 2.3-2.4) with fewer wavelengths (4-7 vs. 7). The PCA+ANN model showed superior performance (RPD 4.3-9.5 vs. 4.2-4.4), using the same number of wavelengths (4-8 vs. 4-6). [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF