Finding Threshold for NDVI to Classify Green Area: Case Study in the Central Thailand

Thailand and other countries worldwide are trying to increase green areas to fight against climate and deforestation issues and improve air quality. To observe a green area on a large-scale using satellite images are eligible. Free-of-charge satellite images such as Landsat 8 offer useful information to produce various outputs. Survey for a green area Normalized Different Vegetation Index (NDVI) is beneficial since satellite images are in charge. NDVI is an indicator that can analyze remote sensing measurements and Geographic Information System (GIS), assessing whether the target being observed contains live green vegetation. Although NDVI is a good indicator for observing a green area, NDVI still cannot classify a green area without thresholds. To find a proper threshold for NDVI classification, we have been through the three land-use types: urban, agricultural, and forest, with multi-temporal satellite images. We develop the software tool using Python code and remote sensing data for classification and accuracy assessment. Both experiments aim to observe a proper threshold that satisfies high prediction accuracy. Finding thresholds for NDVI is required ground truth which is trustable information of a green area and NDVI image. To provide ground truth, we used digitizing method to obtain the information. The type of area experiment uses a different type of area as a variant, and for a temporal area, we use time as a variant. A temporal area experiment observes the same is with different timing to compare the results. As a result, we found that a threshold around 0.325 to 0.367 is suited to observing a green area.