Thornthwaite’s Water Balance Components in Greece with the Use of Gridded Data ^†

Abstract

Thornthwaite’s water balance approach serves as a fundamental tool for assessing hydrological dynamics, particularly in regions vulnerable to aridity and water stress. This study evaluates the performance of gridded datasets in estimating Thornthwaite’s water balance attributes in Greece, leveraging climatic averages of the period 1960–1997. Ground station data from 91 meteorological sites and gridded data from the Climate Research Unit (CRU) of the University of East Anglia were utilized to assess key water balance components. The results indicate that while gridded datasets offer an alternative for regions with limited ground data, local calibration is required due to notable discrepancies. More specifically, it was found that gridded data tended to underestimate precipitation, with estimates approximately 25% lower compared to ground station data. The potential evapotranspiration (PET) estimates using gridded data were more accurate, with underestimation on the order of 10%. Moreover, the gridded data produced overestimations for all of the water balance key components including soil moisture (St), monthly changes in soil moisture (ΔSt), and actual evapotranspiration (AE) compared to the ground station data. The water surplus (S) estimates showed a significant dispersion of values when using the gridded data, particularly in regions characterized by more arid conditions. In addition, the application of gridded data led to a great increase in the aridity index (AI) values, altering the desertification classification of sites from semi-arid to sub-humid or humid categories. These findings underscore the importance of careful consideration when utilizing gridded datasets for hydrological and bioclimatic assessments, particularly in Mediterranean climate regions characterized by a complex topography and temporal climatic variability.

1. Introduction

Climate change is a key factor determining the water availability for natural vegetation and has a significant impact on forests, especially in the Mediterranean [1,2]. The water balance approach, introduced by Thornthwaite [3] and revised by Thornthwaite and Mather [4,5], is a widely known and used methodology adopted by UNEP [6] and UNESCO [7] for bioclimatic assessments and classifications, with emphasis on aridity. Through this approach, precipitation is distributed to support evapotranspiration, increase soil moisture, and store water in the soil and initiate runoff. Several recent works have applied the Thornthwaite methodology to categorize regional climates into aridity categories [8,9] and also for simple hydrological assessments, e.g., Karim et al. [10], Das [11], etc.

This work follows previous studies by Tsiros et al. [8] and Proutsos et al. [9], who studied the aridity index over the Greek peninsula through the application of Thornthwaite’s [3] water balance approach. The present study is aimed at evaluating the performance of a gridded dataset in estimating Thornthwaite’s water balance components to determine its suitability for hydrologic and bioclimatic assessments.

2. Data and Methods

Climatic averages were employed to estimate Thornthwaite’s water balance components. For the recent climatic period 1960–1997, monthly air temperature and precipitation data from 91 meteorological stations were used, as depicted in Figure 1. All stations are installed and operated by the Hellenic National Meteorological Service (HNMS). In addition, gridded monthly temperature and precipitation data from the Climate Research Unit (CRU) of the University of East Anglia [12] were also used for the same climatic periods.

The main components of the water balance were derived or estimated using the two datasets, and the results were compared through a regression analysis. These components include potential evapotranspiration (PET), actual evapotranspiration (AE), precipitation (P), the difference between P and PET, soil moisture (St), monthly changes in soil moisture (ΔSt), Accumulated Potential Water Loss (APWL), water deficit (D), and water surplus (S), as described in Thornthwaite [3] and Thornthwaite and Mather [4,5].

All estimations were carried out on a monthly basis, whereas the aridity index (AI) was calculated annually for all stations, assuming a maximum soil storage water capacity (SWC) of 300 mm. The linear regression coefficients (slope a, offset b, and coefficient of determination R² of the y = ax + b) were assessed to evaluate the performance of the gridded dataset.

3. Results and Discussion

The climatic mean estimates of Thornthwaite’s water balance attributes using data from ground stations and gridded data (CRU) are presented in Figure 2. Across all of the parameters examined, the use of gridded data consistently resulted in the underestimation of water fluxes, as indicated by the slope (a) values of the linear regression lines being less than unity. For example, the precipitation values (Figure 2b) from the gridded data were approximately 25% lower compared to those recorded by the ground stations, yielding a slope (a) of 0.75. Similarly, the potential evapotranspiration (PET) was found to be underestimated by about 11%, with a slope (a) of 0.89 (Figure 2a). The regression lines in Figure 2, however, consistently exhibit low offset values (b), indicating minimal systematic error in the gridded data. It is also interesting to note that the values of the determination coefficients (R²) are relatively high, exceeding 0.73 for almost all of the parameters, except for water surplus (S) during the wet period of the year, which has an R² of 0.37.

The performance of the gridded data for estimating the potential evapotranspiration (PET) is good, while it is moderate for precipitation (P). However, this does not greatly affect the accuracy of the estimated P-PET values, which, like PET, are slightly underestimated. The gridded data also tend to underestimate the soil moisture (St), monthly changes in soil moisture (ΔSt), and actual evapotranspiration (AE) compared to the ground stations. Conversely, the water surplus (S) exhibits considerable dispersion in its values (Figure 2i). St typically represents water quantities contributing to runoff and deep percolation, primarily occurring when the soil’s root zone is saturated, notably during the Mediterranean’s winter months. Despite their small magnitudes, these fluxes significantly impact local water cycles, challenging accurate estimation with gridded data, particularly since they are influenced more by P (where gridded data performance is moderate) than by PET. Conversely, water deficit (D) mainly occurs during summer and is more influenced by PET, which gridded data approximates more accurately than P. Consequently, D values from gridded data tend to be more accurate and less underestimated compared to ground station data. It should also be noted that the D values are generally higher than the S values, consistent with Mediterranean climatic conditions [8,9].

The application of gridded data also results in an overestimation of the aridity index (AI), upon which UNEP’s aridity classification system relies. The average AI from the ground station data in this study was 0.67, while the corresponding average AI using the gridded data was significantly higher (0.80). Considering UNEP’s [6] recommended thresholds for AI classification (arid A for AI = 0.05–0.20, semi-arid SA for AI = 0.20–0.50, sub-humid SH for AI = 0.50–0.60, and humid H for AI > 0.65), applying gridded data introduces high uncertainties and false estimations, as shown in

Table 1. Distribution of sites according to the ground and gridded datasets.

Climate Category	Number of Sites
	Ground Data	Gridded Data
Semi-arid, SA	23	0
Sub-humid, SH	19	18
Humid, H	49	73

. The analysis of the data from 91 ground stations revealed 23 sites as SA, 19 as SH, and 49 as H, with average AI values of 0.40 (sd = 0.07), 0.57 (sd = 0.05), and 0.86 (sd = 0.20), respectively. The overestimation of AI using the gridded data resulted in the classification of all sites into more humid aridity classes. Specifically, the AI values for all stations exceeded 0.50, leading to classification as SH or H, with none classified as SA. Particularly for SA sites (as classified by ground data), the average AI increased significantly to 0.67 (sd = 0.08), compared to 0.75 (sd = 0.12) for SH and 0.88 (sd = 0.20) for H, substantially higher than the AI estimated by ground data.

4. Conclusions

The results of the present study indicate that the use of gridded data can be an alternative for application in sites with limited ground data but it may produce high uncertainties and errors in the estimation of some water balance components. In general, the tested gridded dataset produced underestimations for all of the main parameters of the water budget and generally resulted in more humid conditions than those recorded by the ground stations. The use of gridded data thus should be performed with caution and, in many cases, seems to affect local calibration or the establishment of proper thresholds for aridity classification.