Exploring the capabilities of hyperspectral remote sensing for soil texture evaluation
Hyperspectral remote sensing provides a broad range of spectral bands with up-to-date, cost-effective, and spatially explicit resolution for smart agriculture and environmental management. In this study, we modeled and generated a composite map of soil texture variability using PRISMA images collect...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-12-01
|
| Series: | Ecological Informatics |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1574954125003450 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Hyperspectral remote sensing provides a broad range of spectral bands with up-to-date, cost-effective, and spatially explicit resolution for smart agriculture and environmental management. In this study, we modeled and generated a composite map of soil texture variability using PRISMA images collected from agricultural areas in Iran's dry and semi-arid regions. We explored various workflow combinations of environmental covariates, integrating hyperspectral data with topography (t), climate (c), soil characteristics (s), vegetation-land use (v), and land use to enhance predictions. The analysis clarified the significance of all factors influencing prediction performance. Additionally, we compared the performance of random forest (RF) algorithms with partial least squares regression (PLSR), multiple linear regression (MLR), support vector machine regression (SVR), decision trees (DTs), and multilayer perceptron (MLP) neural networks, addressing the effects of feature selection and irregular soil data on the modeling procedure. Our results revealed that an accurate method for predicting and mapping sand, clay, and silt content involved a combination of t + g + v + c + s covariates offered by the SCORPAN model and the RF model, achieving an R2 of 0.54, 0.25, and 0.33 for predictions of sand, clay, and silt, respectively. Using preprocessed hyperspectral data has demonstrated a commendable performance in predicting soil characteristics, with RF achieving R2 values of 0.40, 0.40, and 0.12 for the respective soil characteristics. The influence of spectral unmixing and vegetation effects on croplands significantly enhanced the prediction and mapping of soil texture using PRISMA data. Furthermore, integrating preprocessed hyperspectral data with DSM covariates resulted in improved predictive performance for sand and clay content, yielding R2 values of 0.64 and 0.51, respectively. These findings highlight the potential of environmental covariates and the modeling approach for enhancing the spatial coverage of real soil status to monitor diversity and refine soil mapping for more effective agricultural management practices. |
|---|---|
| ISSN: | 1574-9541 |