Accurate climate data for specific locations are essential for ecological modeling and genecology. Cost-effectiveness is another major concern for climate data collection. Observations from standard weather stations provide the most accurate climate data, but they are mostly distant from the forest or ecological sites of interest. Onsite weather stations are widely applied to obtain local climate data, but their purchase and maintenance cost is high. ClimateNA can also predict climate variables for specific locations for free and has been widely used as well. Although ClimateNA has been evaluated against weather stations over entire North America (Wang et al. 2016), its accuracy remains a concern, particularly at remote and mountainous areas where standard weather stations, upon which ClimateNA is built, are sparse. In this study, we found that ClimateNA can generate climate variables at reasonably high accuracy, particularly for temperature variables. The response functions developed using the climate variables from the two sources showed similar shapes. Therefore, our results suggested that ClimateNA can serve as a competitive alternative to obtain climate data.
4.1 The accuracy of ClimateNA predictions for temperature variables
We found that the predictions of temperature variables by ClimateNA to be highly accurate. Temperature-related climate variables are often found to be the primary drivers of local adaptation (Hu et al. 2019) and the major environmental factor affecting the performance of plant populations (Rehfeldt et al. 1999; Wang et al. 2006). Further, temperature is critical to moisture-related variables as it affects evapotranspiration, and increasing temperatures are expected to contribute significantly to future drought stress (Tam et al. 2019). Thus, the high accuracy of the ClimateNA predictions for temperature variables is important and encouraging.
For univariate response functions using temperature variables as predictors, we found MAT was the most important climate variable. MAT was widely used as an explanatory variable in previous studies (Pacala et al. 1996; Rehfeldt et al. 1999; Wang et al. 2010; Flynn & Wolkovich 2018; Piao et al. 2019). We also found that the response curves derived from the two climate data sources were moderately strong and very similar in shape using MAT as the predictor. These results suggest that the use of such climate variables from ClimateNA is not likely to lead to bias in developing climate response functions of tree populations.
4.2 The accuracy of ClimateNA predictions for precipitation variables
Compared with temperature variables, precipitation variables predicted by ClimateNA were less accurate, which is in agreement with the results of a study testing local accuracy of ClimateWNA in the southern Yukon (Krebs et al. 2018). The same situation also occurs in a study using gridded data to predict agricultural yields in the US (Mourtzinis et al. 2017). The team found that gridded weather models produced good predictions only for temperature-related variables, while the results for precipitation-related variables are not satisfying (Mourtzinis et al. 2017). These results indicated that precipitation is more difficult to predict than temperature, especially in mountainous areas (Marquı́nez et al. 2003; Germann et al. 2006).
In the present study, several problems were found regarding observed onsite precipitation data. Aside from missing data for both precipitation and temperature records due to uncontrollable reasons, we found that the recorder took precipitation record as 0 for more than a year at two test sites (Jordan River, 48.426°N, -124.023°W, on Vancouver Island and Whitecourt, 54.056° N, -115.791°, in Alberta), which is not realistic. Tipping buckets (also called “rain gauges”) were sometimes plugged by wasp and bird nests, causing the precipitation loggers to miss rainfall events, and no precipitation was recorded during periods of freezing weather because snow is not released from the tipping arm. These factors may have contributed to the underestimate of onsite observed precipitation detected when compared against precipitation records from adjacent standard weather stations (Fig. 3). Consequently, climate response functions developed using precipitation variables from the two climate data sources differed significantly in univariate response functions, suggesting that using precipitation data from onsite weather stations might lead to considerable bias in ecological modeling.
The multiple regression response functions using the two climate variables for building univariate response functions separately, significantly improved the goodness of model fit. However, the problem of the range difference in precipitation remained in the multiple regression response functions. Importantly, we found no clear evidence that the observed data is better than the predicted data for establishing either univariate or multiple regression response functions.
Overall, the highly accurate monthly temperatures and moderately accurate monthly precipitation predicted by ClimateNA are probably attributable to the scale-free feature of the model as it allows climate variables to be predicted for specific locations, whereas other climate data sources rely on grid cell averages. Results of this study suggest that ClimateNA can yield the same level of accuracy for temperature variables and a higher level of accuracy for precipitation variables compared to the observed data from the onsite weather stations in mountainous locations. Underestimates of onsite precipitation could lead to erroneous predictions of forest productivity when response functions are constructed using onsite precipitation. Therefore, particular caution needs to be practiced in using the precipitation data from onsite weather stations.
4.3 Cost-effectiveness of climate models comparing to onsite weather stations
Most temporary field research projects that require climate data use inexpensive ‘onsite’ weather stations maintained by non-specialist crews. Unlike standard weather stations operated by weather agencies, onsite stations generally lack wildlife and vandal exclusion fences, predisposing them to damage. And the high cost of accessing remote weather stations can limit the frequency with which they are maintained. Together, these factors can contribute to lower quality data and greater data gaps compared with standard weather station data. The frequency of data gaps did not increase with station age, suggesting that equipment deterioration was not at fault, nonetheless, 11 of 15 stations had at least one data gap. All these problems combined make it harder and costlier to get precipitation data of high accuracy.
Climate data can be acquired free online from climate models such as ClimateNA, which is considerably more cost-efficient than establishing and maintaining remote onsite stations. Also, climate models can predict climate variables for past and future periods and can derive biologically relevant climate variables. Nowadays, using climate models as a tool for tree growth modeling has become a major trend, and we believe those models like ClimateNA are valuable alternatives to onsite weather stations.
4.4 Limitations and reflection
We used ClimateNA to predict climate data at 11 geographically widespread test sites, most of which are located in remote or mountainous terrain where standard weather stations are sparse, and obtaining accurate predictions can be challenging. Nonetheless, we were able to develop moderately strong response functions, despite a relatively narrow climate range among the test sites (MAT -1.1 to 7.2°C; PPT_sp 65 to 235 mm). With more sites well-distributed in terms of geography and climate gradient, the response functions might be stronger and the results of assessment might be clearer.
Although the test sites involved in the study are located in mountain and remote areas, the entire region of this study is still located within a relative central coverage of ClimateNA (Wang et al. 2016). Thus, our results may not represent the situation of peripheral region, such as Nunavut, North West Territories and Yukon, where the number of weather stations used for developing the baseline data of ClimateNA is limited. Thus, the significance of using onsite weather stations can be much greater in those regions as suggested by Krebs et al. (2018).
For acquiring precipitation data in our study, although ClimateNA is better than onsite weather stations, its predictions were still deviated from the measurement from Environment Canada’s standard weather stations. Thus, if high-quality local precipitation data are required, it may be necessary to establish high-quality onsite weather stations and maintain them carefully.