Muhammad Shahzad Yousaf

Mr. Muhammad Shahzad Yousaf


Research Focus & Metrics

Academic Career Length

1Year (2026-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
30 June 2026 in Preprints.org
Verified
Preprint
Open access
PDF file
HFHassan Ali Fattah
Hassan Ali Fattah
SMShah Mir Mowahed
Shah Mir Mowahed
JDJIQIU DENG
JIQIU DENG
MYMuhammad Shahzad Yousaf
Muhammad Shahzad Yousaf
Accurate nighttime land surface temperature (LSTn) forecasting in data-scarce, topographically complex regions remains a significant challenge for satellite remote sensing and climate monitoring. To address this, the present study introduces the Physics-Guided Transformer Bidirectional Long Short-Term Memory (PG-ST-BiLSTM) deep learning model. For comparison, we employ four baseline models: three deep learning models (TCN, LSTM, and BiLSTM) and one statistical model (linear regression). Using Afghanistan, a topographically challenging and data-scarce region, as a “hard” test bed, the proposed model is evaluated across five major river basins over the period 2001-2024. The proposed PG-ST-BiLSTM achieves an R² of 0.977, RMSE of 2.023 °C, and MAE of 1.534 °C, outperforming baseline models including TCN (R² = 0.961), LSTM (R² = 0.966), BiLSTM (R² = 0.966), and linear regression (R² = 0.926). The model demonstrates strong alignment between observed and fitted LSTn values along the reference line of y = 0.98x + 0.21 closely approaching the 1:1 reference line. All Sen’s slope values are positive, confirming a universal nighttime warming trend across the projected period (2025-2040), with a grand mean of +0.0160 °C/year across all basin-season combinations, suggesting a marginal acceleration of LSTn warming under a trend-stationary climatological assumption, wherein historical warming rates derived from Mann-Kendall and Theil–Sen analyses are assumed to persist through 2040. At the seasonal level, Winter exhibits the highest projected warming rate (+0.1174 °C/year), followed by Summer (+0.0984 °C/year), Autumn (+0.0545 °C/year), and Spring (+0.0504 °C/year), indicating that cold-season and peak-summer nighttime temperatures are projected to remain the most thermally vulnerable through 2040. At the basin level, the hyper-arid Helmand Basin records the highest projected mean Sen’s slope (+0.0796 °C/year), consistent with its minimal vegetation buffering and high nocturnal thermal emissivity, while the Amu-Darya Basin exhibits the lowest warming rate (+0.0460 °C/year), attributable to orographic cooling and comparatively higher soil moisture retention. These results demonstrate the generalizability and reproducibility of the PG-ST-BiLSTM model for nighttime LST forecasting in data-scarce, mountainous regions, with direct relevance to climate adaptation, water resource planning, and cryosphere monitoring worldwide.