Conditioning-controlled retrieval of broadband land surface temperature and emissivity from paired ground-based longwave irradiance measurements

This is the authors' abstract. We don't add key points for this paper.

Accurate retrieval of land surface temperature ( T s ) from broadband longwave radiometric measurements is limited by the nonlinear coupling between surface emissivity ( ϵ ) and temperature, particularly when the available radiative contrast provides insufficient information to distinguish their contributions. A conditioning-controlled framework is presented for jointly retrieving broadband ϵ and T s directly from high-temporal-resolution paired ground-based upwelling ( L ) and downwelling ( l ) longwave irradiance measurements, without requiring externally prescribed emissivity as a retrieval input. The framework constructs a local population of candidate temporal pairs and combines quasi-steady thermal-state screening, numerical and physical admissibility, explicit irradiance-identifiability testing, uncertainty- and information-aware progressive pair selection, physical-state-dependent reciprocal emissivity treatment, fixed eight-update Newton–Raphson refinement, and correlated uncertainty propagation. Field evaluation used 637 datasets from nine SURFRAD and BSRN stations spanning vegetated, mixed, and bare/desert environments. Five datasets lacked the required finite initial irradiances, leaving 632 usable cases, all of which produced final finite retrievals; 628 reached the stable-centre selection criterion and four required the prescribed fallback pathway. Across the 632 field cases, retrieved T s had a bias of −0.162 K, mean absolute error (MAE) of 0.432 K, root-mean-square error (RMSE) of 0.558 K, and R 2 =0.9984 relative to the ground-based validation-reference temperature. A complementary physics-based known-truth experiment independently prescribed ϵ true and T s,true for 6000 scenes spanning 0.85≤ϵtrue≤0.99. In the noiseless experiment, 5924 scenes (98.73 %) yielded successful retrievals; among successful cases, emissivity bias, MAE, and RMSE were 0.0313, 0.0491, and 0.0658, respectively, while the corresponding T s values were −0.335, 1.307, and 3.213 K. Retrieval errors increased toward the lowest-emissivity regime, identifying low-emissivity conditions as the most demanding part of the tested retrieval space. The combined field and known-truth results demonstrate that broadband emissivity and surface temperature can be jointly retrieved from paired longwave irradiance observations when radiative observability and physical applicability are explicitly conditioned before nonlinear inversion.