A new digital cover photography dataset and processing tool for SMAPVEX19-22: How siting and sky condition impact plant area index retrievals in continuous measurement set-ups

Satellite remote sensing is widely used for Leaf Area Index (LAI) retrievals, but calibration and validation efforts require ground Plant Area Index (PAI) values that may be converted to LAI. Digital Cover Photography (DCP) presents an affordable means for covering large regions (∼2 × 33² km²) and multiple years. The Soil Moisture Active Passive Validation Experiment conducted from 2019–2022 (SMAPVEX19–22) utilized DCP data to study vegetation impacts on soil moisture retrievals in forests. This work reports on the DCP tool “EzPAI”, its outputs for SMAPVEX19–22, and clear sky PAI bias identification and its correction. EzPAI features sky condition tracking, multi-tier data screening, data quality flagging, and two-corner thresholding. We found that PAI is overestimated in clear conditions, and our bias correction approach reduced this by 0.2 on average. Massachusetts (‘MA’) and New York (‘MB’) networks attained comparable results for cloudiness (∼67 %) and poor data quality (21 %). Benchmark comparisons to other approaches (DCP tools, Sentinel-2 LAI, LAI-2200c) showed good agreement and performance. EzPAI showed similar results for in situ and Sentinel-2 LAI comparisons, achieving R∼0.9, and some bias (MD= <0.53). For comparisons to the coveR DCP tool the correlation was 0.73 and bias 0.26. For summer PAI totals, CoveR, coverPy, EzPAI and LAI-220c obtained nearly identical results: 4.08±0.35 4.08±0.33, 4.12±0.33 and 4.16±0.77. Differences may be explained in part due to image quality issues (noted at needleleaf canopies), the LAI-2200c data being noisy (σ_spring=0.43, σ_summer=0.72), the different measurement modalities used (satellite, handheld hemispherical, DCP), and our assumption of a constant extinction coefficient (k = 0.65) across 144 site-years. PAI values ranged from 2.73 to 5.16 (3.92 average) and 2.44 to 4.96 (3.80 average) for MA and MB, respectively. Processing time on a Dell Precision Laptop 7560 was 0.87 per image, which was about 3x speedier than coveR.