Accurate estimation of gross ecosystem productivity (GEP) remains a challenge despite its importance in the global carbon cycle. Recently, the relationship between chlorophyll fluorescence and canopy GEP becomes an important scientific concern in plant ecology. However, it remains unclear how they are linked at multiple spatial scales across the growing season. In this study, continuous measurements of net ecosystem CO₂ exchange (NEE) were made in a desert shrub (Artemisia ordosica) ecosystem using eddy covariance (EC) technique from May to October 2015 in Mu Us desert. The real-time fluorescence (F_s) and fluorescence under light (F_m') were also measured in situ using the Monitoring-PAM multi-channel fluorometer. Leaf area index (LAI), the normalized difference vegetation index (NDVI), and environmental factors including photosynthetically active radiation (PAR), air temperature (T_a), vapor pressure deficit (VPD), soil water content (SWC) and precipitation (PPT) were measured simultaneously within the range of carbon flux contribution. We developed a chlorophyll fluorescence-based model with input variables of photochemical efficiency (Φ_PSII), PAR, and LAI to estimate ecosystem productivity (GEP_ChlF). We then compared the applicability of GEP_ChlF with EC-based GEP (GEP_EC) and examined the responses of different parameters to environmental factors at leaf and canopy scales. As a result, the diurnal pattern of Φ_PSII was similar to that of NEE, being mainly controlled by PAR and modified positively by T_a and VPD (P<0.01). The Φ_PSII had a significantly positive correlation with SWC (P<0.01). The GEP_ChlF agreed well with GEP_EC when PAR was 400-800 μmol m^-2 s^-1, with a slope of 0.627 (R²=0.67, P<0.01). The GEP_ChlF was likely to be underestimated under low PAR, leading to a high light interception rate of canopy, thus resulting in a low absorbed photosynthetically active radiation (APAR) in the model. The GEP_ChlF was significantly higher than GEP_EC under high PAR and their relationship was nonlinear, which may be due to Artemisia ordosica released excessive sunlight energy for photo-respiration, nitrogen metabolism, Miller reaction and other processes. It was noted that photo-respiration was difficult to predict accurately because the daytime ecosystem respiration (R_e) was simulated on the basis of nocturnal temperature sensitivity. Compared with GEP_EC, chlorophyll fluorescence-derived GEP_ChlF was more sensitive to environmental fluctuations and had a better relation with environmental factors. Our results confirmed that the GEP_ChlF from chlorophyll fluorescence-based model could replace GEP_EC as a good ecological parameter at canopy scale in consideration of the plant physiological status. We found that excessive radiation, extreme temperature, high transpiration and drought were the main stress factors limiting the photosynthetic process of Artemisia ordosica at different spatial scales. The study provides a method for estimating ecosystem productivity under the stressed environmental conditions on the basis of monitoring chlorophyll fluorescence in relation to environmental factors. The present results may scientifically support decision-making for promoting regional sustainable development.