Land surface models that operate at multiple spatial resolutions require consistent leaf area index (LAI) inputs at each scale. In order to produce LAI from Landsat imagery that is consistent with the Moderate Resolution Imaging Spectroradiometer (MODIS) LAI product and with in situ measurements, an improved regression tree mapping approach has been developed, which uses homogeneous and high-quality LAI retrievals from MODIS and existing LAI measurements acquired on the ground as samples to train a rule-based model based on Landsat surface reflectance data in visible and infrared bands. The methodology employs a simple geostatistical approach for determining sample weight, considering the spatial distribution of the samples and data quality. Results from the Soil Moisture Experiment of 2002 field campaign show that incorporation of available ground-based LAI measurements into the training sample collection improved field-scale estimation of LAI, especially in areas of very high vegetation cover fraction, but did not degrade consistency with MODIS LAI products.