Investigation Of Glacio-hydrological Processes In The Tianshan Urumqi No.1 Glacier Region Using Explainable Deep Learning

Glaciers are indicators of climate change.The glaciers mass balance is closely related to the regional runoff process,and is an important process of the water cycle and energy cycle in cold regions.Quantitative simulation and prediction of it is the key to understanding the hydrological process in cold regions.Tianshan is known as the "Central Asian Water Tower" and is an important water source in Central Asia.The glacier No.1 in the headwater of Urumqi of the Tianshan Mountains is an important supply source of water resources in the arid region of Northwest my country.The changes of glacier No.1 can reflect the climatic conditions of the northwest arid region.Because of its importance and unique geographical location,it is also the only Chinese glacier among the 10 key glacier observation stations monitored by the World Glacier Monitoring Service(WGMS).It is also the only Chinese glacier with a long time series of mass balance observations.Therefore,this paper selects the Urumqi glacier No.1 as the research object to study the hydrological process of the glacier in the arid area.Glacio-hydrological processes modeling is a complex non-linear time series problem with many factors and strong uncertainty.In order to study the meteorological factors and their mechanism affecting the hydrological process in glacier No.1 region,this study uses the deep learning Long Short-Term Memory(LSTM)method for modeling,which has shown a good deal of promise dealing with time series problems by considering long-term dependencies and multiple outputs.In order to analyze the uncertainties of the established LSTM hydrological model,an interpretable model was constructed by coupling Shapley Additive Explanations(SHAP)with LSTM to study the mechanism of hydrometeorological factors influencing glacier mass balance and local runoff.The specific is as follows: Firstly,using the meteorological data build the LSTM prediction model of glacier mass balance.Then,according to the meteorological data and mass balance of different altitudes to establish LSTM model of runoff.On this basis,the SHAP model and LSTM prediction model are introduced to get the explanations model of the glaciohydrological processes.Therefore,it can explain the transmission process and influence mechanism of glacier hydrological information from meteorological factors to glacier mass balance and runoff in alpine glacier area.The main conclusions are as follows:(1)Based on meteorological data,an LSTM prediction model for glacier mass balance at different elevations was established.Using the data from 1988 to 2007 as the training set and the data from 2008 to 2015 as the prediction set,build an LSTM model with one input layer,three hidden layers and one output layer.The mass balance of the accumulation area,the ablation area and the entire glacier was trained in turn to establish the relationship between meteorological factors and material balance.The results show that the model is ideal for the prediction of glacier mass balance.The NSEs of the predicted values in the model for the entire glacier,the ablation zone and the accumulation zone are 0.82,0.79 and 0.59,respectively.(2)Based on meteorological data and glacier mass balance data at different elevations,a simulation model of river runoff of headwaters of the Urumqi River Glacier No.1 was established.Using the data from 1988 to 2007 as the training set and the data from 2008 to 2015 as the prediction set,the river runoff data of the Urumqi glacier No.1 Hydrological Station was studied.Build an LSTM model for river runoff based on meteorological factors and glacier mass balance.The results show that the prediction effect of the runoff model is related to the elevations of mass balance,and the prediction results are best when the mass balance data of the entire glacier is used,followed by the ablation zone,and finally the accumulation zone,with NSEs of 0.80,0.77 and0.73,respectively.(3)Based on the SHAP method,interpretable interpretation of the "black box" modeling results of LSTM.The contribution of characteristic factors affecting glacier mass balance and headwater river runoff was quantified from three perspectives: overall Feature Importance,timestep decomposition and extreme values.The results show that temperature is the most important meteorological factor affecting hydrological processes in cold regions,and the characteristic contribution order that affects the mass balance of glaciers is air temperature,air pressure,precipitation,relative humidity,sunshine hours and wind speed,respectively.The order of contribution of meteorological factors and mass balance to the forecast of headwaters runoff of the Urumqi River Glacier No.1 is air temperature,precipitation,relative humidity,sunshine duration,mass balance of glacier,air pressure and wind speed,respectively.On the monthly scale,the hydroclimatic factors within five months have the obvious effect on the glacio-hydrological processes.Runoff is mainly recharged by precipitation within a time step of 1-2 months,and it is mainly affected by the mass balance of glaciers during 3-5 months.(4)Based on the SHAP value analysis of the extreme event,this study quantifies the contribution of input characteristics to mass balance and runoff.By calculating the SHAP values of the maximum positive balance,maximum negative balance,maximum negative balance,minimum negative balance and maximum and minimum runoff of Urumqi Glacier No.1,the key values of meteorological factors acting on the hydrological process are obtained.The results show that glaciers have a "water tower" effect in arid regions,that is,glaciers store water in the form of ice and snow in the wet and cold seasons,and release water in the form of glacier meltwater in the dry and warm seasons.When the temperature is above 0°C,the relative humidity is about 64%and the air pressure is greater than about 663 hpa,it is not conducive to glacier accumulation.The effect of precipitation on the mass balance of glaciers needs to be considered on the basis of temperature.The increase in temperature causes the ablation of glacial mass.Only when the total amount of precipitation exceeds the ablation amount can the increased precipitation promote the increase of glacial mass balance.