As mentioned in my introductory blog post, I am mapping land use and linear intrusions of the Nethravathi river basin in India. There are no reliable land use maps of the region and hence the map I create will serve as a baseline my geospatial analysis of quantifying forest loss with each intrusion, and fragmentation analysis. The major challenge I had was to distinguish forests from agricultural plantations of rubber, areca, and coffee. This distinction is important so that I don’t overestimate or underestimate the forest loss due to intrusions.

While working on the initial analysis in Madison, I found that using visible bands, indices, or even multi-date imagery was not of much help to get the distinction right. I found that the red edge bands in Sentinel 2 imagery helped in distinguishing the two classes after several trials of running classification for a smaller region. I remember during the summer term, I was struggling to distinguish certain land-cover types while mapping Guatemala City for a lab assignment. No matter how good my training data was, the classifier was not able to bring out the distinction. It so happened that I was giving the classifier too much data to work with while I had to input only certain bands for it to work.

Now that this was sorted, the next task was to get to the field to collect as many data points as I could using a handheld GPS device to create training data as well as verify my test data set. I got to the field as soon as I landed in India. I was on the field after almost a gap of 5 years. What seemed like a challenge at first with my jet lag and the hot and humid weather, gradually felt easier with my parents around who decided to accompany me in the field and drive me around.  

I spent a week on the field covering all the ambiguous test points. This process helped me understand the landscape better. I now have a sense of what lies where and can visualize the fragmentation caused by the linear intrusions in the region. At the end of every field day, I would load the points and tracks to Google Earth Pro, look at each of the points and get my eyes acclimatized to how a feature looked in satellite imagery and connect it back to what I saw on the field and make notes of it. I can now tell a coffee plantation from a forested patch by visually inspecting imagery which I was not able to earlier!

I am currently in my hometown Mysore, working on some of the less interesting things like processing field data, downloading required data sets for my analyses, editing training data and going through multiple runs of image classification. I am also getting to do some new and exciting stuff which I have not done before. I had to visit the Survey of India office to get my hands on toposheets of the region which I will be using as ancillary data. A new road alignment proposed through a forest patch was available on a toposheet. I need to have this road alignment in a usable format for my analysis. So, I spent some time trying to figure out how to geo-reference toposheets and digitize the necessary details I need.

I have also been working on an interactive ESRI story map which will give me an opportunity to put all my analyses in one place so that Wildlife First could use it to communicate the results with stakeholders. So far, I have spent time laying out the basic structure and the interactive visualization which will be integrated into it and it’s been a great learning experience. I feel like all the skills and tools I have learned during the program have come together to help me design and implement this project. I still have a few things to put together before I can start working on the core analysis and I am looking forward to it!