Moderate Resolution Imaging Spectroradiometer Fractional Snow Cover Research Articles

A Novel Machine Learning–Based Gap-Filling of Fine-Resolution Remotely Sensed Snow Cover Fraction Data by Combining Downscaling and Regression

Abstract Satellite-based remotely sensed observations of snow cover fraction (SCF) can have data gaps in spatially distributed coverage from sensor and orbital limitations. We mitigate these limitations in the example fine-resolution Moderate Resolution Imaging Spectroradiometer (MODIS) data by gap-filling using auxiliary 1-km datasets that either aid in downscaling from coarser-resolution (5 km) MODIS SCF wherever not fully covered by clouds, or else by themselves via regression wherever fully cloud covered. This study’s prototype predicts a 1-km version of the 500-m MOD10A1 SCF target. Due to noncollocatedness of spatial gaps even across input and auxiliary datasets, we consider a recent gap-agnostic advancement of partial convolution in computer vision for both training and predictive gap-filling. Partial convolution accommodates spatially consistent gaps across the input images, effectively implementing a two-dimensional masking. To overcome reduced usable data from noncollocated spatial gaps across inputs, we innovate a fully generalized three-dimensional masking in this partial convolution. This enables a valid output value at a pixel even if only a single valid input variable and its value exist in the neighborhood covered by the convolutional filter zone centered around that pixel. Thus, our gap-agnostic technique can use significantly more examples for training (∼67%) and prediction (∼100%), instead of only less than 10% for the previous partial convolution. We train an example simple three-layer legacy super-resolution convolutional neural network (SRCNN) to obtain downscaling and regression component performances that are better than baseline values of either climatology or MOD10C1 SCF as relevant. Our generalized partial convolution can enable multiple Earth science applications like downscaling, regression, classification, and segmentation that were hindered by data gaps.

Assimilation of MODIS Snow Cover Fraction Observations into the NASA Catchment Land Surface Model.

The NASA Catchment land surface model (CLSM) is the land model component used for the Modern-Era Retrospective Analysis for Research and Applications (MERRA). Here, the CLSM versions of MERRA and MERRA-Land are evaluated using snow cover fraction (SCF) observations from the Moderate Resolution Imaging Spectroradiometer (MODIS). Moreover, a computationally-efficient empirical scheme is designed to improve CLSM estimates of SCF, snow depth, and snow water equivalent (SWE) through the assimilation of MODIS SCF observations. Results show that data assimilation (DA) improved SCF estimates compared to the open-loop model without assimilation (OL), especially in areas with ephemeral snow cover and mountainous regions. A comparison of the SCF estimates from DA against snow cover estimates from the NOAA Interactive Multisensor Snow and Ice Mapping System showed an improvement in the probability of detection of up to 28% and a reduction in false alarms by up to 6% (relative to OL). A comparison of the model snow depth estimates against Canadian Meteorological Centre analyses showed that DA successfully improved the model seasonal bias from −0.017 m for OL to −0.007 m for DA, although there was no significant change in root-mean-square differences (RMSD) (0.095 m for OL, 0.093 m for DA). The time-average of the spatial correlation coefficient also improved from 0.61 for OL to 0.63 for DA. A comparison against in situ SWE measurements also showed improvements from assimilation. The correlation increased from 0.44 for OL to 0.49 for DA, the bias improved from −0.111 m for OL to −0.100 m for DA, and the RMSD decreased from 0.186 m for OL to 0.180 m for DA.

Improvement of the Snow Depth in the Common Land Model by Coupling a Two-Dimensional Deterministic Ensemble Model with a Variational Hybrid Snow Cover Fraction Data Assimilation Scheme and a New Observation Operator

Abstract During snow cover fraction (SCF) data assimilation (DA), the simplified observation operator and presence of cloud cover cause large errors in the assimilation results. To reduce these errors, a new snow cover depletion curve (SDC), known as an observation operator in the DA system, is statistically fitted to in situ snow depth (SD) observations and Moderate Resolution Imaging Spectroradiometer (MODIS) SCF data from January 2004 to October 2008. Using this new SDC, a two-dimensional deterministic ensemble–variational hybrid DA (2DEnVar) method of integrating the deterministic ensemble Kalman filter (DEnKF) and a two-dimensional variational DA (2DVar) is proposed. The proposed 2DEnVar is then used to assimilate the MODIS SCF into the Common Land Model (CoLM) at five sites in the Altay region of China for data from November 2008 to March 2009. The analysis performance of the 2DEnVar is compared with that of the DEnKF. The results show that the 2DEnVar outperforms the DEnKF as it effectively reduces the bias and root-mean-square error during the snow accumulation and ablation periods at all sites except for the Qinghe site. In addition, the 2DEnVar, with more assimilated MODIS SCF observations, produces more innovations (observation minus forecast) than the DEnKF, with only one assimilated MODIS SCF observation. The problems of cloud cover and overestimation are addressed by the 2DEnVar.

Snow data assimilation‐constrained land initialization improves seasonal temperature prediction

AbstractWe present the first systematic study to quantify the impact of land initialization on seasonal temperature prediction in the Northern Hemisphere, emphasizing the role of land snow data assimilation (DA). Three suites of ensemble seasonal integrations are conducted for coupled land‐atmosphere runs. The land component is initialized using datasets from (1) no DA, (2) assimilating Moderate Resolution Imaging Spectroradiometer (MODIS) snow cover fraction (SCF), and (3) assimilating both MODIS SCF and Gravity Recovery and Climate Experiment (GRACE) terrestrial water storage. Results show that snow DA improves temperature predictions especially in the Tibetan Plateau (by 5–20%) and high latitudes. Improvements at low latitudes are seen immediately and last up to 60 days, whereas improvements at high latitudes only appear later in transitional seasons. At high latitudes, assimilating GRACE data results in marked and prolonged improvements (by ~25%) due to large initial snow mass changes. This study has great implications for future land DA and seasonal climate prediction studies.

Assimilating MODIS-based albedo and snow cover fraction into the Common Land Model to improve snow depth simulation with direct insertion and deterministic ensemble Kalman filter methods

This study assesses the analysis performance of assimilating the Moderate Resolution Imaging Spectroradiometer (MODIS)-based albedo and snow cover fraction (SCF) separately or jointly into the physically based Common Land Model (CoLM). A direct insertion method (DI) is proposed to assimilate the black and white-sky albedos into the CoLM. The MODIS-based albedo is calculated with the MODIS bidirectional reflectance distribution function (BRDF) model parameters product (MCD43B1) and the solar zenith angle as estimated in the CoLM for each time step. Meanwhile, the MODIS SCF (MOD10A1) is assimilated into the CoLM using the deterministic ensemble Kalman filter (DEnKF) method. A new DEnKF-albedo assimilation scheme for integrating the DI and DEnKF assimilation schemes is proposed. Our assimilation results are validated against in situ snow depth observations from November 2008 to March 2009 at five sites in the Altay region of China. The experimental results show that all three data assimilation schemes can improve snow depth simulations. But overall, the DEnKF-albedo assimilation shows the best analysis performance as it significantly reduces the bias and root-mean-square error (RMSE) during the snow accumulation and ablation periods at all sites except for the Fuyun site. The SCF assimilation via DEnKF produces better results than the albedo assimilation via DI, implying that the albedo assimilation that indirectly updates the snow depth state variable is less efficient than the direct SCF assimilation. For the Fuyun site, the DEnKF-albedo scheme tends to overestimate the snow depth accumulation with the maximum bias and RMSE values because of the large positive innovation (observation minus forecast).

Assimilation of MODIS snow cover through the Data Assimilation Research Testbed and the Community Land Model version 4

To improve snowpack estimates in Community Land Model version 4 (CLM4), the Moderate Resolution Imaging Spectroradiometer (MODIS) snow cover fraction (SCF) was assimilated into the Community Land Model version 4 (CLM4) via the Data Assimilation Research Testbed (DART). The interface between CLM4 and DART is a flexible, extensible approach to land surface data assimilation. This data assimilation system has a large ensemble (80-member) atmospheric forcing that facilitates ensemble-based land data assimilation. We use 40 randomly chosen forcing members to drive 40 CLM members as a compromise between computational cost and the data assimilation performance. The localization distance, a parameter in DART, was tuned to optimize the data assimilation performance at the global scale. Snow water equivalent (SWE) and snow depth are adjusted via the ensemble adjustment Kalman filter, particularly in regions with large SCF variability. The root-mean-square error of the forecast SCF against MODIS SCF is largely reduced. In DJF (December-January-February), the discrepancy between MODIS and CLM4 is broadly ameliorated in the lower-middle latitudes (23°–45°N). Only minimal modifications are made in the higher-middle (45°–66°N) and high latitudes, part of which is due to the agreement between model and observation when snow cover is nearly 100%. In some regions it also reveals that CLM4-modeled snow cover lacks heterogeneous features compared to MODIS. In MAM (March-April-May), adjustments to snow move poleward mainly due to the northward movement of the snowline (i.e., where largest SCF uncertainty is and SCF assimilation has the greatest impact). The effectiveness of data assimilation also varies with vegetation types, with mixed performance over forest regions and consistently good performance over grass, which can partly be explained by the linearity of the relationship between SCF and SWE in the model ensembles. The updated snow depth was compared to the Canadian Meteorological Center (CMC) data. Differences between CMC and CLM4 are generally reduced in densely monitored regions.

Assimilating satellite-based snow depth and snow cover products for improving snow predictions in Alaska

Several satellite-based snow products are assimilated, both separately and jointly, into the Noah land surface model for improving snow prediction in Alaska. These include the standard and interpreted versions of snow cover fraction (SCF) data from the Moderate-Resolution Imaging Spectroradiometer (MODIS) and the snow depth (SD) estimates from the Advanced Microwave Scanning Radiometer for the Earth Observing System (AMSR-E). The satellite-based SD estimates are adjusted against in situ observations via statistical interpolation to reduce the potentially large biases, prior to being assimilated using an ensemble Kalman filter. A customized, rule-based direct insertion approach is developed to assimilate the two SCF datasets. Our results indicate that considerable overall improvement on snow prediction can be achieved via assimilating the bias-adjusted satellite SD estimates; however, the improvement does not always translate into improvements in streamflow prediction. Assimilating the standard MODIS SCF is found to have little impact on snow and streamflow predictions, while assimilating the interpreted SCF estimates, which have reduced cloud coverage and improved snow mapping accuracy, has resulted in the most consistent improvements on snow and streamflow predictions across the study domain.

Evaluation of the MODIS snow cover fraction product

Eleven years of daily 500 m gridded Terra Moderate Resolution Imaging Spectroradiometer (MODIS) (MOD10A1) snow cover fraction (SCF) data are evaluated in terms of snow presence detection in Colorado and Washington states. The SCF detection validation study is performed using in-situ measurements and expressed in terms of snow and land detection and misclassification frequencies. A major aspect addressed in this study involves the shifting of pixel values in time due to sensor viewing angles and gridding artifacts of MODIS sensor products. To account for this error, 500 m gridded pixels are grouped and aggregated to different-sized areas to incorporate neighboring pixel information. With pixel aggregation, both the probability of detection (POD) and the false alarm ratios increase for almost all cases. Of the false negative (FN) and false positive values (referred to as the total error when combined), FN estimates dominate most of the total error and are greatly reduced with aggregation. The greatest POD increases and total error reductions occur with going from a single 500 m pixel to 3×3-pixel averaged areas. Since the MODIS SCF algorithm was developed under ideal conditions, SCF detection is also evaluated for varying conditions of vegetation, elevation, cloud cover and air temperature. Finally, using a direct insertion data assimilation approach, pixel averaged MODIS SCF observations are shown to improve modeled snowpack conditions over the single pixel observations due to the smoothing of more error-prone observations and more accurately snow-classified pixels. Copyright © 2012 John Wiley & Sons, Ltd.

Multiscale assimilation of Advanced Microwave Scanning Radiometer–EOS snow water equivalent and Moderate Resolution Imaging Spectroradiometer snow cover fraction observations in northern Colorado

Eight years (2002–2010) of Advanced Microwave Scanning Radiometer–EOS (AMSR‐E) snow water equivalent (SWE) retrievals and Moderate Resolution Imaging Spectroradiometer (MODIS) snow cover fraction (SCF) observations are assimilated separately or jointly into the Noah land surface model over a domain in Northern Colorado. A multiscale ensemble Kalman filter (EnKF) is used, supplemented with a rule‐based update. The satellite data are either left unscaled or are scaled for anomaly assimilation. The results are validated against in situ observations at 14 high‐elevation Snowpack Telemetry (SNOTEL) sites with typically deep snow and at 4 lower‐elevation Cooperative Observer Program (COOP) sites. Assimilation of coarse‐scale AMSR‐E SWE and fine‐scale MODIS SCF observations both result in realistic spatial SWE patterns. At COOP sites with shallow snowpacks, AMSR‐E SWE and MODIS SCF data assimilation are beneficial separately, and joint SWE and SCF assimilation yields significantly improved root‐mean‐square error and correlation values for scaled and unscaled data assimilation. In areas of deep snow where the SNOTEL sites are located, however, AMSR‐E retrievals are typically biased low and assimilation without prior scaling leads to degraded SWE estimates. Anomaly SWE assimilation could not improve the interannual SWE variations in the assimilation results because the AMSR‐E retrievals lack realistic interannual variability in deep snowpacks. SCF assimilation has only a marginal impact at the SNOTEL locations because these sites experience extended periods of near‐complete snow cover. Across all sites, SCF assimilation improves the timing of the onset of the snow season but without a net improvement of SWE amounts.

An approach to using snow areal depletion curves inferred from MODIS and its application to land surface modelling in Alaska

Snowcover areal depletion curves inferred from the moderate resolution imaging spectroradiometer (MODIS) are validated and then applied in NASA’s catchment-based land surface model (CLSM) for numerical simulations of hydrometeorological processes in the Kuparuk River basin (KRB) of Alaska. The results demonstrate that the MODIS snowcover fraction f derived from a simple relationship in terms of the normalized difference snow index compares well with Landsat values over the range 20 � f � 100%. For f< 20%, however, MODIS 500 m subpixel data underestimate the amount of snow by up to 13% compared with Landsat at spatial resolutions of 30 m binned to equivalent 500 m pixels. After a bias correction, MODIS snow areal depletion curves during the spring transition period of 2002 for the KRB exhibit similar features to those derived from surface-based observations. These results are applied in the CLSM subgrid-scale snow parameterization that includes a deep and a shallow snowcover fraction. Simulations of the evolution of the snowpack and of freshwater discharge rates for the KRB over a period of 11 years are then analysed with the inclusion of this feature. It is shown that persistent snowdrifts on the arctic landscape, associated with a secondary plateau in the snow areal depletion curves, are hydrologically important. An automated method is developed to generate the shallow and deep snowcover fractions from MODIS snow areal depletion curves. This provides the means to apply the CLSM subgrid-scale snow parameterization in all watersheds subject to seasonal snowcovers. Improved simulations and predictions of the global surface energy and water budgets are expected with the incorporation of the MODIS snow data into the CLSM. Copyright  2005 John Wiley & Sons, Ltd.