UNCERTAINTY IN MULTI SENSOR SNOW AREA ESTIMATES FOR THE TUPALANG RIVER BASIN

Abstract

Snow area derived from satellite imagery changes with the classification threshold, observation frequency, spatial resolution, and acquisition date. This thesis quantifies those effects for the Tupalang River Basin. The baseline series used daily MODIS MOD10A1 data for 2001 to 2025 with an NDSI threshold of 30. Alternative thresholds of 20 and 40 were applied to the same workflow. Raising the threshold reduced mean annual snow area from 1441.75 to 1351.67 km2 and mean spring area from 1483.38 to 1387.37 km2. Despite these differences in magnitude, annual Sen slopes remained between -5.16 and -6.01 km2 per year, and spring slopes remained between -6.13 and -6.42 km2 per year. All threshold-specific trends were non-significant. Across four comparison years, Landsat snow area averaged 71.8 percent of the MODIS spring metric, with a mean bias of -387.17 km2 and RMSE of 442.15 km2. The Sentinel-2 estimate averaged 20.5 percent, with a mean bias of -1097.58 km2 and RMSE of 1117.17 km2. These values measure disagreement between observation designs; they are not accuracy scores against field data. Long-term trend analysis should therefore retain the MODIS seasonal index, while high-resolution imagery should be used within a matched multi-date validation protocol.