Towards NoahMP-AI: Enhancing Land Surface Model Prediction with Deep Learning

Mahmoud Mbarak, Manmeet Singh, Naveen Sudharsan, Zong-Liang Yang · 2025

Read Towards NoahMP-AI: Enhancing Land Surface Model Prediction with Deep Learning by Mahmoud Mbarak, Manmeet Singh, Naveen Sudharsan, Zong-Liang Yang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-12919.txt.

Introduction

Mahmoud Mbarak and co-authors present a 2025 paper on NoahMP-AI, a hybrid system that pairs a physics-based land model with deep learning to predict soil moisture. Soil moisture drives drought watching and flood warning, yet the physics model drifts most during extreme events, when its settings leave their tuned range. The hybrid aims to keep the physics and fix its blind spots.

The physics model first generates a rich set of land outputs, from moisture to heat fluxes. A three-dimensional neural network then corrects those fields against satellite soil moisture readings. Tests focus on how well the blend holds up during extremes, where pure physics and pure learning each tend to fail.

Catalog record

CREATORMahmoud Mbarak, Manmeet Singh, Naveen Sudharsan, Zong-Liang Yang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE19 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-12919

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Towards NoahMP-AI: Enhancing Land Surface Model Prediction with Deep Learning Mahmoud Mbarak, Manmeet Singh, Naveen Sudharsan, Zong-Liang Yang Abstract Accurate soil moisture prediction during extreme events remains a critical challenge for earth system modeling, with profound implications for drought monitoring, flood forecasting, and climate adaptation strategies. While land surface models (LSMs) provide physically-based predictions, they exhibit systematic biases during extreme conditions when their parameterizations operate outside calibrated ranges. Here we present NoahMP-AI, a physics-guided deep learning framework that addresses this challenge by leveraging the complete Noah-MP land surface model as a comprehensive physics-based feature generator while using machine learning to correct structural limitations against satellite observations. We employ a 3D U-Net architecture that processes Noah-MP outputs (soil moisture, latent heat flux, and sensible heat flux) to predict SMAP soil moisture across two contrasting extreme events: a prolonged drought (March-September 2022) and Hurricane Beryl (July 2024) over Texas. When comparing NoahMP-AI with NoahMP, our results demonstrate an increase in R² values from -0.7 to 0.5 during drought conditions, while maintaining physical consistency and spatial coherence. The framework’s ability to preserve Noah-MP’s physical relationships while learning observation-based corrections represents a significant advance in hybrid earth system modeling. This work establishes both a practical tool for operational forecasting and a benchmark for investigating the optimal integration of physics-based understanding with data-driven learning in environmental prediction systems. 1 Introduction The accurate prediction of soil moisture during extreme weather events represents one of the most pressing challenges in contemporary earth system science. Soil moisture mediates water, energy, and carbon exchanges between the land surface and atmosphere, making its accurate representation essential for drought monitoring, flood forecasting, and climate prediction [ 1 , 2 ] . However, current modeling capabilities frequently fail during extreme conditions—precisely when accurate predictions are most urgently needed. Land Surface Models (LSMs) have evolved to incorporate sophisticated representations of hydrological, thermal, and biogeochemical processes [ 3 ] . The Noah-MP (Multi-Parameterization) model represents a significant advance in this evolution, offering multiple parameterization options for key processes and demonstrating strong performance across diverse environments [ 4 , 5 ] . Despite these advances, LSMs including Noah-MP exhibit significant biases during extreme events due to parameterizations optimized for mean conditions, inadequate representation of land-atmosphere feedbacks, and structural limitations in representing scale-dependent processes [ 6 , 7 ] . The 2012 U.S. drought, which resulted in $30 billion in economic losses, exemplifies the consequences of these modeling deficiencies [ 8 ] . Two categories of extreme events stress LSMs. During droughts, models must represent slow-evolving soil moisture depletion and vegetation stress responses [ 9 ] . Major LSMs systematically misrepresent soil moisture memory during these conditions [ 10 ] . Conversely, extreme precipitation events such as hurricanes challenge LSMs to represent rapid moisture changes and phenomena like the Brown Ocean Effect, where saturated soils sustain tropical cyclone intensity post-landfall [ 11 , 12 ] . NASA’s Soil Moisture Active Passive (SMAP) mission provides global soil moisture observations that reveal systematic biases in LSM predictions [ 13 , 14 ] . While data assimilation approaches can constrain LSM states with these observations, they provide only temporary corrections without addressing underlying model deficiencies [ 15 ] .Moreover, these corrections can only be applied retrospectively after SMAP observations become available, rather than providing the predictive capability needed before the events occur. Recent advances in machine learning offer transformative potential for earth system modeling [ 16 ] . However, purely data-driven approaches may violate physical constraints and fail to generalize beyond training distributions. This has motivated physics-informed machine learning approaches that integrate physical understanding with data-driven learning [ 17 ] . In land surface modeling, such approaches have shown promise for specific applications including parameter estimation and evapotranspiration modeling [ 18 , 19 ] , yet a systematic framework leveraging the full complexity of state-of-the-art LSMs remains elusive. Here we present NoahMP-AI, a novel physics-guided deep learning framework that addresses this gap through three key innovations: (1) using the complete Noah-MP land surface model as a comprehensive physics-based feature generator, (2) employing a 3D U-Net architecture for soil moisture prediction, and (3) demonstrating robust performance across both drought and hurricane conditions. This work establishes a generalizable framework for integrating physics-based models with deep learning. 2 Materials and Methods 2.1 Study Design and Experimental Setup We designed our study to evaluate the NoahMP-AI framework. Our analysis focuses on the Texas region (25°N-37°N, 94°W-107°W), which provides an ideal test bed due to its diverse climate zones and susceptibility to both drought and hurricane events. For all experiments, we configured Noah-MP v5.0 [ 4 , 5 ] in offline mode at 9 km horizontal resolution to match SMAP observations. The model uses four soil layers (0-10, 10-40, 40-100, 100-200 cm) with hourly NLDAS-2 atmospheric forcing data. Case Study 1: Extended Drought Conditions (March-September 2022) The 2022 Texas drought represents a prolonged extreme event characterized by persistent precipitation deficits, extreme temperatures, and progressive soil moisture depletion. Over 70% of Texas experienced severe to exceptional drought conditions during this period. We generated 684 ensemble members, each covering 2-week simulation periods with perturbed initial soil moisture and temperature states, providing diverse atmospheric forcing sequences for robust training and evaluation. Case Study 2: Hurricane Beryl (July 2024) Hurricane Beryl made landfall near Matagorda, Texas, on July 8, 2024, as a Category 1 storm with maximum sustained winds of 80 mph. This event tests the framework’s ability to capture rapid soil moisture transitions and maintain physical consistency during extreme wetting. We generated 88 ensemble members covering 3-day periods around the hurricane event, with ensemble diversity created through varying initialization times. For both case studies, Noah-MP provides three key physics-based features that serve as inputs to our deep learning framework: surface soil moisture (0-10 cm, m³/m³), latent heat flux (W/m²), and sensible heat flux (W/m²). Table 1 : Summary of experimental design for drought and hurricane case studies Parameter Drought Study Hurricane Study Event period March-September 2022 July 2024 Ensemble members 684 88 Simulation length 2 weeks 3 days Temporal Resolution 3-hourly 3-hourly These features encapsulate the model’s representation of water and energy balance processes, providing rich physical information for the deep learning component. 2.2 SMAP Satellite Observations The SMAP Level-4 (L4) surface soil moisture (0-5cm) product serves as our observational target [ 14 ] . SMAP L4 provides global soil moisture estimates at 9 km spatial resolution (EASE-Grid 2.0 projection) and 3-hourly temporal frequency through the assimilation of SMAP brightness temperature observations into the NASA Catchment land surface model. Quality flags are used to exclude pixels affected by snow cover, frozen soil conditions, or radio frequency interference. For this study, we bilinearly interpolated SMAP L4 data to the 9 km Noah-MP grid and applied quality control masks to ensure training exclusively on high-confidence observations. 2.3 NoahMP-AI Architecture: Physics-Guided Deep Learning Framework Figure 1 : Architecture of the NoahMP-AI framework showing the 3D U-Net encoder-decoder structure. The NoahMP-AI framework employs a 3D U-Net convolutional neural network architecture [ 20 ] specifically adapted for spatio-temporal soil moisture prediction (Figure 1 ). The network accepts a 4D tensor of shape (batch_size, channels=3, time, height, width) where the three channels correspond to Noah-MP soil moisture, latent heat flux, and sensible heat

Read the complete text