Performance and optimisation strategy of a multispectral sensor as part of a newly developed low-cost IoT device for forest monitoring (RemoTrees - beta)

A new low-cost device based on Internet-of-Things (IoT) communication has been developed within the RemoTrees project to monitor climate-change effects in remote forest ecosystems. One of the key component of this device, referred to as the RemoTrees - beta, is a multispectral chipset composed of four sensors (three AS7265X and one AS7341), providing 26 spectral channels covering the range 410–940 nm. The chipset is equipped with a 1-inch diffuser designed to collect hemispherical solar radiation over incidence angles θ ∈ [−90°, +90°], with an angular response close to the cosine law. Here we present laboratory characterisation and calibration results obtained for 15 replicate RemoTrees - beta units. The spectral performance was highly consistent across devices, with central-wavelength variations below ~2 nm. Full width at half maximum (FWHM) values ranged from 19.17 to 47.93 nm, with standard deviations between 0.32 and 1.74 nm and a maximum relative expanded uncertainty of 0.90%. Because the devices will operate under highly variable illumination conditions (time of day, season, latitude, altitude, cloudiness, and canopy cover), optimisation of integration time (IT) and gain (G) is essential to avoid low digital-number (DN) values and insufficient use of the sensor dynamic range. As commonly applied in field spectrometry, automated IT/G optimisation and scan averaging are recommended to maximise signal-to-noise ratio (SNR) and minimise measurement uncertainty. When IT settings alone are insufficient to reach a satisfactory fraction of the dynamic range (≈65 000 DN; target ≥50%), summing of consecutive readings can be used to effectively increase the integration time while limiting saturation risks under rapidly changing sub-canopy light conditions. Radiometric sensitivity was evaluated by varying G and IT. Under optimised settings, SNR values up to ~5000 were achieved. For AS7265X sensors, gains G > 16 combined with IT optimisation increased SNR by up to ~4×, while for AS7341 gains G > 2 with IT optimisation yielded improvements up to ~5×. Detector nonlinearity contributes an expanded uncertainty of up to ±2.98% (k = 2) if uncorrected, which decreases to ≤±1.24% when nonlinearity correction is applied. The calibration coefficients derived from the tested devices showed moderate inter-device variability, with a maximum variation of approximately 10% for each spectral band. The RemoTrees - beta light sensor demonstrates stable spectral performance, high achievable SNR, and manageable inter-device variability, supporting its suitability for large-scale deployment in forest monitoring networks. Proper optimisation of integration time, gain, and signal averaging is essential to fully exploit the sensor dynamic range and minimise uncertainties under highly variable illumination conditions. Ongoing field deployment will further validate these strategies and refine operational protocols for long-term climate monitoring applications.