Influence of sample size and machine learning algorithms on digital soil nutrient mapping accuracy.

Area of Science:

• Agricultural Science
• Environmental Science
• Data Science

Background:

• Accurate soil nutrient mapping is essential for sustainable agriculture and environmental management.
• Machine learning (ML) offers powerful tools for predicting soil properties, but performance is influenced by data volume.

Purpose of the Study:

• To evaluate the impact of varying sample sizes on the prediction performance of five ML algorithms for 14 soil properties.
• To compare the effectiveness of Multi-layer Perceptron (MLP), Random Forest (RF), Extra Trees Regressor (ETR), CatBoost, and Gradient Boost (GB) in soil nutrient prediction.

Main Methods:

• Trained five ML algorithms on datasets with sample sizes ranging from 25 to 800 for 14 soil properties.
• Utilized 574 environmental variables from digital terrain and Sentinel-2 imagery as predictors.
• Evaluated prediction accuracy using Lin's concordance correlation coefficient (CCC) and Root Mean Squared Error (RMSE).

Main Results:

• Prediction accuracy generally increased with sample size for all ML algorithms, with diminishing returns beyond a certain point.
• RF, ETR, CatBoost, and GB outperformed MLP, showing better prediction performance across various soil properties.
• Micronutrients showed a notable improvement in prediction accuracy with increased sample sizes.

Conclusions:

• An optimal sample size can be determined to achieve accurate soil nutrient predictions efficiently.
• Selecting appropriate ML algorithms alongside optimal sample sizes is key for maximizing prediction accuracy in soil mapping.
• The study provides guidance for resource allocation in soil survey and mapping projects.