Mountain agriculture extraction from time-series MODIS NDVI using dynamic time warping technique

ABSTRACT

The study attempts to extract Mountain Agriculture using an optimized Dynamic Time Warping (DTW) algorithm having endpoint constraints. The DTW was applied over a time-series annual stack of Normalized Differential Vegetation Index (NDVI) using a set of reference time series profiles for three agriculture classes (i.e. double cropping, single cropping, and horticulture) and the pixel-wise similarity is examined to identify the agriculture classes. In addition, Euclidean Distance (ED) was used to compare DTW-based result. The detection accuracy of each class was assessed using Google Earth-based agriculture sample, and the spatial agreement of resultant map was assessed with high-resolution reference data using Pareto boundary technique. The sample based accuracy evaluation reveals that DTW algorithm performed better for double and single cropping agriculture detection in compared to the horticulture. Overall, DTW-based agriculture map (0.81 ± 0.01) yielded higher overall accuracy in comparison with ED-based agriculture map (0.75 ± 0.01). The Pareto boundary-based spatial agreement analysis using high-resolution reference data also shows the dominant performance of DTW based agriculture map than an ED-based map. DTW performed better than ED, in terms of optimal distance (OD), in ten out of eleven districts. However, reliable spatial matching (OD less than 0.23) between DTW-based map and reference agriculture map was observed in lower elevation region, especially in Hamirpur (OD = 0.06), Bilaspur (OD = 0.09), Shimla (OD = 0.19) and Una (OD = 0.20) district.

Acknowledgments

Authors are thankful to the anonymous reviewers for their insightful comments to improve the original manuscript. Authors also would like to thank NASA MODIS team and NRSC, ISRO, India for providing all the required data freely.

Disclosure statement

No potential conflict of interest was reported by the authors.