Phytoplankton chlorophyll-a (Chl-a) concentration serves as a crucial indicator for assessing water eutrophication status. Conventional monitoring approaches face significant limitations: laboratory analyses are time-consuming and labor-intensive, while in-situ sensors suffer from biofouling interference, low accuracy, and high maintenance costs. Traditional satellite remote sensing techniques are unsuitable for high-precision and real-time monitoring due to incorrect atmospheric correction, technical complexity, and poor temporal resolution. The emergence of hyperspectral proximal sensing technology has effectively addressed these challenges, significantly improving Chl-a concentration monitoring efficiency. This study employed a novel portable hyperspectral proximal sensing water quality monitoring device, collecting 533 synchronized in-situ Chl-a measurements across eight lakes, reservoirs, and rivers from 2021 to 2024. High-accuracy Chl-a concentration inversion models were developed and compared through both linear regression algorithms and machine learning approaches to achieve real-time Chl-a concentration monitoring. Comparative analysis of models based on linear regression, Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Support Vector Machine (SVM) algorithms revealed that the XGBoost-based model demonstrated superior performance (R2=0.87, RMSE=6.02 μg/L, MAE=3.98 μg/L). This innovative methodology enables simultaneous spectral acquisition and Chl-a concentration estimation, streamlining field monitoring procedures while reducing technical barriers and significantly enhancing operational efficiency.