Low-cost sensors and multitemporal remote sensing for operational turbidity monitoring in an East African wetland environment

Steinbach, S.; Rienow, A.; Chege, M. W.; Dedring, N.; Kipkemboi, W.; Thiong’o, B. K.; Zwart, Sander Jaap; Nelson, A. 2024. Low-cost sensors and multitemporal remote sensing for operational turbidity monitoring in an East African wetland environment. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 17:8490-8508. [doi: https://doi.org/10.1109/JSTARS.2024.3381756]

Many wetlands in East Africa are farmed and wetland reservoirs are used for irrigation, livestock, and fishing. Water quality and agriculture have a mutual influence on each other. Turbidity is a principal indicator of water quality and can be used for, otherwise, unmonitored water sources. Low-cost turbidity sensors improve in situ coverage and enable community engagement. The availability of high spatial resolution satellite images from the Sentinel-2 multispectral instrument and of bio-optical models, such as the Case 2 Regional CoastColor (C2RCC) processor, has fostered turbidity modeling. However, these models need local adjustment, and the quality of low-cost sensor measurements is debated. We tested the combination of both technologies to monitor turbidity in small wetland reservoirs in Kenya. We sampled ten reservoirs with low-cost sensors and a turbidimeter during five Sentinel-2 overpasses. Low-cost sensor calibration resulted in an R2 of 0.71. The models using the C2RCC C2X-COMPLEX (C2XC) neural nets with turbidimeter measurements (R2 =0.83) and with low-cost measurements (R2 = 0.62) performed better than the turbidimeter-based C2X model. The C2XC models showed similar patterns for a one-year time series, particularly around the turbidity limit set by Kenyan authorities. This shows that both the data from the commercial turbidimeter and the low-cost sensor setup, despite sensor uncertainties, could be used to validate the applicability of C2RCC in the study area, select the better-performing neural nets, and adapt the model to the study site. We conclude that combined monitoring with low-cost sensors and remote sensing can support wetland and water management while strengthening community-centered approaches.