EASTERN INDIA ECO-FORECASTING III
A MULTI-SENSOR APPROACH TO ENHANCE THE PREDICTION OF MANGROVE BIOPHYSICAL CHARACTERISTICS IN ODISHA, INDIA
Across the globe, mangroves play a major role in coastal ecosystem processes mitigating erosion and serving as barriers against storm surges. India holds approximately 5% of the world's mangroves, over half of which are along its east coast. Situated in the state of Odisha, Chilika Lagoon and Bhitarkanika Wildlife Sanctuary sustain mangrove sites of local importance in need of effective management. This study demonstrated the use of Terra, Landsat, and Sentinel-1 satellite data for spatio-temporal monitoring of mangrove health for both sites. Specifically I will highlight the historical analysis of land cover maps was produced using Landsat 5 and 8 data to determine decadal changes in mangrove area estimates between 1995 and 2017. This analysis was used to predict land use-land cover change or fragmentation of Bhitarkanika mangroves. Based on IPCC data availability, the soft prediction map for 2050 showed the probability of mangrove risk to disturbance in the eastern part of Bhitarkanika. This study revealed the advantages of using a multi-sensor approach to monitor mangrove health and inform monitoring protocols.
For more information about the project see this link.
Credit: This project was a group effort and all information on this page and the link was produced by Abhishek Kumar (Project Lead), Isabel Miranda, Maria Luisa Escobar Pardo, Taufiq Rashid, Shanti Shrestha, NASA DEVELOP National Program.
The land-cover maps were made using Google Earth Engine Explorer with a random forest algorithm. These are the classification results with an overall accuracy of 84% for 1995, 82% for 2004 and 86% for 2017. An accuracy assessment was calculated in TerrSet to create an Error Matrix indicating the overall accuracy. Visually we can note significant fragmentation of dense mangrove and increases in agriculture and open mangrove land cover.
The total amount of loss of dense mangrove is 9.28 square km from 1995 to 2004 and the total amount loss from 2004 to 2017 is 21.44 square km, indicating more loss occurred from 2004 to 2017. Focusing on a particular part of our study area we can note than in 1995 areas of open mangrove (red) in 2004 were replaced by dense mangrove. In 2017 we can note that locations of dense mangrove were again lost becoming open mangrove. In addition 70% of the total dense mangrove that was lost from 2004 and 2017 changed to open mangrove. There was also a 24.4% gain in dense mangrove from 1995 to 2004.
The driver variables we used were different for each transition and were selected based on the strength for the variable to predict change for each transition. For example, pixels that are transitioning from dense mangrove to open mangrove are most influenced by distance to roads, towns channels, previous disturbance and elevation.
Multilayered Perceptron (MLP) Neural Network was used to produce a transition potential image that describes the probability that a transition will occur in the landscape and is used to predict future land-cover change for 2050. Finally using the Change Prediction Tab in the Land Change Molder (LCM) in TerrSet was used to specify the date 2050 and use the Markov Chain Analysis to model our transitions in land cover.
One of the two models of change is a soft prediction map that indicates a scale of vulnerability that shows the risk of mangroves to disturbance in the future and the area’s potential to change to another land cover. Red/orange locations indicate medium to high vulnerability and locations of yellow and blue indicate lower vulnerability.