Application of a Hidden Markov Model to remotely sensed snow cover measurements of a Himalayan basin

Abstract

Remote sensing is used to monitor snow cover in areas without sufficient in-situ observations such as the Himalayas. However, commonly used remotely sensed snow cover products such as those provided by the Moderate Resolution Imaging Spectroradiometer (MODIS) have reduced accuracy in mountainous areas due to the negative influence of atmospheric effects and terrain in these regions. This thesis investigates the ability of a Hidden Markov model to limit erroneous misclassifications and reduce unrealistic estimates of snow cover. A Hidden Markov model is able to utilise a series of input observations, MODIS snow cover products in this case, and use these to calculate the probability of them representing the ground state. These probabilities can then be used to model the most likely series of states, this effectively provides a dynamic filter than can mitigate the problems faced when remotely sensing snow in mountainous areas. This approach was evaluated by applying a Hidden Markov model to MODIS snow cover measurements of a sub-basin in Eastern Nepal. The Hidden Markov model resulted in 16% and 9% increases in agreement with corresponding evaluation data than when compared to the two equivalent MODIS snow cover products that provided input data. The results also highlighted the ability of this approach to reduce data noise. Comparisons of snow covered area over time showed improved consistency in measurements when comparing the model output to both the daily and 8-day composite MODIS snow cover products. The ability of a Hidden Markov model to employ a dynamic filter that can utilise entire sequences of observations could offer improved accuracy when compared to other time-series filtering methods. This is because it is able to utilise a much a larger amount of observations without compounding losses in accuracy or resulting in reductions in temporal resolution. This study shows the potential for a HMM approach to provide a robust and flexible method for processing ‘noisy’ data such as remotely sensed snow cover measurements. Show more