The distribution of Sal forests in Jharkhand

We used the information to train satellite images and built a model to detect Sal forests all over Jharkhand. The model is based on a simple principle – the frequencies of light reflected by the leaves of Sal trees is different and distinct from the light from trees of other species. This is known as the spectral signature of a species, and it includes the values of light reflected across different frequency bands. With advances in satellite sensor technology and the frequency of data availability, it is now possible to build accurate spectral signatures using images from multiple seasons. The insight here is that easier to detect Sal trees during its flowering season, when no other tree species has a similar light frequency reflectance.

Geeky stuff

We used images from the Sentinel II system provided by the European Space Agency. These images are high-resolution (~10m wide pixels) and are available at a frequency of once every five days. This allows us to use images during specific seasonal cycles to increase our accuracy of detection. For building our model to detect Sal trees in the landscape, we used a combination of frequency bands in the Blue, Green, Red, Infrared, and Short-Wave Infrared parts of the spectrum. We divided the field data on the location of Sal trees collected in December into two sets – training and prediction. We used the training dataset to build the model using machine learning tools and then tested the accuracy of our results on the prediction dataset. We were able to achieve ~90% accuracy in our model.

The map of the distribution of Sal is the first step in building an inventory of Seasonal Forest Products in Jharkhand. We will now use the model to estimate the number of Sal trees located within the boundaries of each village. Using available data and research on average production of seeds from Sal trees, we will extrapolate the potential value based on recent trends in market prices for raw and processed Sal seeds.

Over the next two months, we will extend the distribution modeling and estimation of market value to eight other Seasonal Forest Products (Amla, Hirda, Beheda, Mahua, Karanj, Chironji, Kusum, and Kadamb). At the same time, we will work with local communities and their representatives to validate the outputs of our models. This information from the ground will help to further refine the models and improve our accuracy. This exercise will allow us to construct the first scientific inventory of forest-based industrial raw materials in India. Such an inventory is necessary for attracting industry partners to collaborate with communities for procurement.