Soil organic matter estimation by using Landsat-8 pansharpened image and machine learning
| Autor: | Abdelmejid Rahimi, El Mostafa Ettachfini, Abdelkrim Bouasria, Khalid Ibno Namr |
|---|---|
| Rok vydání: | 2020 |
| Předmět: |
010504 meteorology & atmospheric sciences
Artificial neural network Soil test Soil organic matter Multispectral image Decision tree 04 agricultural and veterinary sciences 01 natural sciences Soil quality 040103 agronomy & agriculture 0401 agriculture forestry and fisheries Environmental science Image resolution Predictive modelling 0105 earth and related environmental sciences Remote sensing |
| Zdroj: | 2020 Fourth International Conference On Intelligent Computing in Data Sciences (ICDS). |
| Popis: | Considering the significant position of soil organic matter (SOM) in soil quality and maintenance, and its role in the functioning of soil physicochemical and biological processes, it is essential to monitor frequently the SOM status and its dynamics. It is a time-consuming and expensive task if we depend exclusively on chemical analysis, particularly in a semi-arid irrigated zone and with intensive agricultural activities and a very fragmented landscape. It is the Sidi Bennour region, which is situated in Doukkala Irrigated Perimeter in Morocco. Data from satellites could be a good alternative to conventional methods and fill this void with low costs. There has been a great deal of interest in satellite image prediction models, especially with free and abundant availability of satellite data. This work intends to predict SOM using Decision Trees (DT), k-Nearest Neighbors (k-NN), and Artificial Neural Networks (ANN). The soil samples (369 points) were collected at 0-30 cm of depth and the laboratory analysis was carried out. A multispectral Landsat-8 image was acquired and then calibrated. An image pansharpening processing was applied to produce a PAN image with 15m of resolution from 30m image resolution (MS). The obtained results indicate that the ANN model outperformed the other predictive models for both images (MS and PAN) with R2= 0.6553 and R2=0.6985, respectively. The statistical RMSE of predictive models was 0.2153 and 0.2014, and MAE was 0.1682 and 0.1573 for both images, MS and PAN respectively. For this predictive model, the image pansharpening could increase the prediction accuracy of R2 by 4.32%and reduce the RMSE by 1.39%. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|