The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume XLII-3/W12-2020
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-3/W12-2020, 31–35, 2020
https://doi.org/10.5194/isprs-archives-XLII-3-W12-2020-31-2020
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-3/W12-2020, 31–35, 2020
https://doi.org/10.5194/isprs-archives-XLII-3-W12-2020-31-2020

  04 Nov 2020

04 Nov 2020

FOREST DISTURBANCE DETECTION BY LANDSAT-BASED NDVI TIME SERIES FOR AYUQUILA RIVER BASIN, JALISCO, MEXICO

Y. Gao, A. Quevedo, and J. Loya Y. Gao et al.
  • Centro de Investigaciones en Geografía Ambiental, Universidad Nacional Autónoma de México, 58190 Morelia, México

Keywords: Deforestation, Forest degradation, Magnitude of change

Abstract. Time series data have been applied for forest disturbance detection. The validation of detected changes is challenging partially because the validation data are often not readily available. Unlike multi-temporal change analysis, time series analysis not only detects areas of change but also reports time of change. Both spatial and temporal accuracy are therefore important for the accuracy assessment. Ayuquila River Basin (ARB) is one of the early action areas in Mexico for the implementation of REDD+ initiatives under UNFCCC. In ARB, shifting cultivation and cattle grazing often take place, resulting in degraded forestland. Sub-annual forest disturbance detection and estimation contribution to the improved local forest management and REDD+ implementation. Landsat-based NDVI time series data covering 1999–2018 were analysed using linear regression and the breakpoints of change and the magnitude of change were detected. Breakpoints with magnitude of change ranging from (−0.05) to (−0.2) were verified during a field campaign in October 2018. Here the magnitude of change is related with NDVI differences. Areas with magnitude of change higher than (−0.2) were identified as false changes. Verification data were generated by visually interpreting time series Landsat images of 2016–2018. In this way, areas with forest loss were identified. By stratified random sampling, 683 points were applied for the verification including 511 points of forests and 172 points of forest loss. It yields 75.84% for the overall accuracy of the change detection; for the detected forest loss as a category, the user’s accuracy is 88.89% and the producer’s accuracy is 0.46%. A possible reason for the very low producer’s accuracy is that the selected magnitude value (−0.2) is too low and some of the detected changes were filtered out.