Publications

Praveen, A; Jeganathan, C; Mondal, S (2023). Mapping Annual Cropping Pattern from Time-Series MODIS EVI Using Parameter-Tuned Random Forest Classifier. JOURNAL OF THE INDIAN SOCIETY OF REMOTE SENSING, 51(5), 983-1000.

Abstract
Monitoring agriculture growth at different seasons (i.e., Rabi-winter crop, Zaid-summer crop, Kharif-monsoon crop) is an important requirement to understand annual cropping pattern dynamics for food security-related policy and strategy formulation. Cropping pattern is not uniform across a region and hence satellite observation of different time-period over a year is inevitable in such cases, but the associated big-data processing and accuracy requirements makes the problem more pertinent to study. Non-parametric machine learning (ML) algorithms have a higher ability to deal with such information extraction problem related to high-dimensional time-series satellite data, especially to infer multi-growth agriculture patterns. The current study aims to explore one of the popular ML algorithm, i.e., Random Forest (RF), along with two conventional supervised algorithms [i.e., Maximum Likelihood Classifier (MLC) and Spectral Angle Mapper (SAM)] for annual cropping pattern mapping in Bihar State of India using MODIS time-series Enhanced Vegetation Index (EVI). The RF algorithm was tuned to test its sensitivity with diverse combinations of model parameters, training sample data, and input variables. Tuning options for training samples and input variables have provided a possibility to understand the optimality and infer important temporal periods to detect cropping patterns with less data. In addition, the analysis also revealed that though a high degree of collinearity (i.e., r >= 0.9) exists in time-series EVI data due to the presence of multiple growing and decaying phases of crop, it does not negatively affect the model performance. The blind random sample-based assessment reveals that RF excelled in accuracy (81.47%) over MLC (67.47%) and SAM (52.53%).

DOI:
10.1007/s12524-023-01676-2

ISSN:
0974-3006