Classification of Precipitation Intensities from Remote Sensing Data Based on Artificial Intelligence Using RF Multi-learning
A new strategy based on random forest (RF) classifier multi-learning is elaborated for the rainfall intensities classification from remote sensing data provided by Meteosat Second Generation (MSG) satellite. It consists in carrying out a series of classifications to isolate a class at each treatment...
Gespeichert in:
Veröffentlicht in: | Journal of the Indian Society of Remote Sensing 2023-04, Vol.51 (4), p.733-742 |
---|---|
Hauptverfasser: | , , , , |
Format: | Artikel |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Zusammenfassung: | A new strategy based on random forest (RF) classifier multi-learning is elaborated for the rainfall intensities classification from remote sensing data provided by Meteosat Second Generation (MSG) satellite. It consists in carrying out a series of classifications to isolate a class at each treatment (learning/classification). The idea developed here is to allow the RF classifiers to better separate the classes by proceeding one by one. To do this, the training of a set of RF models independently of each other is carried out. The first learning is performed to isolate the first class. Pixels belonging to the first class are excluded from the database during the second training. This second learning is performed on the remaining pixels to isolate the second class. The pixels corresponding to the second class are also excluded from the database during the third training. The process continues until the separation between the two last classes. Thus, for n classes, we perform
n
−1 learning. The learning and validation of RF models (hereafter called Multi-RF) are realized by comparing input data from the MSG satellite with output data from Setif meteorological radar. To evaluate the Multi-RF, the results obtained are compared to those obtained by applying the standard RF (Uni-RF) and to One versus All RF (OvA-RF). The comparison results indicate superior performance for the proposed scheme. Indeed, the evaluation statistical parameters obtained for Multi-RF are better than those obtained by Uni-RF and OvA-RF. |
---|---|
ISSN: | 0255-660X 0974-3006 |
DOI: | 10.1007/s12524-023-01665-5 |