Estimation of Global Warming Response of Snowpack and Development of Analysis Technique for Precipitation Particle Observation

Snowpack and snowfall are strongly related to snow disasters, climate, and water cycle, and they would be affected by the global warming. Therefore, the former part of this doctoral dissertation revealed the response of snowpack to global warming along with its uncertainty including snowpack amount and quality in Hokkaido, a northern island of Japan. The estimation was conducted by using a combination of multiple general circulation models (GCMs), a single regional atmospheric model (RAM), and a one dimensional multi layered snowpack model. The forcing of the snowpack model was taken from dynamically downscaled (DDS) data from GCMs for the present 1990s climate and GCMs in a decade when the global mean temperature has increased by 2℃ from present conditions. The result showed that snow covered days would robustly decrease by about a month over Hokkaido. Additionally, the results show that the seasonal maximum snow amount would also decrease by 30-40% in southwestern and eastern Hokkaido, mainly due to a decrease in the total precipitation amount until a timing of seasonal maximum amount of snowpack. In contrast, the changes in snow amount was uncertain in northern and eastern Hokkaido mainly due to the uncertain changes in storm tracks, wintertime monsoon, and topological precipitation introduced by the different projections among GCMs. The decrease in snow amount would be larger in the beginning of ablation period, corresponding to a timing of wetting snowpack from bottom to top, and so the peak of snowpack amount would be shifted earlier and the thickness ratio of melt forms would increase as a result. The thickness ratio of hoar category would decrease except at high altitude region. In the latter part of this doctoral dissertation, a new method to estimate an optimal mixed joint probability density function (PDF) from precipitation particle size and velocity distribution (PSVD) data was developed looking ahead to an advanced forcing data to snowpack model in the future The new method developed in this study searches a locally maximum log likelihood within a realistic parameter range using the expectation maximization algorithm assuming that the particle velocity followed the normal distribution and the particle size followed the Gamma distribution. The performance was evaluated by estimating PDF from a PSVD data set randomly sampled from a population prepared in advance and by comparing the estimated PDF with the population. The results showed that the new method successfully estimated the population. The results showed that the new method successfully estimated the populations even if the populations was constructed with multiple type of precipitation particles such as graupel and rimed aggregates, not, not ever estimated previously. Furthermore, the method was helpful to discard frequently-observed erroneous data with unrealistically large fall-velocity. The new method could also distinguish liquid and solid particles.